Technology Day 2009 - "The Mind's Eye”
MEAD WYMAN: Good morning and welcome to Technology Day 2009. And welcome to those of you who've come from afar back to visit MIT. Welcome also, somewhere out there, the web participants who are participating with us today via the internet. I'm Mead Wyman. I'm with the class of 1962. I am a member of the class of '62. And I am the Chairman of this year's Technology Day Committee. The--
No applause is really necessary yet.
The MIT Alumni Association ascribes great importance to the Technology Day program. There are some reasons. I'm sure there are many reasons for this, but there's one that's particularly compelling for me. We know that when you come back to MIT for a reunion, or a for a visit, or for whatever, you many times would like to have some exposure to the world class research and education which is taking place here. The Tech Day program is a principal vehicle for responding to that expectation.
Of course, it's impossible to think about giving you a comprehensive view of everything that's happening at MIT. That would simply be impossible. And so it falls to the Technology Day Committee to fashion a program which allows you, the audience, to take a look at one of, or several of, the areas of major concentration at the Institute. Today, we're going to address the areas of neuroscience and artificial intelligence.
We're very fortunate to have three outstanding professors from MIT to give presentations. As a matter of fact, I can't imagine that it would be otherwise, that we have three outstanding professors, but we're very delighted to have professors Saxe, Sinha, and Winston. These folks will lead us through the current thinking and work that's being done in their areas of specialization. And we really appreciate their willingness to participate with us today in the proceedings.
Before we start, I have a few administrative matters which I have to cover with you. As a fellow nerd, I am required, or it will be necessary for me to ask you again to check and make sure that you have turned all of your electronic devices off. Electronic devices which could possibly distract you, could distract the speakers, or your neighbors. Of course, we're talking principally about cell phones. And I should point out that we do make an exception for hearing aids and pacemakers.
With regard to the question and answer-- question and answer period, which follows each one of the speakers presentations, we will be asking you to fill out on the index cards that were given to you at the outset here-- your questions, write them out. And the ushers, toward the end of the presentations, will come down through the aisles and collect your questions, bring them up to the front, where they'll be organized, and then pass them on up to the podium. This has worked very successfully for us. It did work very successfully for us last year, allowing us to process a larger number of questions than we could when we had people standing at the microphones.
I hope that many of you had a chance to see the video which was playing before the start of the program this morning. This is a 10-minute video, which is a very interesting portrait of MIT. I noticed that the sound was not-- it wasn't possible to hear the sound. And for those-- for that reason, and for also for those of you who may not have seen the video, we want-- we want to play that at our 30-minute intermission so that you can see it. Many of us who've seen that find that it's a very moving portrait of the Institute. So it will be shown the last 10 minutes of our 30-minute, intermission. And I could highly recommend that if you haven't seen it, you come and take a look at it.
All right, we are on a very tight time schedule today. And so I wanted to mention that my introduction of our speakers will be quite brief. We can do that in part because there are very extensive biographical information, there's very good biographical information and abstracts of the presentations in your Technology Day program. So let's get started.
One of the very special parts of Technology Day is the opportunity for us to hear and to talk with our president, Susan Hawkfield. Susan Hawkfield-- here's where you'll see the brief introduction. Susan Hawkfield is the 16th president of MIT. She is the first woman. And she is the first life scientist to be a president of our wonderful Institution. Prior to being the Provost at Yale University and President at MIT, she was, and continues to be, a professor in neurobiology.
I will-- I hope that's enough, Susan.
Well, you beat me to the punch. I wanted to say welcome President Hawkfield.
SUSAN HAWKFIELD: Thank you so much. Thanks. Well, good morning, everyone. It's great to see you all. Mead, thank you for that kind and brief introduction. We're going to work on making it even briefer for next year. I want to also thank Mead for chairing the Technology Day Committee, and all the members of the Committee, for putting together such a spectacular program today.
The last time Mead and I saw one another was when we were joined by several hundred MIT students at the NCAA competition, where our basketball team, for the first time in its over 100 year history, had made to the NCAA. It was an enormously exciting event, set of events.
And one of my favorite stories is the first night. I think Mead was there the first night. I could only make it the second night. But the first night, the opposing team played at that school's arena. So we had several hundred, they had several hundred students. And their students, their fans, were taunting ours by holding up signs that said, your algebra won't help you now.
We won that night.
In any case, this program this morning is only part of what has been so far a spectacular weekend. I saw many of you at the Tech Night at the Pops on Thursday night. You know, I'm becoming kind of connoisseur of Tech Night at the Pops. I think it was the best we've ever had. Lots of fun, great, great performance, good program. And then yesterday, I declared the commencement was a success. It did not rain until after we finished awarding over 2,000 degrees.
Exactly. And it is just-- it's an exciting weekend. The mood on campus is very high. And it's just wonderful to have you all here. So thank you for coming to Technology Day. You are MIT's very best ambassadors. And we hope to equip you for your ambassadorial roles by providing some of the information. As Mead said, it can't be a comprehensive overview of what we're doing at MIT, but it'll be a little insight into some of the spectacular work that's going on on campus.
And for me, as a neuroscientist, it's a little bit like getting three desserts. So I'm very much looking forward to these marvelous presentations today. Now, before we go any further with the formal part of today's program, I want to capitalize on this moment of having so many alumni here in the audience to introduce two enormously important people. One, we all need to thank, and the other we're going to welcome. So I would invite Sherwin Greenblatt and Judy Cole to join me up here.
Come on up. Don't be shy. Sherwin is-- and Judy also, these are not people who are used to being in the limelight. So as they're coming up, I'm going to talk about first Sherwin. This past year has been a year of many transitions for the Alumni Association. They moved from seven different locations in three different buildings, into one spectacular facility at 600 Memorial Drive. And I hope many of you have had a chance to see that this weekend.
This year has been a time of searching for a new leader for the Alumni Association. And through all of these transitions, the Alumni Association has been more than ably led by Sherwin Greenblatt. Sherwin in my mind is kind of the canonical MIT alumnus. He, of course, graduated from MIT. He has had a great career. He started as employee number one at the Bose Corporation. And then a couple of years ago, after a whole career at Bose, he retired as president of the Bose Corporation.
And I would say he retired, but really he tried to retire, because his retirement activity was running the Venture Mentoring Service for MIT. Which many of us wouldn't consider that to be retirement. So he's demonstrating what I found to be an almost universal characteristic among MIT alumni, the distinct failure at being able to retire. If any of you need advice, Sherwin is an expert.
So when I first started, in the summer of my first year at MIT, we needed an interim executive vice president for the Institute. And I called on Sherwin to take on that role. And when the three months that I had recruited him to serve turned into more than a year, he continued to serve ably. And then, when our new executive vice president and treasurer, Terry Stone, was found, and she started, Sherwin went back to retiring at the VMS.
But then, when we needed an interim leader as the EVP of the Alumni Association, of course, I turned to Sherwin again. But as the months went on, he became increasingly anxious to leave, because he's ready to go on to a new job, another new job, as a grandfather. So last week, he celebrated the birth of Bryce Jennings Greenblatt, his first grandchild. And so I have some gifts to thank Sherwin for his service and to get him on his way into his next job. Please join me in thanking him.
SHERWIN GREENBLATT: That is really nice.
SUSAN HAWKFIELD: Thank you.
SHERWIN GREENBLATT: Oh, thank you.
SUSAN HAWKFIELD: And actually, I think this is also for you.
SHERWIN GREENBLATT: Okay. Thank you. I appreciate it.
SUSAN HAWKFIELD: Now, we wouldn't have let Sherwin leave, I wouldn't have let Sherwin leave, unless we knew that the Alumni Association would be in terrific hands. And so it is a very great pleasure to introduce, for those you have not already met her, Judy Cole, our new EVP and CEO of the Alumni Association.
Now, I won't take the time to formally introduce her. I hope that you all have read her background. And I do hope that many of you will find an opportunity to meet Judy this weekend. She could not be better prepared for us. Everyone who interviewed Judy knew at once that she would "get it" about MIT. There's a kind of intensity, and humor, a devotion, an ability to work really hard. And believe me, and I hope all of you can believe me now, Judy is just a terrific new leader for the Alumni Association. So please, welcome Judy Cole to MIT and to the AA.
JUDY COLE: Great, thank you so much.
SUSAN HAWKFIELD: You're welcome. Thank you.
Okay, now my task is to give a comprehensive tour of what's been going on at MIT this year. Mead described how we don't have the comprehensive in Technology Day, but I feel I've got to touch a lot of things. So it's going to be a whirlwind tour. I apologize. The punch line of the race across the campus this morning is that it has been, as you all know, a very demanding year. But we have had a great year here at the Institute.
I want to be sure that I leave sometime to hear your questions. So let me just kind of lay out the three things I want to talk about. The first is MIT's role serving in these difficult times. The second is I'll give you an update on how we're coping with the financial downturn, including where we are on our campaign for students. And then I will describe one important critical source of our momentum for the future.
So while the pattern of activities at the end of the year has a kind of timeless nature to it, that timelessness and the sameness of the events, the wonderful repetition of the events, is thrown up against the really extraordinary difference that we face this year. Just one measure of where we are this year compared to last year.
On June 6th this last year, the Dow was at 12,200. On Friday, it was at 8,760. And as I said, I'm going to talk about the impact of that downturn on MIT in just a little while. But despite the rigors of dealing with that, we have been on an enormously productive track. The new administration in Washington has very articulately stated that it values science, analysis, research. These are the things that MIT values.
On inauguration day, our lecture halls were filled with people watching the inauguration remotely. And at the point when President Obama said that it was time to restore science to it's rightful place, everyone leaped to their feet and applauded. And my guess is there are not too many places in America who received those words the way they were received here at MIT. What I often say about this administration, and my great hope is, it will be cool to be smart, what a relief. And--
And the hope in this is that the values that MIT holds so dearly will become more mainstream as our nation grasps the importance of science and technology, innovation, entrepreneurship for our future. More substantively on April 27th, President Obama gave a speech at the National Academies. And he made a very strong declaration of his support for science. And I'm just going to read one passage from what is a truly amazing speech. And I would recommend it to all of you.
He said, "At such a difficult moment, there are those who say we cannot afford to invest in science. That support for research is somehow a luxury at a moment defined by necessities. I fundamentally disagree. Science is more essential for our prosperity, our security, our health, our environment, and our quality of life than it has ever been." We believe this here.
And perhaps most important, President Obama embraces a knowledge based innovation economy. This is the principle of growth economics that was laid out so cogently by Professor Robert Solo, a Nobel Prize winner here in our economics department. Restoring the economy of the country will demand the same kind of fuel that powered the information technology and biotechnology revolutions. And the basic concept is as you invest at the federal level as generously as you can in research, that research turns into innovations. Those innovations are taken off campuses, out of research labs, to start new companies. And out of those companies come entirely new industries. This is a perfect assignment for MIT.
We have many opportunities to serve in Washington right now. And we're doing so in a number of ways. Members of our faculty have been secunded from our campus down to Washington to have leadership roles in the OMB, on the Council of Economic Advisors, at the Department of Defense. As I said, many of our faculty, but also many of our alumni are serving in government roles. There are a large number-- there's a large number of people from the PhD program in our Department of Economics serving in the administration.
Two members of PCAST, The President's Council of Advisors on Science and Technology are MIT faculty. And I participated with President Obama in a briefing on green technology opportunities in mid-March. Following that, we hosted on campus Congressman Ed Markey, Carol Browner, and the Science Advisor, John Holdren, for a session on the energy legislation that has been moving through the House. And in mid-April we hosted-- sorry, the beginning of this month-- we hosted Secretary-- of May-- in May we hosted Secretary of Energy Steven Chu who gave the Coral Taylor Compton lecture here. Now both Secretary Chu's lecture and the session with Markey, Holdren, and Brown are available on MIT World, if any of you wants to watch that, as is my event with President Obama at the White House.
Now MIT is primed to serve in other ways, not just in Washington, but here on campus. And I will briefly review where we are on two of our large initiatives. One on energy and the other, at this wonderfully interesting interface, where life sciences is joining the physical sciences and engineering. For this audience, I don't need to describe the four urgent reasons why we need to be very serious about tackling the energy challenge. In terms of energy demand, we are on track to double energy use worldwide by 2050 and perhaps earlier.
Energy security is a pressing problem. We depend on undependable countries for a lot of our energy supply. And as far as the climate, there is general agreement that we are contributing to global warming. And there is recent data from MIT professor Ron Prinn's work and the MIT Joint Program on Science and Policy of Global Change that's showing a marked acceleration beyond what had been predicted in terms of global warming. By the way, you can get to Ron Prinn or the Joint Program's website very easily and see the data that they've been studying.
And now, of course, we have the added incentive, shall we say, of economic downturn. And I believe that fundamental economic growth is really the only path to getting our nation out of the economic difficulties we're in right now. MIT's Energy Initiative, we call it MITEI, was officially launched two and a half years ago. The first phase of this, we relied mostly on industry partnerships. And we've raised over $200 million to fund research and education in energy. The Energy Initiative, from its very beginning, has been radically cross disciplinary. This means it reaches across all of our five schools, maybe not every department, but wow, close to every department is participating in some way.
At this point Ernie Moniz, the director, and Bob Armstrong, the deputy director of the Energy Initiative, estimate that about 20% of our faculty are participating in the Energy Initiative. Our students are, as you might anticipate, wildly excited about this. There is a mostly graduate student led Energy Club, one of several student energy organizations. The student led Energy Club, the last time I talked about it, I was told it had 1,500 members, but that data was a week old. And some student shouted from the audience-- no, it's 1,700. It's a lot, there are a lot of members.
Most are students. And these are graduate students who intend to be energy professionals. It's an inspiring group. So the Energy Initiative takes us from policy and economics to concepts for new buildings and cityscapes, to, of course, basic science and new technologies. We have roles to lead, both in policy and in research, but also in education. For those of you who are interested in some of our policy pieces, if you get to the METEI website, you can get the link to the policy pieces we've already done on the future of coal, the future of nuclear, the future of geothermal.
There's a new report evaluating the cap and trade system in the EU. And we've got a couple of new reports underway, one on the nuclear fuel cycle, which will be absolutely critical, understanding what we're going to do with nuclear waste if nuclear is going to be part of the energy solution going forward. Another study on natural gas and one on solar. And those new studies will be coming out in the next year or so. On the research frontier, there are no silver bullets. We do not believe that there is a single technology that is going to get the world from our current mix of use to a mix of use that will actually sustain the new demand and will reduce the amount of CO2 we push in to the atmosphere.
So we are pursuing a portfolio of technologies. We divide them into the transitional and the transformational. There are technologies that we use today that simply have to be improved. And they can be improved in the short term. And we're working very hard on those kinds of technologies. And the transformational are all those marvelous alternative technologies that we dream about using, but we're not ready yet to use them at scale.
Now, while I say there are a lot of different technologies under development, over these last couple of years, the focus of the Energy Initiative has really come to rest on solar. And solar, both in terms of materials for conversion, and, of course, the wonderfully exciting area here at MIT is the development of new storage technologies. There are over a dozen new battery technologies under development now at MIT.
Now people say, well, why solar? I mean, besides that it's such a great challenge. I say, well, Willie Sutton robbed banks because that's where the money was. And solar, that's where the energy is. And one of our faculty who does work on solar said, well, we should be using all that energy for doing something more than just heating asphalt. I think we can figure that one out.
I mentioned the huge student interest. In the education domain the Energy Initiative has a task force on education that brought forward this year a very exciting new minor in energy. Most of our majors and minors at MIT exist in departments or in particular programs. The energy minor, the governance of which was approved by the faculty at our May meeting, stretches across, again, all five of our schools and reaches into many, many departments. And will provide our future energy professionals with the kind of education that will support them from the economics of energy, to the science of energy, to the design of energy. And they will be able to really lead, we believe, as they leave MIT. That minor will be rolled out this fall.
As I mentioned, the first stage of the Energy Initiative was largely powered through industry partnerships, because we didn't have confidence at the time that we could secure the kind of funding that we would need from the federal government. So we had wonderful industry support. I mentioned over $200 million. The Department of Energy has changed its direction, happily. And we are doing a very good job at securing the new funding that's coming from the Department of Energy. We have also been the enormously grateful recipients of some important philanthropic gifts.
And the reason why private philanthropy is so important for the Energy Initiative is there are areas that industry, and frankly, the federal government just are not interested in supporting. We don't get a lot of support for education. So the energy minor is going to really rely on private philanthropy. And the other place where it's hard to get support from either industry or the government is support for new professorships. We have, believe it or not, some gaps in where our professors cover, and so we would like to have five more, five new professorships in energy, two of which have been generously funded by alumni. So that's a very quick tour of the Energy Initiative.
Let me turn next to this exciting area, where the life sciences is coming into conversation with the physical sciences and engineering. And it's happening all over campus. About a third of our faculty in the School of Engineering are either using the tools of life sciences or collaborating with the life sciences in some way. And it is not all about biomedicine. One of the most exciting battery projects that we have going on is a project that Angie Belcher is leading, from the Department of Material Sciences and Engineering and the Department of Biological Engineering, that has discovered a way to make viruses incorporate battery materials and then self assemble into the anode and cathode for batteries. Just one example of tools of the life sciences being used outside of biomedicine.
One place that is very much in the biomedical domain, where this convergence with life sciences, and engineering, and the physical sciences is happening in our new cancer activity. Those of you who have walked along Vassar and Main, see another big building going up. That is the Koch Institute for Integrative Cancer Research. It's a very exciting part of campus. If you look at the buildings that sit around it and what goes on inside them, this is really where the life sciences and engineering are coming into conversation. It's between the Ray and Maria Stata Center, where the Department of Engineering and Computer Science sits, and the Department of Biology, in the Koch Biology Building. It is in the center of what I call "the Great Circle."
This new building is a very interesting advance on what MIT has done in cancer. People often don't know that MIT has worked in cancer. But our Center for Cancer Research was established in 1974. The leader then was Salvador Luria, a phenomenal biologist, who had already won the Nobel Prize. Luria, with the Department of Biology, recruited 12 biologists into the Center for Cancer Research. Of those 12 biologists, four have won the Nobel Prize so far.
In addition to great science appreciated by the Nobel Prize committee, the fundamental discoveries underlying the new targeted cancer drugs came out of MIT's Center for Cancer Research. But when I spoke with Tyler Jacks, now Director of the Center for Cancer Research, or was the Center for Cancer Research, about where the future for cancer research at MIT was moving, he described incredible, exciting activity that was going on, where the life scientists and engineers were beginning to talk to one another in incredibly productive ways.
So it will not surprise you that the Koch Institute for Integrative Cancer Research, which now puts the Center for Cancer Research in our history, will be populated by about a dozen cancer biologists and a dozen engineers, who are interested in helping us figure out how to diagnose, treat, and prevent cancer in new ways. It's very, very exciting. It's exciting for the faculty. The students have enormous interest. And we believe that this is an approach that really is almost uniquely MIT, and we have a unique way to approach it.
People have asked me, why do we think this combination of biology and engineering can really help in our fight against cancer. And one of my responses, of course, is well, we're already doing it and making huge advances. But another is that it's been done in other fields. If we look at heart disease and stroke, over the last 25, 30 years, in the United States, we decreased death from heart disease and stroke by 63%, an astonishing product of investments, mostly out of the National Institutes of Health. And if we look at what were the critical discoveries and the critical advances, it's both biology and engineering. It's been both drugs and devices. The biology was understanding the metabolism of cholesterol and developing drugs to make cholesterol metabolism work better in those who didn't, so the statins. And it's also devices, drug eluting stents, these devices that prop open clogged vessels. So we believe that the same kind of approach in cancer will result in the same kind of wonderful advances.
The building, as I said, is rising fast. This afternoon, there's going to be an event for a wonderful plan to raise some money to help us support that building. And we do need a lot more support to get that building up and get it working well. Part of that building will be the Swanson Class of 1960 Biotechnology Center. We're going to be celebrating that at an event this afternoon. I invite you to join us. It's going to be at the Broad Institute at 3:30. And what the Swanson Biotechnology Center will be is a celebration of the life of Bob Swanson, as I said, and MIT alumnus, the founder of Genentech, and truly a 20th century legend. As I said, if you'd like to join us, please do.
Let me now turn to the question that really is on everyone's mind, although there are other things that are so fascinating. And it's this question of the economic downturn and our finances. Of course, it affects MIT. In December, I was describing a situation where all of our revenue streams would be compromised. Now I say that most of our revenue streams will be compromised. Certainly, we are anticipating decreased revenues from our endowment, from tuition, and from gifts. But one of our revenue streams clearly has not, and doesn't look likely to be compromised, at least in the near term. And that's research.
The Provost, Rafael Rife, and I, and Terry Stone, the EVP and Treasurer, have been focused very much on our research volume, which was frankly declining for many years. This year, for the first time in I think about five years, we have an increase in research volume, by about 6%, very good news. And that's before the kinds of programs that are coming out of the ARRA or the stimulus bill, or the really great new budgets that the administration has put forward for the science agencies.
Let me talk a little bit about our endowment. All large institutional endowments have lost value. We have been planning for an endowment decline on the order of 30%. Last year, July 1st, our endowment stood at about $10 billion. So we have been operating toward an endowment that would be $7 billion. The fact is, we got a report from our Chief Investment Officer, Seth Alexander, this week, and he says, right now, we're tracking about 20%, maybe 25% down. That's very good news. Because a couple months ago, it really was 30%. And we don't know we're going to finish there. But we're going to continue with the planning assumption of an endowment value loss of about 30%.
This means, in practical terms, that we have to take $150 million out of a $1 billion General Institute budget. And our plan is to do that in two or three years. We are going to take 5% out. We have taken 5%, actually almost 6% out for the year going forward. And we have a number of activities going on to get the next 10% out, including something called the Idea Bank. Many of you may have actually contributed to the Idea Bank. This is a web tool where people can enter their ideas for how we might do things better, more efficiently, more economically. And you can then vote on ideas. It's been a very interesting online conversation.
We also have put together, have charged a task force for budget planning. The task force has almost 200 people on it. They come from across the Institute-- faculty, staff, and students. And the task force has nine working groups, each of which is focused on a particular area of our activities. And they really do span everything we do across campus. The initial report from that group will come out at the end of this month. The final report in October. So that we can implement some of those, or hopefully, many of the ideas that are coming out of the task force for the budget for FY11.
Let me talk about tuition a little bit. The financial need of our students, obviously, is increasing. Last year, kind of midway between semesters, about 20 students came forward because their families had suffered a decline in their financial situation. This year about 40 students did. We're planning to increase our financial aid budget for next year by about 10%, to make it just over $80 million. So we expect that we will be increasing our expenditures in financial aid, even while the funding that supports it has been declining.
In the realm of gifts, obviously we rely so, so deeply on private philanthropy. Gifts fund any number of things that we do that we just can't fund any other way. And I would say the good news is not as bad as we expected this year. So thank you, all of you, you have been generous in ways that I would say I could not have anticipated. Last November, December, it has really been an extraordinary year.
We measure our philanthropic progress in two ways. We look at cash that's come in. And then we also look at pledges for new gifts, pledges for future commitments. In terms of cash, we're down only about 5% from last year. So this is very good. Most people are fulfilling on the commitments they've made in the past. But as I look forward, I'm a little bit more nervous about looking forward, because our new gifts and new pledges for those gifts are down on the order of about 25% from last year.
Now I say that, but you have to remember last year, we had an extraordinary year. So if we compare where we are with gifts and pledges to a three year average, we are down on the order of 8% or 10%. So we're doing fine, but we have to really think-- incorporate into our budget planning an anticipated less powerful stream of gifts than we have in the past. There have been stories of just extraordinary generosity this year that are enormously heartwarming and enormously encouraging.
One of our alumni, who is enormously generous for his capacity, every year he gives us $1,000. And it is-- we have considered this at the limit of his capacity, actually a little bit beyond it, wrote a lovely letter to us in December saying that the economic downturn caused him to really think about what was most important to him, what really mattered in the world. And so this year he sent us $2,000. Amazing.
And one thing that has come through many of these stories this year is that while a lot of markets have lost in value, the market for meaning has really increased in value this year. And MIT, I think, really does deliver on real value. I talked about our response to the budget reductions. I have to tell you, and you will all know this, this is not easy. We're putting, I think, a very important positive face on it. We think it's an opportunity to do things that we couldn't do before. We just didn't have reason enough to do before. But it just is simply, it's an unhappy process.
Any decision we make will not be greeted unanimously with approval. And I think the community has done a wonderful job of just being quite practical and quite hardheaded about the things we need to do. We want to come out of this with an Institute that is leaner and stronger, that is more strategic. We have to preserve our core strengths and values. It's an opportunity for us to actually think about what those are.
I will call out just two of them. One is financial aid. We are committed to retaining our policy of need blind admission and need based aid in meeting the full need of our admitted students. We have to do that, because one huge source of strength at MIT are the remarkable students we bring to the campus. We also are going to continue to hire faculty. Because it is important to refresh our faculty by bringing young people into the Institute. And that's something we're going to continue to do as long as we possibly can.
How do we sustain this kind of activity? Obviously, I've been talking about your generosity, but private philanthropy has been a huge part of it. We launched our Campaign for Students last October, just as the market was starting to tank. They tell me every time that MIT launches a campaign, that's what happens. We should stop doing that. The Campaign for Students was designed as a birthday present for our 150th birthday in 2011. We're actually doing quite well on it. We've raised about 65% of our target, which is $500 million. And I know that many of you have been generous in contributing to the campaign. And I really can't thank you enough.
The campaign's target is students. We're raising money for undergraduate financial aid, for graduate fellowships, for money to enhance our educational environment, and also for the living environment for students. We've got a lot of work to do on all four of those areas. And the Campaign for Students will accelerate it.
So let me just close with a description of one huge source of our momentum. And that is our spectacular students. We have now admitted our class of 2013. The numbers scare any one of us who got into college years and years ago. Applications this year went up 17%, to a record high. Good news in terms of the world seeing MIT as a place that's going to pave a path to the future. What that means, of course, is that our admission rate was very low. We admitted 10.6% of the students who applied.
Yield, 65% of the students we offered admission to have decided to come to MIT. And this is quite amazing, considering the environment. A number of our competitor schools have stopped using any kind of early admission program, which means students are applying to more and more schools. And so the face off when they've been accepted in a lot of schools, to decide for MIT and not another school is very, very much tougher. And also, as you all know, there are some very generous, very aggressive financial aid policies in some of our schools, which frankly, we just can't match for some of our students.
Any way you described this class takes your breath away. 90% of them graduated in the top 5% of their classes. 20% are academic all stars, meaning that they won some national or international competition. 20% are stars in music, the arts, or athletics. But my favorite statistic, always, is this one. 19% of next year's freshman class are the first in their families to go to college.
Now this year's class, as you know, is the first class that I admitted as freshmen. So I know a whole lot of them. And as they came up to shake my hands yesterday, it was all I could do to keep from just saying, can you please, you're so wonderful, can you stay. But no, the world needs you, go ahead. But they are remarkable students. And I can tell you, I've met a lot of those 19% who are the first in their families to go to college. And these are remarkable people. And I often say, and many of you have heard me say this, that MIT is a place where the American dream comes true.
And they come to MIT. MIT changes their lives. And then they go out and change the world. This year's graduating class showed me that again and again. This year's Student Awards are breathtaking. In terms of the elite national awards, we have had a year like none other in the past. Two of our students won Rhodes Scholarships, four won Marshall's. 13 won Fulbrights. We've never had a year like this. It just shows you that the world appreciates the kind of student that comes out of MIT.
Four Marshall's, let me just give you a sense of that-- that's 10% of the Marshall Scholarships awarded this year, came to MIT students. These students just heighten our sense of mission, and responsibility, and stewardship. And to serve them and those to come, we have to maintain an unwavering focus on producing the innovations and the innovators, even in, or I would say, especially in difficult times like these today.
In pursuing that mission, we depend absolutely on the loyalty, the enthusiasm, the generosity of our alumni. You are indeed our information army. You carry the spirit, the values, and the good of MIT into the world. And I can't thank you enough for that. Let me stop now and take some questions. Thanks for coming to Technology Day.
MEAD WYMAN: [INAUDIBLE]
SUSAN HAWKFIELD: Whatever you say, Mead. Short, okay, one word answers, yes and no questions. Yes, sure.
MEAD WYMAN: The first question, I'm going to go from the general to the specific, the first question is, strategically, as we look ahead, will MIT continue to concentrate in these two areas of energy and integration of bioengineering, or do you have some new ideas?
SUSAN HAWKFIELD: Yeah, so these two ideas are very big ideas. And I think these two ideas will continue for, my guess, is a decade or more. But even while these are two of the major initiatives that reach across campus, there are probably hundreds of fantastically interesting and important projects going on. And so we shouldn't think that these are the only things that MIT is doing, nor that they will be the only initiatives going forward.
You will hear about some exciting stuff that's going on in the brain sciences, in the intelligence world, and everywhere around us there are brilliant new activities cropping up.
MEAD WYMAN: Thank you. I'm going to try to do this one, because I think it's the least controversial. There was a lot of interest in the energy information that you provided. Here's a question, which is, if we're convinced about global warming and are considering a major alteration in the economy, and the production of greenhouse gases through cap and trade, why doesn't MIT advocate for a serious study of a contrarian view, to see if there are other ways that we might improve the output or reduce the output of greenhouse gases?
SUSAN HAWKFIELD: So this is a bit controversial, and I don't mind talking about it. But I always preface this by saying, I am not the expert. So I rely on the experts at MIT. For those of you who are deeply interested in the science behind this, I recommend the website for the Joint Program on Climate Change. It has lots of data, lots of analysis by very, very interesting and thoughtful scientists.
In terms of other approaches, the other thing I'd recommend in terms of what kind of economic and policy approach should be taken, Denny Ellerman did a study-- I mentioned this-- on the EU cap and trade, which is very interesting. Obviously, the purpose of the study was to determine whether it can be effective, whether that's the direction we should apply at MIT. So the energy, the MITEI, the website has a huge amount of data on it. And we are interested in pursuing any number of alternatives, because we don't know that we have a single right answer yet.
MEAD WYMAN: Okay, fine, thank you. The last question then would be, it seems that the transformational approaches to alternative energy are on a very long time frame for development. What can be done to accelerate that?
SUSAN HAWKFIELD: So what can be done to accelerate it is to fund the research. They are on it a long time frame. Everyone dreams about an all solar to all electric energy equation. We're a long way away from that. We're a long way away from it because the conversion technologies are too expensive and insufficiently developed. Storage technologies are not ready for prime time yet either. One of the things that I find on sobering is the national investment in energy research.
The budget for basic R&D annually at the government is $4.5 billion. That's a lot of money, you can say. But the kinds of really world changing innovations that have come out of the health sciences come out of an NIH budget that now is about $30 billion a year. So our investments have been small and they've been episodic. So we've been on a roller coaster of federal investments in energy research. And you just can't make progress that way.
So we have to commit to a sustained, high level of funding of R&D. The companies aren't going to do it. The federal government has to do that. But on the industry side, we also have a fascinating anomaly. That if you look at the investments of, let's just say, pharmaceutical companies in R&D. They spend somewhere between 15% and 20% of revenues on R&D. And a lot of innovation comes out of those investments. If you look at the energy sector, on the other hand, energy companies, American based energy companies, invest one quarter of 1% of their revenues in R&D.
So we have, on the one hand, if you look at what has happened in biomedicine over the last 30 years, it is miraculous. But to get those kinds of miraculous changes, you need to invest, federally at the level of about $30 billion a year, and the companies also need to be investing. In the energy domain, we just haven't had the kind of focused, sustained investments, either in the private sector or in the government to really change the game. So follow the money. We really as a nation have to commit to changing this.
I'll tell you one other sobering insight into our country's interests in changing the energy equation. Of the 14 or so companies that have signed up to be our industry partners that are funding the kinds of research that I think we need to make solar viable, to come up with new kinds of storage devices, new kinds of batteries, only two of those companies are American based. And it's not that we haven't been approaching the American based energy companies as aggressively as we've been approaching the non-American companies.
So just as a nation, we really haven't focused on this in a way that would create the kind of transformation that I believe we need. We need it for the economics, as well as-- just put climate change aside, there are plenty reasons that we need to change the energy equation.
MEAD WYMAN: I'm afraid we have to stop. It's such a pleasure to have you spend time with us. And I wish we could spend more time, but I'm afraid we have to--
SUSAN HAWKFIELD: Thank you, Mead, it's always a joy.
MEAD WYMAN: Thank you very much for coming.
Our next speaker is Professor Rebecca Saxe. Rebecca is an assistant professor of cognitive neuroscience at MIT. Her main area of research is in the theory of the mind. And this morning, Professor Saxe will take us into how the brain creates the mind. Rebecca, welcome.
MEAD WYMAN: Can I help you with anything?
REBECCA SAXE: No, I'm good.
Thanks so much. Thank you, Susan, for that amazing introduction to MIT. Thanks Mead, and thanks everybody, for coming to listen to me. So on quite a different note today, I'm going to talk to you about the problem of other minds. And the problem I want to talk to you about is slightly different from the standard problem of other minds you might have heard about as an undergrad at MIT.
The old problem of other minds, the one that puzzled philosophers, used to be, do other people have minds? How can we know whether other people have minds? Because it's possible that maybe you have a mind, but everybody else in this room is just a really convincing robot designed by the MIT Media Lab. And I think after Patrick Winston's talk today, that might not seem so implausible.
But for the moment, I'm going to assume that the answer to this question is yes, other people have minds. Many people have minds. Many of them are in this room. And so we're not going to talk about that today. A second problem of other minds, one that puzzles novelists, and parents, and teachers, and spouses, is why is it so hard to know other minds? Or actually, why is it so hard to change other minds?
Why is it so hard to figure out what people want, and what they think, and to change what they want, and to change what they think? And I think this problem is often put most eloquently and specifically by novelists. So for example, in one of my favorite statements, Philip Rothen, in American Pastoral wrote, "And yet what are we to do about this terribly significant business of other people? So ill equipped are we all to envision one another's interior workings and invisible aims."
I think that's a really eloquent statement of this problem we all face. And as a teacher, and as a spouse, I think this is an incredibly important problem. But as a scientist, it's actually a different problem that I'm interested in, which is, why is it so easy to know other minds? So in particular, I want to pose to this problem. I want to convince you that it's a puzzle why it's so easy, and that it's a puzzle that we're beginning to start working on.
So in particular, it's so easy for all of us as human adults to see, and to recognize, and to think about other minds in a number of ways that are quite remarkable. So one way is from very little information that you might get from a scene or a moment, you can make really accurate and powerful inferences about what somebody is thinking. So you can guess what this woman is thinking from just a momentary exposure to one image of her, or what this man might be thinking.
And you might say, well, that's because you recognize those images, you've seen things like that before, you've been in those places. And so an important insight from social psychology is that's not true. We can take all of that away and give you just a little bit of movement. This is six seconds of geometric shapes moving. And from this you can see relationships between the two characters. You can see the concern of the triangle, the nervousness about jumping, the decision to jump. The genius of flipping the trapdoor before the evil circle comes in. All of that, out of six seconds of the movements of two geometric shapes.
So that's one sense in which it's amazingly easy. We get lots and lots of information from apparently very little in the stimulus. But another way to be amazed at how easy it is to think about other minds is to think about what it is that our brains are having to do. So we, as human adults, can figure out a problem like this. Juliet doesn't want to marry Paris, and she asks the Fryer, who tells her she can pretend to be dead. And the Fryer sends a message to Romeo, who doesn't get it.
So when he gets the other message, he thinks Juliet is dead. So when he sees Juliet, he decides to kill himself. We can follow that kind of crazy complicated logical stream, which is kind of amazing. Or I can say a sentence to you, which has one mental state embedded inside another, inside another. Like I hope I'll convince you to believe that the brain does what I think it does. That's four mental states, one inside another.
And as a neuroscientist, what I try to figure out is, well, how does the meat inside your head make sense of the sentence I just said. So in particular, we know a little bit about the structure of the brain. We know that it's a group of cells called neurons, connected to one another. And they can send really simple messages from one to another. So a neuron fires. That is, it sends basically an electrical spike down to the next cells to which it's connected. Every neuron is sending many, many, many messages out. Everyone is receiving many, many, many from the neurons to which it's connected.
And the pattern over space and time as those spikes come in determine whether the next cell sends a message on or not. So we have a huge network of cells sending these incredibly simple messages. And this is the huge explanatory gap of neuroscience. Is how do we get from these simple messages being passed from one cell to another in an enormous network to constructing the meaning of this sentence with four mental states embedded inside one another?
And the honest answer is, we have no idea. So neuroscience, the project of human cognitive neuroscience, is to try to bridge that gap, to try to go from the way neurons function. Which is pretty similar between humans, and monkeys, and mice, and actually sea slugs, those pretty similar function of neurons, to the really unique capacities of the human brain, in terms of the cognition and the mind. And so that's a project of human cognitive neuroscience, which is using amazing modern technologies that let us study human brains directly by taking pictures of them without cutting holes in anybody's head.
So that's my research. And what I'm going to talk to you about today is the research I've been doing in one particular area, which is looking at this kind of thinking that we do with our minds, thinking about other people's thoughts. And the first thing I want to tell you about that is that, although it's easy for us now to think about Romeo and Juliet, it doesn't start easy. In fact, we can watch children slowly and effortfully building up an idea of how their minds work. And watching that developmental history in childhood can give us clues about what to look for in their brains.
And so to show you about that, I'm going to introduce you to a task that we do with children. This is called the false belief task. And in this task, this child is going to be asked to think about and make predictions about what somebody else will do, given their thoughts. And in particular, this is a five-year-old, who's going to be asked to do this. And what I'm going to show you after that is a three-year-old. I'm going to show you a well-known transition between three and five. So we're going to start with a five-year-old, who's pretty much got it.
- His name is Ivan. Do you know what pirates really like?
- Pirates really like cheese sandwiches.
- Cheese? I love cheese.
- Yeah. So Ivan has this cheese sandwich, and he says, yum, yum, yum, yum, yum, I really love cheese sandwiches. And Ivan puts his sandwich over here on top of the pirate chest. And Ivan says, you know what, I need a drink with my lunch. And so Ivan goes to get a drink. And while Ivan is away, the wind comes, and it blows the sandwich down onto the grass.
- And now, here comes the other pirate. This pirate is called Joshua. And Joshua also really loves cheese sandwiches. So Joshua has a cheese sandwich. And he says yum, yum, yum, yum, yum, I love cheese sandwiches. And he puts his cheese sandwich over here on top of the pirate chest.
- So that one is his.
- That one's Joshua's, that's right.
- And then his is on the ground.
- Yeah, that's exactly right.
- So he won't know which one is his.
- Oh, so now Joshua goes off to get a drink. Ivan comes back, and he says, I want my cheese sandwich. So which one do you think Ivan is going to take?
- I think he's going to take that one.
- Yeah, you think he's going to take that one. All right, let's see.
- I told you.
- Oh, yeah, you were right. He took that one. So now, Ivan has his cheese sandwich. And now Joshua comes. And Joshua says, I'm really mad at you Ivan. Why is Joshua mad at Ivan?
- Because he took his.
- Yeah, he took his sandwich, that's right.
- He didn't know which one was his, because he didn't know the wind was going to blow.
- That's right, he didn't know. That's exactly right.
REBECCA SAXE: Okay, so that's a five-year-old all over the problem of somebody else having a false belief. But if we look just about a year and a half earlier, this is a three and a half year old who's heard the same story. He's going to have quite a different answer.
- And now Ivan's coming back. And Ivan says, I want my cheese sandwich. Which sandwich is he going to take? Do you think he's going to take that one? Let's see what happens. Let's see what he does. Here comes Ivan, and he says, I want my cheese sandwich, and he takes this one. Uh-oh. Why'd he take that one?
- His was on the grass.
REBECCA SAXE: So this three-year-old, first of all predicts that the pirate will take the sandwich that's really his, he'll take the one on the grass that's really his. He won't not take the one, where he left his. And then faced with contradictory evidence, even when he sees the pirate taking the wrong sandwich, the three-year-old doesn't think of the possibility that it's a mistake, or that he has a false belief. The three-year-old instead comes up with some other reason why he might want to take the other sandwich. In this case, he might want to take the other sandwich because his was on the grass.
So developmental psychologists for the last 30 years, since this pattern was first observed in the 1980s, have been studying this transition, trying to understand what's the difference between the mind and brain of a three-year-old and a five-year-old, such that the five-year-old is all over what somebody else might be thinking and what that will cause them to do, whereas the three-year-old seems to be missing this huge aspect of other minds, namely the possibility that they might have a false belief.
Now what will probably come as no surprise to parents in the room, but took developmental psychologists a lot longer to get to is that five-year-olds are not done developing their ability to understand other minds. They've still got lots to learn. And in particular, what we've recently started doing is asking children a slightly different question. And when do this, we see quite a different and much later developmental trajectory.
So in this case, we're going to ask children not just what somebody will do, but to make a moral judgment about their action. So I'll show you first what the three-year-old says.
- Here comes Joshua, and Joshua says, I'm really mad at you, Ivan. Why is Joshua really mad at Ivan?
- Because he take his sandwich.
- Yeah. Now is Ivan being mean and naughty for taking Joshua's sandwich?
- Yeah. Should Ivan get in trouble for taking Joshua's sandwich? Yeah, all right.
REBECCA SAXE: So it's maybe not surprising that the three-year-old thinks Ivan should get in trouble, because he doesn't understand that Ivan took the sandwich because he thought it was his. But now here's the five-year-old. And remember, this five-year-old with all over understanding that Ivan thought that the sandwich he was taking was his. But look what happens when I asked him the moral question.
- Joshua's really mad and he's crying. Was Ivan being mean and naughty for taking Joshua's sandwich?
- Uh, yeah.
REBECCA SAXE: And it's not until seven that children start giving the answer the grown ups would give.
- That's right. He's taking the one on the box. But uh-oh, here comes Joshua. And Joshua says, I'm really mad at you Ivan. Why is Joshua really mad at Ivan?
- He thought that he took his sandwich.
- Oh, so can I ask you something?
- Should Ivan get in trouble for taking Joshua's sandwich?
- No, because the wind took it.
- Oh. Is Ivan being mean for taking Joshua's sandwich?
REBECCA SAXE: My favorite version of this was one when I asked him when his mother was in the room should Ivan get in trouble, and he looked straight at his mother and said, no.
So what we can see here is two pieces of a really complicated developmental story, where children are learning more and more how to think about other people's minds and use that to make decisions. So between three and five, the children are learning how to predict what somebody will do given their false beliefs. And it's not until seven that they start to integrate those ideas into more complicated judgments, like moral judgments.
So those patterns in the development give us a hint of what to look for. And when we now go back to the human brain and ask, well, what brain regions are the children and even we, as adults, using when we're trying to solve these kinds of problems? So to study this, we put human adults into an fMRI scanner. And we asked them to solve the equivalent kind of problem, but now made a little bit more appropriate for grownups.
So here's an example. Somebody would be lying in the scanner. And they're reading a little story like this. John told Emily that his car was a Porsche. Actually John's car was a Ford. But Emily doesn't know anything about cars. And then we ask, true or false, Emily thinks John's car is a Porsche. So this is more or less pirates and cheese sandwiches for grownups. And then we can look at, well, what regions of their brain are showing more metabolism, more activity, while the grownups are reading these stories, compared to very similar logical problem solving that doesn't require them to think about anybody's thoughts.
So these are the results from what we call typical human adults. They're MIT undergraduates. And you can see that there's a group of brain regions shown here in the bright colors that show more metabolism when these undergraduates are thinking about other people's thoughts, compared to when they're thinking about other logical problem solving. Just to reassure all of you and me, we did recently do this with real adults and it works just the same.
I'm going to talk today about one of these brain regions that I've been studying quite a bit, which is called the right temporal parietal junction. It's above and behind your right ear. And this brain region seems to be playing some very specific role in the human capacity to think about other minds. Just to give you a sense of how reliable this is, these are the brains of just three individuals who walked in off the street. We gave them 12 stories to read about beliefs, 12 stories to read about other things. And in each one, you can see the same little brain region is lighting up.
This is kind of amazing from a neuroscientist perspective, because that kind of reliability, having the same brain region show up in each individual who walks in off the street, is the kind of thing we'd seen before for visual brain regions that let us see, for motor control brain regions that control our actions. But it's really rare to see something that's systematic and reliable for a function as complicated and as abstract as thinking about other people's thoughts.
One of the things that we've been doing now is studying how these brain regions grow in the life of a child. So as you saw, there's reason to believe, based on the development, that these brain regions, or at least their functions, are maturing really slowly over the time course of development. And so we can take advantage of that, to try to watch that development happen in real time. This is a project we just started this year. We scanned our first group of children.
Here's the feedback we got from one child. We showed him his brain. He said afterwards, I discovered my brain fills up my whole head. I'm really smart. Another child, gratifyingly said, I can't wait to go again. Which is actually wonderful, much more enthused than MIT undergrads were. These are the children sitting in front of our practice scanner, where they practice being in that environment.
And what we know so far is that first of all, the same brain regions are-- we can see them in children thinking about somebody's thoughts as we do see in adults. So that's good. We can find these brain regions and study them. But interestingly, they're not exactly the same in the children as they are in the adults. So what I'm going to show you here is just a measure of how specialized this brain region is. Does it really do just thinking about other people's thoughts? Or does it do some related other things as well?
So this is going to be on a scale from 0, not at all specialized, to 100% specialized. And in adults, this brain region is almost entirely specialized for thinking about other people's thoughts. It does almost nothing else. But in eight and a half to 11-year-olds, it's somewhat less specialized. And between five and eight and a half year olds, it's even less specialized. So while we can find this brain region and start to study it, it seems like it's very slowly acquiring this highly specialized function in the human cognitive life. So that's very intriguing and one of the things we're hoping to follow up in the next few years.
But another thing we've been looking at is, in the adults, we've been looking not just at when does this brain region get recruited, but also what does its recruitment, its activity, predict, in terms of more complicated or real life decisions that people might be asked to make. And so, in particular, we wanted to ask adults to make moral judgments, just like we asked the children, and see if we could see this brain region predicting the kinds of moral judgments that people would make.
So we designed-- again, this is a grown up version of the pirates problem that you saw before. This one might even be set at MIT. So in this problem, participants again will be lying in the scanner, and they'll read a story like this. Grace and her friend are taking a tour of a chemical plant. And while they're on tour, they stop for coffee. And Grace is making the coffee. And her friend asks for some sugar in her coffee.
And next to the coffee machine is a jar of white powder. And this white powder is actually a deadly poison. But the jar is labeled poison. So Grace thinks that the powder is poison. Grace puts the powder in her friend's coffee. Her friend drinks it and dies. And the participants are asked, was it morally permissible for Grace to put the white powder in her coffee? Now, just to check, does anyone think it was morally permissible? Because if so, I'm getting my coffee somewhere else.
So we can change little versions of the scenario and ask a slightly different question. So for example, in some cases the jar might be labeled sugar. So it's a white powder next to the coffee in a jar labeled sugar. In that case, when Grace thinks the powder is sugar, that's pretty reasonable. And now she puts it in her friend's coffee and her friend dies. And we ask the participants, was it morally permissible for Grace to put the powder in her friend's coffee? And hopefully, more of you now think it maybe was morally permissible for Grace to put the powder in the coffee.
Now again, we can do it the other way. The white powder by the coffee might actually be sugar but be labeled deadly poison. And in that case, when Grace puts the powder in her friend's coffee, her friend is fine. But we can still ask, was it morally permissible for Grace to put the powder in her friend's coffee. And again, I hope you have the intuition that it wasn't morally permissible.
So these are the moral judgments that we collected from participants when we just asked them these kinds of questions. So what I'm going to show you here are, across the bottom are the cases where the friend is fine, versus when the friend dies. And the blue bars are when Grace thinks the powder is sugar. And the red bars are when Grace thinks the powder is poison. So if the powder really is sugar, and Grace thinks that it's sugar, everybody says it was morally permissible to put it in the coffee. That was totally okay.
But if Grace thought the powder was poison, even though it really was sugar, people say it was completely wrong to put it in the coffee, even though nothing bad actually happened. And by contrast, if Grace thought it was sugar, even though she killed her friend by accident, people say it was mostly okay, not entirely okay, but mostly okay to put that in the friend's coffee. And then in the case where it was poison and she knew it was poison, people say it was definitely not okay to put it in the coffee, which is good news.
So now we can look at, well, what's going on in people's brains while they're making these decisions. And in particular we took-- the one most interesting case in some respects was the case where Grace accidentally poisoned her friend. So she thought the powder was sugar, but then her friend died. And when we asked this question, how much should Grace be blamed for accidentally poisoning her friend, people actually sort of disagreed. Some people say she should be blamed not at all. Some people say she should actually be blamed somewhat.
And we can look at, well, what's going on in this brain region in those people while they're making that decision. So what I'm going to show you is different people, each person is going to be a dot. Along the bottom I have put how much activity there was in this brain region. And then on the vertical is how much blame did they say that Grace deserves. So if they say more blame, that'll be higher up. Less blame, that'll be lower down.
And so what you can see is that people whose RTPJ, this brain region, have lots of activity, were the same people who were really paying attention to Grace's belief, that she thought that it with sugar, and saying she deserved very little blame. So at the far end, they have very high activity in this brain region and they assign very little blame, paying most attention to the belief. Whereas on the far left side, we have people who have very little activity in this brain region. And they're the people who are paying very little attention to the belief. They're saying she deserves a lot of blame because her friend died.
So it looks like activity in this brain region predicts quite a consequential fact, namely how much moral judgment people will assign-- how much moral blame people will assign for accidents. But of course, a much stronger proof of that that's really what's going on in this brain region would be not just to observe that correlation, but to directly interfere with function in that brain region, and try to see if we could thereby interfere with those moral judgments.
There's basically two ways that you can interfere with the function of a human brain region. One is that you can wait for a case of injury. So you can wait for a case of a gunshot wound or a stroke that selectively impairs that brain region. But another way you can do it is a technology called transcranial magnetic stimulation that allows us to pass a magnetic pulse through the skull into a healthy individual and transiently disrupt the function of that brain region.
So I'm going to show you a video of TMS. The first thing I'm going to show you is I'm trying to show you that it's a magnetic pulse. So the first thing you'll see is me putting a quarter on the coil that delivers the pulse. And then you'll hear a click sound. The click sound is when the machine goes off. And when the machine goes off, you'll see what it does to the quarter.
There's the quarter. Okay, it's quite an effective way to fling quarters. So now we're going to apply that same magnetic pulse to my skull. And because it's not easy to see moral judgment, we're going to do it to the part of the brain that controls my right hand. So this is Marina positioning the coil over my skull. And she's using this brain image of my brain to figure out where she should position it so that she can get it located over the part of my brain that controls my right hand. And then, again, you'll see what happens to my hand.
REBECCA SAXE: We were trying to get me to flip the corner. All right, so you can see that whenever the pulse goes off, it causes a small, but significant contraction of the muscle in my right hand by causing those brain cells that control my right hand to fire. So now we can do a similar thing to people who are making moral judgments.
In this case, of course, we're not going to change their right hand, we're going to try to interfere with this brain region that they use for thinking about other people's thoughts. So what I'm showing you here are the moral judgments of these same people while they're receiving TMS, but they're receiving it to some other brain region, a brain region we don't hypothesize to have any role in moral judgment. And as you can see, they're making pretty much the same judgments that I showed you before.
The judgments are they say it's okay for the blue bars, when she thinks that it's sugar. Not okay for the red bars, when she thinks that it's poison. And it doesn't make as much of a difference whether the friend was fine or dies. It matters more what Grace thought. So now we can ask, well, what happens to those moral judgments if we've been applying TMS to this brain region, the right temporal parietal junction.
And so the first thing is, we don't interfere with their ability to do the task at all. And so this is their moral judgments after TMS to the right RTPJ when Grace thinks the powder is sugar, and it is sugar, and those judgments are unaffected. Now, if we take the case where Grace thought that it was poison, but it was really sugar and her friend was fine, normally people say that was really totally not okay. She thought it was poison and so it was really not okay to put it in the coffee.
But after TMS to their RTPJ, they say it was significantly more okay. So it was more okay for her to put what she thought was poison in her friend's coffee as long as it wasn't really poison. And then we have the opposite case, where she thought that it was sugar, but it was really poison. And people say that-- adults say it's mostly okay to put something you think is sugar in the coffee, even if it turns out to be poison. Remember, that's not what the five-year-olds say.
And so after TMS to the RTPJ, we can make adults look more like five-year-olds. They say it's less okay to put something you think is sugar in the coffee if it's really poison. When she thinks it's poison, and it is poison, they still know that that's really not okay.
So what I hope you've learned from my talk today about how the brain events the mind is, in particular, for adults, thinking about other people's minds is amazingly easy given what we know about how the brain functions and prompts a big challenge to neuroscientists to come up with an explanation of how a bunch of neurons wired together and firing together can give our brains the amazing capacities they have.
But it wasn't always easy. The discovery of other minds is a very slow and protracted process in childhood. I say slow, but sure, because one of the amazing facts is, of course, all children come to it. They just come to it slowly and effortfully. Almost all children. I want you to remember that there are special brain regions in the human brain that seem to play some special function in this task. And in particular, these brain regions are an exciting place to look for figuring out how neurons accomplish this job. That this brain region matures slowly over the course of childhood. In fact, even the 11-year-olds that we studied don't have fully mature brain theory of mind brain regions.
And finally, that by changing this aspect of brain function, we can also change your mind, in this case, the moral judgments that you make. But I want to give the last word to the novelists, and parents, and teachers, and spouses, via Philip Roth, in American Pastoral, who concludes after the question that I started with that "The fact remains that getting people right is not what living is all about anyway. It's getting them wrong that is living, getting them wrong, and wrong, and wrong. And then on careful reconsideration, getting them wrong again." Thank you.
MEAD WYMAN: Thank you, Rebecca. That was quite fascinating. And by the way, I'm very much impressed with what it generated in terms of questions, quite a few. I hope-- I hope we can have time to use some of these. Let me ask you this one first. Does the development of the brain as we grow up tell us about how the brain evolved?
REBECCA SAXE: That's a great question. In some ways, we assume so, and in some ways we hope so, and in some ways probably not. So the way in which we assume so is, there's an old principle of biology, that development or otogeny, the course of development of an individual organism, in some sense recapitulates biologeny, recapitulates the evolutionary history. So this is an idea that will be familiar to people who know that, for example, a fetus in a very early stage of development has a tail. So there's aspects of evolutionary history of an organism that are encoded in the development of that organism.
And so in some sense, we could hope that that process and those signatures continue on all the way throughout development. And so by studying what develops first, versus what develops later, we might be able to study something about what was evolutionarily old or what was evolutionarily new. So that's the sense that we assume so, is that by extension on a biological principle.
The second answer is that we hope so, or rather that it's an empirical question. So insofar as we can, we're trying to simultaneously study both development and the evolutionary history of the human brain. And see where do those two independent empirical programs converge on a similar answer. So one case is in these brain regions that I study. I showed you evidence that functionally they mature very slowly in the course of childhood. Another thing about those brain regions is that they're among the most expanded in the human brain, relative to other mammalian brains. And so it seems like those two independent empirical clues can help us zero in on the mechanism of evolution and the mechanism of their function.
And then the third sense, probably not, is that another aspect that, of course, hugely influences child development is culture, and parenting, and all the environmental factors that govern the growth of children. And those factors will surely influence the way that the human brain develops, and in ways that can't perfectly recapture the evolutionary environment.
MEAD WYMAN: I hope I can do all of these justice. Here's the next one, is the RTPJ ever found on a person's left side?
REBECCA SAXE: That's an interesting question. So when we look for all the brain regions in the brain that seem to be playing some special role in thinking about other people's thoughts, there's a brain region on each side, one on the right and one on the left. In the people that we study, we see that the right, the brain region on the right is more specialized for thinking about other people's thoughts than the one on the left. Now, we, by design, study only right handed people. That's because most right handed people-- all human brains pretty much have slightly different functions on their right and their left. And for most right handed people, language functions are on the left.
And for left handed people, it's 50/50. And so when we're trying to study things that might be different on the right and the left, we study right handed people, because we think that will give us the best bet of seeing consistency from one participant to another. My left handed friends are actually very annoyed by this. They never get to have their brains scanned. So we mostly study right handed people. And in those people, the right region seems to be more specialized.
So we don't know, could we see the opposite in left handed people? That's one question. Another interesting observation from language, which is much better understood than social cognition, is if an adult damages their left hemisphere, the brain regions on their left that support language, the adult loses the capacity to speak and understand language. A massive stroke in your left hemisphere severely damages language production and language comprehension.
But if an infant has the same kind of stroke, around the time of birth-- in fact, sometimes infants are born with almost no left hemisphere whatsoever-- those infants can acquire normal language. And that language function is supported by the right hemisphere. And so it seems like around the time of birth, the two hemispheres, although the left is the preferential target of language, the two hemispheres are more balanced in terms of what they have the capacity to acquire. And it's through the course of maturation that these functions get assigned sort of more permanently to one side or another.
And so there's every reason to believe, although no direct evidence, that in a child with a massive stroke to their right hemisphere, that they could, in fact, acquire normal social cognition on the left, as long as that damage happened early enough in life. But that's only an inference. There's no direct evidence of that yet.
MEAD WYMAN: The questions keep coming. And they're really quite good. I'm not sure we can get through all of them. But there are several questions about the impact or the study of autistic people. And I think what I do is just combine two questions here. And you can kind of work on it as you like. Any thought, as the writer says here, any thought to studying autistics in this area? The second one says, people with some forms of autism, I believe it's called Asperger's--
REBECCA SAXE: Asperger's
MEAD WYMAN: Asperger's Syndrome, lack the ability to read indications from others' faces. Have you studied the brains of these people?
REBECCA SAXE: Yeah, those are great questions. So one of the reasons that the ability to think about other minds became a target of a neuroscientific investigation, not just a psychological investigation, was the observation that children with Asperger's and autism have selective difficulty on the kind of false belief problems that I showed you these, the three children solving. So very soon after those tasks were first invented, scientists in England observed that kids with autism and Asperger's had unusually-- had unusual difficulty with those problems relative to very similar logical problems. And that was the first indication that there might be separate parts of the brain responsible for solving those problems, that could then be separately disabled by disease.
And so it's been an interest all along in the process of studying how these brain regions function in typical adults and typical children to also understand the time course of atypical development. So the study that I showed you, where we were scanning typical children was actually just the very beginning of a large program project that I'm doing in the Department of Brain and Cognitive Science, with my collaborators, Professors Nancy Kanwisher and John Gabrieli, and supported by the Alison Medical Foundation. And the goal is a massive study of typical development and development with autism.
We're aiming to study 100 typically developing children and 100 children with autism. And to specifically look at the development of these brain regions between the ages of 5 and 10. Of course, the hope is that since these brain regions are related to thinking about other people's thoughts, and children with autism seem to have a selective difficulty in thinking about other people's thoughts, we'll be able to identify the neural basis of that difficulty. And then maybe use these brain responses, either as a measure of interventions, like a way of measuring whether interventions are successful, or as a target of interventions, a way of deciding what will be an effective treatment for autism. And so that's definitely part of our plans for the immediate future.
MEAD WYMAN: There's quite a bit of interest in how your research might be applied to different categories of people. And with this one, I'll give a plea to perhaps be very careful in the way that you state the question. The question is what difference does sex play in your findings? I think what the person meant to say was--
--what difference does gender play in your findings?
REBECCA SAXE: I guess I shouldn't be making MIT jokes in response to that question. So we're very interested in whether-- in the broader question there. So I talked a minute ago about disease, about whether the differences in this brain region is related to differences that cause disease. But there's another question, which is, are differences in these brain regions related to just differences among individuals, differences between people, all of whom count as normal, especially around here.
And so there's many dimensions of differences between people that might be related to differences in brain function. I mean, of course, all differences in personality are going to come down to differences in brain function in one way or another. And so we've recently started a sort of survey, where-- so, by now, we've collected these brain regions, the theory of mind brain regions, we call them, from almost 1,000 people over the course of the last 10 years.
And belatedly, we're now starting to collect a lot more information about those people as we're scanning them, so of course, age and gender, but also personality trait information and psychiatric history. And we're interested in whether we can see correlations between differences between people and differences in the function of these brains. So until recently, I always said that I'd never seen a gender difference in these brain functions. And that was certainly true in the smaller samples we were looking at. But now that we've started to look at a much bigger sample, we actually are starting to see small gender differences, and related actually to differences, reported differences in empathy between individuals.
And the difference is interesting. It's that, on a personality questionnaire, women report greater empathy on average than men. And that's related to this brain region is actually showing greater activity, which you can interpret as working harder in people with lower empathy than people with higher empathy. And so as a consequence, we see more activity in men than women. But that difference is totally explained by the empathy difference. And so once you match people for their reported empathy, then men and women still look the same.
MEAD WYMAN: Thank you. I'm going to work a little bit on this theme of different categories of brains. This is one that clearly was proposed by someone who must be close to me in age. It says, have you researched what happens to this decision making or capability in much older minds? Is there a deterioration of understanding in making moral judgments?
REBECCA SAXE: I'm so outnumbered in this room. You can't ask me that question. So we've been looking at it a little bit and in various different ways. We have looked at moral judgments on our moral questions in younger and older participants. In that case, we found that the older participants tended to be a little bit more punitive. They tended to say, if you killed your friend, well, and made us a little worried. But for the most part, when we look at the brain functions, what we see is actually striking similarity in the location and function of these brain regions from the five-year-olds we've studied up to the 60-year-olds that we've studied.
Recently, we've begun studying more grownups, partly because we've started studying congenitally blind adults. We were interested in how these brain regions developed in people who had never seen another person's face, never seen a human action. And we were interested in whether that would cause these brain regions to develop really differently, suggesting that their development depends on seeing people, or whether actually these brain regions might develop very similarly, if it's just thinking about other people that's necessary. Blind children certainly think about other people just as much as sighted children.
So we were scanning a population of adults who were born blind. And they ranged in age from 30 to 60. And so we also scanned a group of sighted people, who ranged in age from 30 to 60. And so, as I mentioned, the older adults, the older sighted adults, showed the same brain regions with very similar functions as the five to 10-year-olds that we had been scanning. Really interestingly, so did the congenitally blind adults. So in these adults who've never seen anybody, these brain regions still have exactly the same position and function as in somebody who got visual exposure of other people.
MEAD WYMAN: There are a number of, again, situational categories that people are wondering if that's given you different experience or different experimental outcomes. The questions are more or less around is there a difference between left handed and right handed people? Do you see any difference in children or adults from different social backgrounds or family backgrounds that perhaps may have been abusive or nurturing, or versus nurturing? And finally, maybe I'll lump together with this, and you can handle it as it makes sense. When sociopaths are studied, are their brains different, in terms of this recognizing or understanding what's going on in other peoples' brains?
REBECCA SAXE: This is great. I feel like I'm being asked for my entire undergraduate course.
So, a simple way to answer this question would be to answer the most specific version. So to say, in terms of these brain regions and this function, what's the difference for kids from different backgrounds or for sociopaths versus typical people. And that would be easy to answer, because the answer is, we have no idea. We haven't done any of that yet. This is a new research program.
One of the things I'm not sure I remembered to say is that this technology in this field is less than 10 years old. We're just giving birth to an entire science. It feels like being in the Wild West. You can try anything. Nobody has any idea what's going to be the most productive way to do this science yet. Which is a really exciting time to be a young scientist, although occasionally, there's a sense of vertigo.
So when we look to related fields, where people have asked these kinds of questions, the most related field is the developmental psychology that I showed you in the videos, where you don't ask directly about the brain development, but you ask about the development of the thinking ability in children, again from different backgrounds or different groups. There's a few really intriguing hints in the literature about that.
So one thing that I find really intriguing is there's a little bit of evidence that kids with a really good ability to think about other people's minds are actually somewhat protected from the effects of parental abuse. So if you look at kids who start out with either a really good ability to think about other people's thoughts, or not so good ability, and then you look at the consequences for their lives of living with abusive parents, the kids who were really good at thinking about other people's thoughts are protected from the negative consequences. So that's very intriguing in terms of thinking about theory of mind, among other things, as a self-protective or a compensatory strategy that lets us protect ourselves from other people, not just lets us engage with other people.
One interesting way to think about psychopaths-- there's a little bit of neuroscience research on psychopaths. Actually, this research is always fun to read, because they compare psychopath murderers to normal murderers.
In our population of normal murderers--
So when you look at that, again there isn't direct research on this topic, but the hints are that their ability to know what other people are thinking isn't impaired. What's impaired is their ability to care. And so, it's interesting to think of those as two different parts of the problem. One is sort of the problem solving part, figuring out what somebody else's motivations would be. And the other part is being emotionally engaged by that. Those might be supported by different neural systems. So, I'll stop there.
MEAD WYMAN: The questions were really very interesting. We can't get to the bottom of them. Thank you very much for responding to them. And thank you very much for the presentation.
Our next speaker is Pawan Sinha, who is an Associate Professor of Vision and Computational Neuroscience at MIT. His area of expertise is in understanding how the brain recognizes objects in the visual world. This morning, Professor Sinha will give us some insights into how we learn to see. Professor Sinha, welcome.
PAWAN SINHA: Thank you. Thank you. It's a true privilege for me to be here and to be able to share with you a journey that my students and I have been on for the past few years. It's a journey that's in part scientific and in part personal, but it's also a journey that illustrates how MIT's impact is felt in the lives of people who might be thousands of miles away from our campus.
So let me start by introducing the people who have been gracious companions. These are my students. That's Ben Ballas, who recently graduated with a PhD, and that regrettably is not the interior of my lab.
Ethan Meyers-- Ethan is the one in the middle.
Scott Garland, not in India. And Ming Ming and Yuri Ostrovsky. Ming is a postdoctoral fellow, who will be taking up a faculty position at Dartmouth this fall. And Yuri is a graduate student, who will be finishing his PhD this fall. And here they are, pretending to be good Indians in New Delhi. And Meg Molson, who's a postdoctoral fellow. And Tappan Gandhi, who's is a doctoral student. And in the middle is one of our subjects.
So the journey begins with a very simple observation. And that is, that we are highly visual creatures. It's hard to imagine any aspect of our interaction with the real world that might not benefit from the visual modality. This has been the domain that has interested us greatly. And specifically, as Mead mentioned, the question of abiding interest to us and to many scientists in the general domain of neuroscience is, how does the brain learn to see?
This is a question that has deep implications for many areas of neuroscience and beyond. Understanding the process of development might help us better understand the mature visual system. It might also help us develop better diagnoses and interventions for conditions that lead to problems in visual processing. And visual processing might just be a gateway for understanding many other kinds of processing mechanisms in the brain. The brain might use the same kinds of principles of operation, not just for vision, but for other modalities as well. And finally, very practically, and as I think Patrick will exemplify, understanding the brain and how it learns to see can help us make autonomous computer vision systems.
So the question to my mind and to most of our minds is undoubtedly important. And then the challenge becomes, how can we go about studying this question? So whenever we talk about a problem of learning, how does the brain learn to do X or Y, the impulse is to work with this beloved creature, the human infant, and to try to study how an infant begins to acquire different capabilities, in our case, different visual capabilities. This is certainly a very direct way of approaching the problem, but it has all the kinds of challenges that you might expect.
Babies are not interested in being our experimental subjects. They would rather sleep than give us good data. And this is something that I learned first hand with the birth of Pam and my son. This is Darius at the age of five days. So when various was born, I thought this is going to be such a gold mine of data.
But very quickly I was dissuaded of that idea, both by Darius and by his mother.
So this approach has its challenges. What else can we do to study visual learning? One other approach is to work with adults, normal adults, but to put them in settings that they're unfamiliar with, to show them bizarre, new kinds of objects that we have designed in our favorite computer graphics program, such as the one that you see here and to ask them to learn to recognize objects from this class. And this approach has yielded some interesting insights. But the results, as you might imagine, are contaminated by the tremendous amount of experience that an adult is bringing to the table.
So it's fair to say, I think, that the approaches that we have on hand have significant limitations. An alternative approach is this, to work with children who have sight onset late in life. So what kinds of children might these be? These would be children who have been allowed to remain blind for several years, even though their blindness is treatable. So certainly, we cannot ethically create a population of the children. But we can perhaps rely upon some quote, unquote, natural experiments, finding individuals who, for one reason or another, have been left blind, have been left untreated, even though their condition was curable.
The problem with this approach is that such cases are extremely rare. They're almost impossible to find. In fact, it's estimated that over the past Millennium, fewer than 20 such cases have been studied in any detail. So it's hard to come up with a systematic scientific program using this approach. And now, with the greater development of medical resources, especially in the West, these cases are getting harder and harder to locate.
So is this then just a thought experiment? Something that's fun to consider, but hard to implement? And for me too, it was just a thought experiment, until I began to look a little closer at the country. I grew up in. That's India. And what I discovered was somewhat startling. And I want to share with you some of the facts and figures about childhood blindness in India. And what I'm going to tell you is going to be somewhat of a sobering counterpoint to the stories that you might have become more familiar with from the popular press about the growing economic clout of India.
So childhood blindness, or blindness in general, in India, it's a massive problem. India has the world's highest number of blind children. Well over a million children are blind. The incidence of childhood blindness is well over three times that in the developed countries. And if you include adults as well, then it's estimated that over 1 in every 100 Indians is blind.
What are the causes of childhood blindness in the country? There are several causes. Here are the two primary ones-- corneal scarring, caused by things like vitamin A deficiency, as you see on the left, and congenital cataracts, children being born with the condition that we typically associate with old age. And there are several other causes as well, some of which are listed here-- anomalies of the entire eyeball, anomalies of the retina, and so on.
And some, somewhat shockingly, some traditional remedies. What do I mean by that? What happens is, that many villages in the country are completely devoid of medical care. So a medicine man, who really does not have any background in medicine, he takes on the responsibility of prescribing treatments. And these treatments very often do more harm than good. And I want to show you one example of that.
Here's a three-year-old girl who had congenital cataracts. She was brought to the medicine man. And instead of suggesting that the girl be taken to the hospital for a simple cataract removal surgery, the person decides to drive the demons out of the child by burning her abdomen with red hot iron rods. So that's the level of ignorance that we have to contend with. Simple conditions like conjunctivitis, that just need eye drops, are worsened by the kinds of concoctions that are put into the eyes of the children. And the children go blind as a result.
Considering all of these causes, it turns out that nearly 60% of all childhood blindness in the country can be treated or prevented. But how much is actually treated? Less than 20%. And the reason for that is-- there are several reasons. One of the primary ones is that most of the current interventions, most of the current programs for treating blindness are focused on adults. And the quality is extremely dubious. And just to convince you of that, I show you here a few of the headlines just from last year.
These are outcomes of eye camps, treatment camps that are organized by private or governmental clinical teams to treat cataracts or other eye anomalies out in the field. And the success rate of these items is sobering, or the lack of success rate of sobering. This was just from 2008. There are several other causes of the untreated, of the vast population of untreated children. Fewer than 10% of all India's hospitals are equipped to do pediatric surgeries. Most of the ophthalmologists are concentrated in the big cities, New Delhi, Bombay, Calcutta, Madras. While most of India's population lives out in the villages. And many are desperately poor, so they can't make their way to the cities to get treatment for their children.
This untreated childhood blindness has devastating consequences. It hastens death. The life expectancy of a blind child is 15 years shorter than that of a sighted one. And that's for the lucky few who make it past early childhood. The childhood mortality rate amongst the blind is over 50%. Fewer than 50% of the children who are blind make it past the age of five. Less than 10% of these children get any kind of education. And less than 1% are employed as adults.
So India suffers a tremendous economic loss, if one were to think of it just in a purely mercenary manner. India is suffering a loss to the tune of $10 billion a year, which has more money than what's needed in order to tackle this problem at the root. But that's not happening and the problem stays with us. So it's clear that there's a humanitarian need that needs to be addressed. And it also became clear to me, when I saw these facts and figures, that this humanitarian need was in perfect synergy with the scientific need I had mentioned, of finding a child who can have sight onset late in life.
So this perfect synergy between the humanitarian and the scientific missions led us to launch Project Prakash. Prakash is the Sanskrit word for light. And we hope that through this project, we can bring light into the lives of children, and also simultaneously shed light on many of the deep problems of neuroscience. The humanitarian goals of Project Prakash are the obvious ones. We want to provide treatment to as many children as we can, and also to build awareness amongst the villagers and amongst the doctors in the cities as to the magnitude of the problem.
We started out fairly modestly on this humanitarian mission. I would go out to villages every summer or winter, whenever I'd head to India from MIT. Sometimes accompanied by my father, who most of the villagers thought was the real doctor and I was the assistant. And my father would go along with that. So I would examine these children. And with my limited knowledge of ophthalmology, I would try to identify cases of treatable blindness and take them to the hospital for treatment.
So it was well-intentioned, but a very minor effort, given the magnitude of the problem. So over the past couple of years, as the project has gained a little more traction, we have enlisted the support of hospitals in India, and have gotten them to launch outreach initiatives, where teams of medical workers can go out into villages or into schools for the blind and examine many children in one go. And I want to show you a short video clip of one of the examination sessions, of screening sessions. This is from a school for the blind in New Delhi.
- Good morning.
- Good morning.
- Good morning.
PAWAN SINHA: That is the case of microphthalmus, an eyeball that's too small. That's an extreme case of microphthalmus, called anopthalmos. It's hard even to discern the presence of an eyeball. These are permanent cases of blindness. We cannot treat these cases.
But every so often, we come across children who show sensitivity to light or the ability to tell what direction light's coming from. And that's the first indication that the condition might be treatable.
So after that field examination, we bring the children over to the hospital. This is one of the hospitals we're working with.
The hospital has world class pediatric facilities, developed in part because of a donation from the Ronald McDonald Foundation. So you see many posters of Happy Meals in the pediatric wing.
I don't know whether the lighting is good enough. But as we zoom into the eyes of this child, you will see the whites inside his pupils. And those are congenital cataracts.
The nystagmus is the searching eye movement that you're seeing. And that indicates that the blindness had a very early onset, within the first three or four months. The hospital has a good surgical facility for children. And now you're going to see the same child, three weeks post operatively. We remove the bandages from his right eye.
We've conducted many such screening camps and treatments, mostly centered around Delhi. And overall, well, these are some of the preoperative and post-operative photographs of three children. We've screened well over 1,000 children and provide treatment to over 200 of them.
Thank you. And there are also some other intangibles that we've been able to achieve, like bringing the problem of childhood blindness onto the radar screens of the people who can make policies. So the members of parliament, questions about childhood blindness have been raised in the parliament, directly because of the work that we've been doing. And the Supreme Court of India has now mandated that every child, before being admitted to a school for the blind, has to go through, has to be examined by a registered ophthalmologist. So there are no needlessly blind children who are ensconced in schools for the blind.
That's the humanitarian side of Project Prakash. The scientific aspect of the project is to understand how vision develops after the onset of sight. So what we do, in a sense, is this. Let's say we find a child who has light sensitivity or the ability to see what direction light's coming from. And we judge that this is a treatable case of blindness.
The child is provided with treatment. And the treatment is entirely unconditioned on any agreement on the child or the family's part to participate in the scientific studies. So we treat many more children than the children that we actually work with. There is no ethical dilemma about quid pro quo. After the child is treated, every week or so, we have the child run through a simple battery of visual tests to try to see what is the arc of visual development that they are exhibiting, which visual abilities are coming on first, and which are the ones that are following.
And this arc of development provides us a very rich source of information about detecting potential causal dependencies between the early developing skills and the later developing ones. And we do this follow up for as long as it's practically possible in order to determine what are the asymptotic levels of skills that these children have and what kinds of skills are they perhaps permanently compromised at. And that's very important to know in order to develop the right kinds of targeted rehabilitation procedures.
Very recently, we have gained access to a functional brain imaging system in New Delhi. And now we can combine our behavioral assessments with neuroimaging assessments. So we can begin to make structure function correlations. We can see as a child is acquiring or is exhibiting a particular proficiency, what are the aspects of brain function that have changed perhaps in order to bring about that behavioral change.
Some of the milestones that we've achieved so far. This is a fairly young project, as projects go. We've been doing this for about three or four years. So our accomplishments are, I would say, modest. But we are proud of the little bit that we've done so far. One of the big dogmas in neuroscience that comes to us from work with animals is the dogma of critical periods, that if the brain is deprived of sensory information for the first few months, or first couple of years of life, the brain is not going to be able to make use of that information, even if that sensory modality is restored.
And what we are finding is that that's just not true. Even individuals who are treated in adulthood, that they've been blind until the age of 20 or 30, you restore vision, and they can acquire very significant visual function. So this is beginning to challenge one of the main ideas in neuroscience. Some of the specific data that we've gotten about the time course of object learning has allowed us to develop computational models of visual learning. This is a paper on the development of face learning in these children.
And very recently, we have also begun to tackle some longstanding philosophical questions. Here's one of the big ones. It's referred to as the Molyneux problem. Over 300 years ago, Molyneux frame the question-- imagine that you have a person who has grown up blind, and while blind, this person has learned to tell the difference between a cube and a sphere by touch. And now through some miracle, as an adult, the person gains site. And you put the cube and the sphere in front of the person. Would this person, without touching the cube and the sphere be able to tell which is which?
This has deep implications for the empiricist and the nativist viewpoint, and also issues of how shape, or information in general, is encoded across different sensory modalities. This question has remained open for the past three centuries. And with Project Prakash, for the first time, we have data to actually answer this question. So any guesses as to what the answer is? How many of you think that the answer is a yes, that the person will be able to tell immediately?
Okay, so that's about 50%. That's exactly the right kind of answer I wanted. So about half of you think that the answer is yes. And the other half presumably think the answer is a no. So it's an interesting question to ask then. It's not a foregone conclusion. We find that the answer is a no. Immediately after a slight onset, children are entirely at chance at transferring the haptic knowledge to the visual domain. But interestingly enough, give them a week and they can do this at ceiling level, at 100%. So even though they are at chance on day one, on day seven, they are at 90% or 100%. So whatever is happening in the brain is happening very, very rapidly. And we are currently trying to understand what are the mechanisms of such rapid learning.
Our work has been profiled in a few scientific and popular venues. Here's an article from Nature. And here's one from Time Magazine. And this is gratifying, because ministers in the Indian government are not reading journals like Neurocomputing or even Nature. They are reading things like Time. So for us to be able to go to them with these articles, it's much greater-- it carries much more weight than going to them with a stack of scientific papers. So it gives us more leverage.
One of the projects that we have been working on recently, that I wanted to tell you about, just in a little more detail, because it touches on a few other directions in the lab, is the project related to visual integration. And let me first illustrate what the problem is that we are tackling. This complex landscape that you're observing is an image, where I have represented the brightness of each point in the image with height. So the hills and valleys correspond to how bright or dark the corresponding region of the image is.
And I've done that just couldn't make the point that in an image, there are no easy cues for telling you where one object ends and the other one begins. If I were to ask you to parse this into distinct objects, it's a very difficult task. And yet, we do this all the time. So what you see on the bottom left is the original image. And it's easy for us to say where the boundaries of different objects are, even though the information that we're getting is extremely complex. How does the brain learn to do this task of visual integration, of putting together the world into meaningful chunks?
Here's another way of saying the same thing. Given any image, such as the one you see on the left, in the fine structure, this image is made up of many different regions of different colors, different brightness. Somehow the brain can put together subsets of these regions and come up with meaningful assemblies. And we are interested in understanding how this happens. And we are using Project Prakash in order to gain some insights into this developmental process.
I'll show you data from three children, because they span a nice age range, from seven to 29. But we have gotten similar data from about 13 additional children. So we feel very confident about what I'm going to mention. So after these children were treated, we had them run a very simple battery of visual tests, probing their ability to parse a world meaningfully, or parse the images meaningfully.
So I'm sure it's hard to see in the back, but this is an outline of a circle overlapping a square, or a filled circle overlapping a transparent square, and so on, a colored square overlapping another colored square. So we show the children these images and we ask them to say how many objects do you see, where are they, if you can name these objects, please name them. And the results that we get are amazingly consistent across different children, irrespective of age.
So here's a short video of the 29-year-old, four weeks after treatment, trying to interpret these images.
- [NON-ENGLISH SPEECH]
- Next one, please.
- [NON-ENGLISH SPEECH]
- Okay, triangle, next. Yeah?
- [NON-ENGLISH SPEECH]
- Square, next. [NON-ENGLISH SPEECH]
- [NON-ENGLISH SPEECH]
- Triangle. [NON-ENGLISH SPEECH]
- [NON-ENGLISH SPEECH]
- [NON-ENGLISH SPEECH]
- [NON-ENGLISH SPEECH]
- [NON-ENGLISH SPEECH]
- [NON-ENGLISH SPEECH]
- Square and circle. [NON-ENGLISH SPEECH]
- [NON-ENGLISH SPEECH]
PAWAN SINHA: His visual motor coordination is kind of poor, but you can get the sense of roughly what he's trying to [INAUDIBLE]
- [NON-ENGLISH SPEECH]
- [NON-ENGLISH SPEECH]
PAWAN SINHA: He's very consistent.
- [NON-ENGLISH SPEECH]
- [NON-ENGLISH SPEECH]
- [NON-ENGLISH SPEECH]
PAWAN SINHA: One of my graduate students suggested that even if he had said yes, we would have subtitled this no.
We wouldn't have.
- [NON-ENGLISH SPEECH]
- [NON-ENGLISH SPEECH]
PAWAN SINHA: So it's interesting, he was able to recognize the triangle when shown on its own, but when embedded in the field of lines, he cannot. [INAUDIBLE] here. He fails. And just to put some numbers up, for control subjects, these tasks are trivial. They are at ceiling, at 100% performance on all of these tasks. These subjects are at floor. They perform very consistently, but they don't parse the world quite as we would parse it. So they are all at 0%.
And, in fact, with three dimensional objects, when we ask them how many objects do you see out there, they just see it as three different objects. The world is broken up into many pieces. In their tracings, you get the same sense of over fragmentation. So given the black figure, SK is tracing it as two L's and one little thing in the middle. And here is another child. I've asked him to copy this shape.
- [INAUDIBLE] Very good. [NON-ENGLISH SPEECH]
PAWAN SINHA: And now, I am asking him to do a trace on the star itself.
- [NON-ENGLISH SPEECH]
PAWAN SINHA: And what I want to point out is just how piecemeal that tracing strategy is. Every little segment in that figure is being traced as a separate entity.
- Okay, very good.
PAWAN SINHA: What does this have to do with real world recognition? So we showed these children images of commonplace objects. And what we found was that their recognition rates were fairly low, ranging from 18% to 34%. So very few of these images are being recognized. And we'll have a little more to say about this in just a moment.
So the images that they are unable to recognize, if you ask them to point to where the objects are, even if they can't name the objects, they point to each region of a different color or luminous. So for the cow, each black splotch becomes its own object. In fact, for a ball, even the shadow becomes its own object. So the world is broken up into many little pieces. And the brain seems to be unable to put all these pieces together.
However, if we introduce a different kind of a visual cue, the performance changes.
- [NON-ENGLISH SPEECH]
- [NON-ENGLISH SPEECH]
PAWAN SINHA: In the case, what you saw earlier. And now Ethan has set it in motion.
- [NON-ENGLISH SPEECH]
- [NON-ENGLISH SPEECH]
- Right. That's right. [NON-ENGLISH SPEECH]
- [NON-ENGLISH SPEECH]
- [NON-ENGLISH SPEECH]
- [NON-ENGLISH SPEECH]
PAWAN SINHA: So notice that the two figures are always staying overlapped. It's never the case that we are completely separating them. And yet, with motion, almost a magical change comes about. Just to compare the performance without motion, which is at 0%, and with motion, they are going up to 60, or 70, or 80%.
So now, back to the real world recognition. I said that the performance was quite poor, but it's still interesting to ask, given that the recognition performance is not zero, there are some images that these children are recognizing and others they are not. What is the difference between the images that they do recognize and the ones that they don't? So here for SK, one of the subjects, is the set of recognized versus unrecognized images. Outlined in red are the ones he is recognizing and the others he is not.
And just looking at these images, it is not immediately clear what's the difference between these two sets of images. They seem about equally complex. They're about has colorful. So what might be the rule? And we couldn't figure out what the possible rule was, until we began to see these results with motion. And then we asked, how about partitioning the set of images we are showing these children into images that depict motile objects versus those that depict immotile objects.
So if you do that, this is the partition that you get. In red are the images that people would rate as being motile. And in blue are the ones that we would rate as being immotile. And if you compare the recognized set for our subjects, in red are the ones that they're recognizing and in blue are the ones they're not recognizing. So it's a remarkable congruence between the criterion of motility and the recognition performance. Statistically, a very significant congruence. And we get the same congruence for all of the children that we have tested so far. There are exceptions, of course.
And this exception holds across all of the children that we work with. So here is--
- [NON-ENGLISH SPEECH]
- Coca-cola, Pepsi.
- [NON-ENGLISH SPEECH]
- [NON-ENGLISH SPEECH]
- [NON-ENGLISH SPEECH]
PAWAN SINHA: So I talked about MIT's impact, this is the impact of American marketing, 6,000 miles. You can go to the remotest village in India and there will be one hut that's painted either entirely blue or entirely red. And that's the Coke or Pepsi distribution site.
So here's the emerging pattern of results from Project Prakash. We find that the early stages of sight onset are invariably coupled with integrative difficulties. The world is over fragmented. These integrative impairments, we believe, compromise recognition skills. But motion provides important information regarding object segregation and integration to begin piecing together the world. And motion, importantly, also appears to create representations of objects that can then be used by the brain to recognize these objects in static images.
So we can ask, are these the initial stages of a learning program that will eventually lead the brain to being able to parse static images correctly? Or is this a strange learning pathway that will never allow these children the normal facility that we have with static image parsing? And we didn't have a way of addressing that question, until we followed these children several months after their treatment.
So here is SK, 18 months, post operatively.
- [NON-ENGLISH SPEECH]
- [NON-ENGLISH SPEECH]
- [NON-ENGLISH SPEECH]
- [NON-ENGLISH SPEECH]
- [NON-ENGLISH SPEECH]
- [NON-ENGLISH SPEECH]
- [NON-ENGLISH SPEECH]
- [NON-ENGLISH SPEECH]
- [NON-ENGLISH SPEECH]
PAWAN SINHA: And he can do some fairly complex ones also, but I won't take the time to do this. And we find the theme thing across all of our children. The time at which they begin to show these quote, unquote normal skills varies from child to child, and it's correlated with the age at which they were treated. But we find that the same progression holds. And these are just data from a few children. I'll skip through this.
So what tentative inferences can we draw from these results? The first one is this is the obvious one. Just given the fact that these children are learning, we need to rethink our notions of brain plasticity. The idea that the brain is plastic only in the initial couple of years of life, and then it somehow loses its malleability, that seems to be to be discarded. And the strict notion of critical period seems not to hold.
The second inference is a more specific one. Given the privileged position and the privileged role of motion that we are observing, and given what's been reported in the infant literature, where babies first become sensitive to motion cues well before they become sensitive to any other kinds of cues, cues of color, or brightness, or form, the hypothesis that we are formulating is that this privilege role of motion is not by happenstance, but rather motion serves a critical bootstrapping function for visual learning. This is the building block over which much of the rest of visual learning rests.
Ongoing work involves refining the concept of critical periods. So instead of just applying this blanket notion of critical periods on all visual skills, we are trying to figure out a more nuanced version of this idea, which skills are subject to a critical period and which are not. And from the second inference of the role of motion, we are taking two different directions. One is the development of a computational model of visual learning. This is a fairly ambitious undertaking, where we are trying to create a computational system, that we call Dylan, whose goal is to take in visual information, essentially unannotated raw video, of the same kind that a baby or a child and Project Prakash would get, and begin to discover automatically what are the meaningful chunks in this world. So to behave like a human child.
And the second is some tentative work in autism. Quite surprisingly, Project Prakash is leading us to some interesting hypotheses in autism. And I'll wrap up in just a couple of minutes. But let me mention what the tentative connections are. So here's what we find from Project Prakash. The early stages of site acquisition are accompanied by integrative difficulties. And there's a vast literature in autism which suggests that children with autism who might have otherwise normal visual function, experience integrative difficulties in analyzing visual and other sensory information.
In Project Prakash, we find that integrative impairments compromise recognition skills. And there are reports from the autism literature suggesting that autism is correlated with recognition impairments. But most interestingly for us, we are hypothesizing in Project Prakash that motion plays a critical role for the development of integrative skills. And if you look at the literature in autism, a few papers that have probed motion perception have found that motion perception is compromised, significantly compromised, in autism.
So the working hypothesis that we have is that at least some correlates of autism might be manifestations of integrative impairments, which might in turn be caused by a basic limitation in dynamic information processing. And here's where I'd like to request the lights to come down. I want to show you a couple of examples of tests of motion perception that we are conducting in our lab. What you're going to see here is first a control subject playing Pong, our home brewed version of this video game Pong, that hopefully many of you are familiar with. When I've given this talk to some of the undergraduates, they don't even know what Pong is.
So what you'll see is the child controlling the paddle with the mouse. Here's the ball that will bounce around. And in blue is our record of the child's gaze, where are they looking while they're playing Pong. So what's interesting to note is how anticipatory the gaze is. Well before the ball hits the paddle or the wall, the child begins to look at the point of impact. So they are able to make cues off the motion information so far to predict what the trajectory is going to be in the future. And they can bring the paddle to the right place in advance of the ball actually getting there.
Now let's look at a similar kind of recording of eye gaze from a child with autism. Notice how often the child follows the ball, rather than anticipating the trajectory. And other experiments that we are doing seem to bear out this idea, that the ability to integrate information, dynamic information over time, seems to be significantly compromised in autism. And I don't have time to flesh out the implications of that hypothesis. But it makes connections to many, many aspects of the broader autism phenotype.
So here's the causal chain that we are developing in the domain of autism. We eventually want to understand the hallmark deficits in autism, the social skill impairments. But we are trying to figure out what might be factors leading up to it. And at the moment, we find that, or we hypothesize that this is one of the key building blocks that might underlie many different aspects of autism. So project Prakash have unexpectedly led us to a potential hypothesis about an impairment that might explain some aspects of autism.
So to summarize Project Prakash has been this amazingly gratifying project for us. It has allowed us to have-- to be able to contribute to multiple aspects of our work, certainly in blindness alleviation. It has provided us insights regarding brain mechanisms of plasticity and learning. It has given us some clinically relevant hypotheses. It has seeded our work in computational modeling. And the students who have been involved with Project Prakash, the postdoctoral students, the graduate students, and the undergraduates, have all found this to be an important component of their MIT training.
The challenges for Project Prakash are significant. If you think of this green disk as all of the children that we have treated so far, this is what we face. This is what we've treated and this is the number of children awaiting treatment. So what we are doing is now scaling up our work by creating the Prakash Center for Children, which combines health care, education, and research all in one campus. So the idea is to have a hospital, a school where these children that we are treating can be educated immediately after they regain sight. And then we couple that with an on-site research facility.
And we've have decided to locate this in this little town called Rishikesh in the foothills of the Himalayas, because it has good access to the population centers of the country. This is the Gangetic Plains, which have about 35% of India's population. And it's desperately poor and desperately in need of medical facilities. And it doesn't hurt that Rishikesh looks like this. So every summer, it would be a pleasure to go over there and do this work.
So let me conclude by saying that Project Prakash has been an incredible opportunity for us, to not only tackle some interesting neuroscience questions, but also to directly impact for the better the lives of many children that we work with. Thank you very much.
MEAD WYMAN: That was really good. We have a bit of a dilemma here because of time. As the morning has gone on, your questions have gotten better and better. And it would be very interesting to ask Pawan these questions, but I'm afraid we just don't have time. We're going to have to hold ourselves to two questions. And it was very hard to figure out which ones, because your questions were, in fact, very good.
But let me give you two. And we will have to be very brief, because we have a limit at the end. There are two questions about the difference between color and shape. And I'll try to combine them. Maybe I'll just read both of them.
Does background color affect what the person that you're studying sees? It seemed dark backgrounds were recognizable, light or white not background. The second aspect of this, and you have to handle them, I guess, both together. Since color and shape both represent different pieces of one thing for people with restored sight, does levels of luminescence of light hitting the object, create a new piece or shape from the whole of these people? Do you understand what's being asked there?
PAWAN SINHA: Kind of, perhaps.
MEAD WYMAN: Well, make it brief.
PAWAN SINHA: Okay, I will make it brief. The first question is easy to answer. And that is the background luminous, even though it's an interesting observation, given the images you saw, that maybe dark backgrounds are easier-- objects are easier to recognize on dark backgrounds than on lighter ones, we find that that correlation does not hold. We have tested these objects on many different kinds of backgrounds. And that does not matter. What really matters is the criterion of motility.
And as for the second question, perhaps the questioner is asking, does luminous play an important role in the formation of these representations? And indeed, it does. It interacts in an important way with the dynamic information. So dynamic information links together pieces that have stood out. The initial parsing into the regions is based on luminous. And then those pieces get glued together via motion.
MEAD WYMAN: It's very interesting to see the impact of dynamism or change, motion, in your work, but I'm afraid we won't be able to handle this. There were a number of very good questions querying whether the impact of your study affected certain other conditions that we're more familiar with. For instance, and maybe we just have to make this one or two of the examples. I'm sorry for that. But we just--
PAWAN SINHA: I'm sorry for running over time.
MEAD WYMAN: No, that's okay. Does your research lead to any understanding of dyslexia or other cognitive diseases?
PAWAN SINHA: That in particular is an interesting connection. Dyslexia has been associated with an abnormality in what's referred to as the [? Magnal ?] system, so visual processing in the brain can be roughly divided into the [? Parabol ?] system and the [? Magnal ?] system, with the [? Magnal ?] system being more concerned with dynamic processing. And there are a couple of influential theories which suggests that dyslexia might be related to an impairment of the [? Magnal ?] system.
So we are following that up to try to see whether there is a connection between the children we are studying and the population of dyslexic children.
MEAD WYMAN: Professor Sinha, thank you so much for your presentation.
Patrick Winston is the Ford Professor of Artificial Intelligence and Computer Science at MIT. His area of expertise is currently in how vision, language and motor facilities account for intelligence. This morning, Professor Winston will tell us about a computer with common sense. Welcome, Patrick.
PATRICK WINSTON: I am a member of a Naval advisory committee, which takes me to San Diego for two weeks each summer. And I have the great privilege, therefore, of visiting the San Diego Zoo. In fact, two weeks from today, at about this time in the morning, I'll be making my entry into the San Diego Zoo. And my first stop will be the orangutans. I love watching the orangutans, because they're so human like in how they act and what they do. And I especially enjoy watching them stand on four legs. Because as you can see, their back legs are really short and their arms are really long.
And I look at them and I say to myself, if those fellows had just pitched back a few degrees, I would be today speaking to a large hall filled with people with long orange hair. And we would be the ones that would be in the zoos. After I spend some time with the orangutans, of course, I make my way to the chimpanzees. I especially feel obligated to visit the chimpanzees, because they are our closest cousins. And when I started in my field, we had no idea how similar we are to chimpanzees, but now we do know.
We know that from a DNA perspective, we are chimpanzees to about 97%, or maybe it's 98%. It keeps going up. I presume that it will--
Eventually-- I assume that eventually it will fall short of reaching 100%, so that we can preserve a little of our remaining dignity.
But I want to know why I am different from a chimpanzee. And I'd like to know why the humans that populate with our planet 100,000 years ago were not like us, even though they looked like us. It was only about 50,000 years ago that we, of the human species, began to do anything interesting. After about 50,000 years, we started building pyramids. Eventually, we wrote plays, did mathematics, built universities. And somehow, something has made us different and I'd like to understand what it is.
So I'll spend my allocated time discussing that with you. And in the event that some of it should seem dry, I feel obligated to make you a promise. And the promise is that when I'm finished, you'll be smarter than when I started. And by this, I don't mean that you'll just be more informed. What I mean is that you'll be smarter. So let me proceed then into the main part of the show.
I'll start with a articulation of a goal. This is an aggressive goal. And it invites several questions. One question is, why am I so optimistic? Another is, exactly what is a computational theory of intelligence? And a third is, exactly how long do you plan to live?
For this, the answer comes from biology. Biology teaches us that the energy producer in our cells are the mitochondria. We inherit the mitochondria only from our mothers. That's where all the mitochondrial DNA comes from. My mother is 98 and determined to outlive her sister, who had 103 is still a moving target. So perhaps my claim is not as outrageous as it might seem.
But onto the question of why be optimistic, one reason to be optimistic about this happening and having this happen at MIT, is because I sit at a roundtable with colleagues like Rebecca and Pawan, and many others spread throughout the Institute. So we have at MIT the ingredients for a great initiative against the deep scientific question of what is it that makes us humans different from chimpanzees, what makes us intelligent.
And, of course, we're encouraged in this kind of cross disciplinary activity, because our president is well known to have an appetite for large efforts against substantial social problems and scientific inquiries. But why are we in artificial intelligence and computer science at this table? It's because our perspective is a little different. We like to look at the problem of intelligence from the perspective of the kinds of-- well, let me put it this way.
If we're thinking about the difference between a computer and a person, there's quite a lot of difference at the bottom level. Computers are dry inside. We're wet inside. They have transistors. We have neurons. So down at that level, we're quite different. At the architectural level, we're still quite a lot different. Our anatomy is different from that of a computer. The architecture is different. As Rebecca and her colleagues continue to work away at that problem, we learned about the kinds of things that are very different in our own anatomy from what we would expect to find in a computer.
But eventually, we reach the computational level, where we ask computational questions. How is it possible to compute that? What is the nature of our universe that makes it possible for us to see how far something is? What makes stereo vision possible? All these kinds of questions are computational questions that have computational answers, that are to some extent independent of the kinds of apparatus that is at the very bottom level, whether it's a neuron or a transistor.
So we, who do this kind of work, are interested in what are the computational questions, what are the faculties by which humans think, and how can we understand those faculties at a computational level. One of those faculties, of course, as Pawan has emphasized, our vision system. And here's the first of my major claims. We think with our eyes.
In the early days of artificial intelligence, we regarded vision is merely an IO channel. But in fact, we think with our eyes. It's not just an IO channel. If you're not sure about that, let me give you a puzzle. How many countries in Africa are on the equator? I don't imagine that many of us would stake our life on our answer to that question. But let me ask you the question again. And now something extremely remarkable is happening. I've used my language apparatus to communicate with your language apparatus. And your language apparatus has instructed your visual apparatus to scan across that red line, counting the countries as it goes.
So there's a remarkable ability of our human language system to marshal and direct the perceptual faculties that seem not just there for IO, but therefore problem solving. Sometimes the kinds of things that we do with our visual system deal with things that aren't even there. There are things we imagine. So if I say to you, for example, John kissed Mary. And then I ask you a question did John touch Mary, you all know that it must have happened, because you have some sort of image, blurry perhaps, but you see that image of John and Mary locked in an embrace. And you can read the answer of the imagined image with your visual apparatus.
So vision isn't just an IO channel. It's the locus of very profound kind of problem solving. So as an engineer, how do I think about it? Well, I think that if I can't build it, I don't fully understand it. So using the kinds of clues that I get from the rest of the Round Table, I and my colleagues try to build systems that can see and use a vision system, not just as a recognition apparatus, not just as recognition apparatus to recognize faces or objects, but also as a problem solving mechanism.
So the kinds of experiments, the kinds of programs we write are programs that look into the world and assess what's happening out there. And here you see a photograph of a student of ours, Stephanie, who is being watched by a program, who is trying to figure out what she's doing. And she's going to do-- she's going to jump. And so you'll see a bar light up just over the jump arrow. As she jumps and you see that blue, that's a program that's looked into the world and decided that it's seeing a jumping activity.
Here's another. Stephanie's working with her friend. The friend is going to give the ball to Stephanie. Stephanie is going to take it. So you'll see both give and take light up. There goes the ball. And both give and take light up, because whenever you give somebody something, they take it from you. So that's an example of how we're trying to take what used to be thought of as a purely symbolic and linguistic kind of enterprise and connect it to the perceptual apparatus by which we interact with the world.
So we think with our eyes. We also think with our hands. I'm not so sure about that, but there are examples. You may have noticed in Pawan's last movie that the subject he was working with was doing a great deal of tracing around those objects with his fingers. We seem to use our hands a lot when we think. I once had a student advisee of mine who developed a severe case of RSI, Repetitive Stress Injury. So she was all locked up in a contraption with lots of Velcro and stuff. And she couldn't move her hands.
And I asked her if it really impaired her work a great deal. And she said, well, it doesn't do anything to my humanities subjects, because I can always dictate my papers. But the one thing I can't seem to do is mathematics. I thought that was interesting. She couldn't move her hands. And that impaired her ability with mathematics.
That's, of course, folklore at this level. But I was sitting at the commencement exercises yesterday waiting for something to happen. And my colleague, Bob Berwick, started telling me about an experiment, recent experiments, which have been conducted with small children who are learning mathematics, and instructed to use their hands as they work mathematical problems that are akin to algebra. And the ones that were instructed to use their hands do much better than the ones who were not using their hands.
So it seems in interesting ways that we think with our eyes and we think with our hands. And, of course, most significantly, we think with our mouth.
Let me give you the experimental evidence for that. Well, first the anecdotal evidence. I was talking with Danny [? Hellswell, ?] who, a lot like I was, was once a graduate student working with Marvin Minsky. And Danny came into my office one day and he said, have you noticed that Marvin has a short attention span? I said yes, I've noticed that. He said, whenever I talk to him and try to suggest an idea, he's usually figured out the idea within a couple of seconds. And I can't actually explain it. Have you experienced this? I said, yes, I have.
And then he said have you noticed that the idea that Marvin is guessing you have is actually better than the idea that you're trying to explain to him?
Yes, I've noticed that, I said. And he went on to suggest that-- he went on to ask, do you talk to yourself when you solve problems? And I said, yes, sure, I do. Well, what if that inner conversation we have with ourselves is an important part of the problem solving process, if it's actuating mechanisms in our minds, brains, that make it possible for us to recall and use information that we've learned at the time. That's folklore. Let's look at some experiments.
These are done by a community of psychologists, most notably Liz Spelke, who did some of the work than I'm about to report, when she was at MIT. You imagine a rat. The rat is in a rectangular room. All the walls are painted white. I suppose that's why they're black here. All the walls are painted white. And while the rat is watching, some food is placed in a small basket.
Then, the rat is spun around. And the question is, where does the rat go to look for the food? And the answer is, it's pretty smart, it only goes to two of the four possible corners. It goes to the two where the food could actually be after it's been disoriented. So it's able to use the shape of the room to decide which of the two corners to go to.
Now, if you substitute a small child, you get the same result. Usually, you substitute a toy or something of that sort, instead of the food. Still goes to the two corners. And of course, if you use an MIT person, well, you can go back to food again, but you--
You still get the movement to the opposite corners. And all we've learned so far is that, from the perspective of this experiment, rats and people are pretty much the same. I like to tell my students that we're just rats with big brains. But now we repeat the experiment with one wall painted blue. And for us, that breaks the symmetry. So, what happens to the rat. You spin it around, see where it goes. It doesn't seem to use the blue wall.
Even though if you put the food in a blue basket, it would have no trouble whatsoever picking out the right corner. But somehow, its having difficulty combining the rectangular shape of the room with that blue wall. And amazingly, if you use that small child, you get the same result. They can't seem to use the blue wall. It's only the MIT professor--
--that figures this all out. Now there are some questions that come up when you do these. And one is, at what age does the small child become an adult? And the answer is about five, which is shockingly late, I think. But you can ask the next question, is, well, what characterizes that time when the child becomes an adult? And it's the exact age at which the child has begun to use the words left and right in their own descriptions of the world.
Now, I had to say that carefully, because they understand left and right at an earlier age. But this is the age when they start to use left and right in their own descriptions of the world. It's if they have something in their linguistic apparatus, in their language apparatus, that makes it possible to combine the information from the blue wall with the information about the shape of the room to get the answer right.
But now there's one further thing that's done with this sort of work. You now ask the adult to undergo the experiment again. But this time while the adult is spinning around in a chair, you ask the adult to read something back to you that you read through the adult at the same time. So it's a little bit like simultaneous language translation, only it's English to English. Some people can do this very easily. Some people take a few minutes to get onto it. But it's something we can all do within a few minutes.
If you do that with a human adult, you reduce them to the level of a rat again. Somehow, you've jammed the language channel and the human adult becomes no better than a rat at figuring out where the food is. Since that's a professor, the undergraduates love it.
So that's rats, children, and adults. Can we make this more meaningful? I'll tell you about some of the experiments of Michelene Chi. Now I wouldn't normally tell you about work performed at another university, but I'll make an exception for Michelene Chi, because she happens to be married to an MIT graduate, and her son just picked up a degree from President Hockfield yesterday. So I'll allow myself to talk about Mickey's work. And I'll also allow myself, because it's, I think, very instructive.
She has people work physics problems. You probably all remember 801. They're problems like that. And she instructs them. And some are good at it and some are not so good at it. She divides them into two groups. The ones that are good at it and the ones that are bad at it. And as they work problems, he has them talk out loud. And she takes-- she makes a count of how often they say things to themselves while they're solving the problem. Some of the knowledge is just plain problem solving knowledge, like, oh, I'm stuck, or I need some help here. Other knowledge is physics knowledge. These forces must be equal, things of that kind.
So taking into account the best and the worst, measuring how often they say things to themselves, the worst, in the invisible white font there, on the average talk to themselves and say things to themselves about 10 times in the course of solving a problem. What about the ones who are best? 35 times. So it seems that the students who are good at this talk to themselves a lot. What we don't know is whether if a student is instructed to talk themselves a lot, they'll do better. But it's certainly the case that the ones who do do better have a tendency to talk to themselves a lot.
So what do we do about this as computer scientists? Well, we're MIT people, so we make models. In fact, I think that's the distinguishing characteristic of an MIT education. We learn to make models of everything. [INAUDIBLE], we'll make a model of that too. So if we're going to make models of that, what kind of models are we going to make? Well, one kind of model is a model of what kinds of things we humans say when we use language.
And we've developed a system that has 18 different ways of capturing the meaning of a sentence, not just the syntactical analysis, but what the sentence actually does by way of telling us about the world. Many of these sentences are sentences like that one-- the bird flew to the top of a tree. And you'll see a light up in yellow. That's the trajectory apparatus. It talks about movement along a path. It talks about locations, places, types of motion.
Some of the conversation that we have as we talk to other people is about physical space and some of it is about the movement along trajectories in very abstract spaces. So here a country is moving toward democracy. That's about as abstract concept as you can get. But it's still, from the perspective of this system a movement of something along a path in some kind of space.
When we have that sort of thing understood, when we know what sentences do, then we can think about how we can build systems that use their imagination to solve problems. A while ago, I gave you the example of John kissing Mary. Let me show you some recent work, work you wouldn't have seen if you were here a month ago, that suggests how a computer might solve a problem with its imagination. Now in order to make a font big enough for you to see, some of the text will be truncated. But I hope it will be audible on the house sound system.
So imagine that a ball falls into a bowl.
COMPUTER: Okay, I will imagine that a ball falls into a bowl.
PATRICK WINSTON: So a synthesizer as imagined the action of falling into a bowl. Now here's an analog to John kissing Mary.
COMPUTER: Okay, I will imagine that the man runs into a woman.
PATRICK WINSTON: Well, we don't have a very elaborate synthesizer, but box will have to do.
COMPUTER: Yes, I have learned from experience that contact between a man and a woman appeared because a man runs into a woman.
PATRICK WINSTON: So because it has perceptual apparatus, it can solve the problem with its perceptual apparatus. Of course, we're not limited to blocks. We can imagine other kinds of things.
COMPUTER: Okay, I will imagine that a man gives a ball to a man.
PATRICK WINSTON: These two slugs are former graduate students of mine, now distinguished faculty. You can even imagine things that can't--
COMPUTER: Okay, I would imagine that a man flies.
PATRICK WINSTON: It doesn't get very far, but that's part of the point. So we think with our imagination. We think with our eyes. We think with our hands. We think with our mouth. We think with our imagination. And we think with stories. I think that much of education from childhood on is about stories.
In the beginning, we tell our children fairy tales so they won't run into the woods. We scare the hell out of them. Later on, we teach literature and history. History because it's supposed to repeat itself. Literature because it's about the human condition. And then we go to professional school, and we learn law, medicine, business, engineering, science, all those subjects are largely taught by the kinds of stories that we tell.
So the question becomes, can we-- using what we've done so far, with our visual system, and our language system, and our all kinds of systems, imagine that it be possible that a program could capture the meaning that comes across in a story. So here's not Macbeth, but just the plot of Macbeth. And there are one or two things I'd like you to especially notice in here. There are no sentences containing the word because. So if the system is to get anything out of the story, it's going to have to get it by applying some background knowledge, general knowledge about what goes on in the world.
How do we know if it knows anything? Well, we can ask it questions. We can ask, for example, how's Duncan doing at the end? And it should know that he's not doing very well, because he's dead. We might say, why did MacDuff kill Macbeth? That's a little harder, but to answer that question would have to know that if you kill somebody's friend, you harm them as well. Then you might elevate it to a higher level still, and say, well, now that I see you can understand and fill in the gaps between the explicit statements in the story, can you recognize concepts like revenge or Pyrrhic victories?
So let me show you what we've been able to do in the past month or so. This is what's in the story. This is not just a syntactical analysis of the story, but this is a report on the knowledge that's been extracted from the sentences and put into the kind of knowledge framework that I showed you when I talked about what we extract from sentences. Now, you can't see that, of course, so I'll blow it up a little bit.
There's one of the elements that has been extracted from the story. With the elements that have been extracted from the story, we fill in the gaps with background knowledge. Once again, that is, of course, invisible. So let me blow a little of that up. Yeah, Duncan's dead. You know that because he got killed. You know that Macbeth harmed MacDuff, not just Duncan. But that's because MacDuff is a friend of Duncan's. And if you kill somebody's friend, they're not real happy about it.
So we have both stuff that's explicit in the story and stuff that's not. It's all connected together by relations that explain and predict. And once we have that in place, then we can begin to ask if there are patterns in the explanation and prediction connections that would answer the kinds of questions that we might have been asked back in our good old humanities classes. So we might ask, for example, is there a success in this story, and a pattern will be found. Blown up, it looks like this.
Yes, there is a success because Macbeth wanted to be King. He murdered Duncan as a means to achieve that end. Duncan died. Because Macbeth is his successor, he becomes King. And so we come back again to the left side, and we see that Macbeth is deliriously happy as a consequence of that. But then, we can see if there's a mistake. And a mistake pattern is found too.
Macbeth murdered Duncan. That leads to Macbeth harming MacDuff. That leads to MacDuff killing Macbeth. And so Macbeth is not so happy in the end after all. One of those connections is only used in explanation. It's the one that says if you dislike somebody, perhaps you'll kill them. Unfortunately, this doesn't have the strength of a prediction. But it's there, only because if there's no other kind of explanation available in the story, that's the explanation that's presumed.
So we can see other sorts of stuff in here too. There's our Pyrrhic victory. It's a combination of a success and a mistake. There's a revenge. And what we have in the end is a claim that we are the stories that we've been told. Some of them we experience directly, but many of them are surrogate experiences, that are taught to us in school through the subjects we take, taught to us in our classrooms, and taught to us through religion. And they're taught to us through fairy tales. And sometimes they're taught to us through propaganda, because that occurs when someone directs us to a story that has the answer that they want.
And as you look at examples like that, you might say, well, that's not exactly my concept of what a Pyrrhic victory is. But that's okay. If you shake hands with the person on the right and shake hands with a person on the left, as the MIT saying goes, they will think differently about what a Pyrrhic victory is too. It's remarkable how we humans tend to assume that when we have a good thought, everyone else will have the same good thought and will think the same way. Symmetrically, we humans often suppose that when we have a defect, it's ours exclusively.
Many years ago, when I was in third grade, there was a dark conversation between my teacher and my mother because I was having trouble learning the multiplication tables. I've never learned the multiplication tables. That's a secret I kept till well into my faculty career. But then, I discovered that my colleague, an extraordinarily intelligent friend, Jerry Sussman, can't remember the multiplication tables either. We both cheated our way around it by remembering all of the squares and would work off the squares. So if you ask either of us what 7 times 8 is, we would say, well, let's see, 7 times 7 is 49, so it must be seven more than that, 56, is that right? Good
To give you one other example, many people, many very bright people, including my wife, have a slight pause between asking them which direction left is and they're identifying which way it actually is. I, myself, do not have that perception. If you say left, I feel like I know immediately that it's over that way. But my wife has, like many people, a noticeable delay between saying left and identifying which way left is.
I'll never forget one day, we were driving along, and she was navigating. And she said, turn left. So I turned left. And she said, what did you do that for?
And I said, I thought you said to turn left. And she sheepishly looked down and discovered that her arms were crossed.
So you might ask me, well, what is it all good for? Well, of course, I think it's good for filling in part of the puzzle of what it is that makes us humans intelligent, the story part especially. What would you do with story understanding if you had systems and understood stories. What if you insisted on a practical application. If you were to hold me in a corner at knifepoint and ask for an example, I would cite an example that I learned about from a military historian some years ago.
He was commenting on how difficult it is to see what's going on in the fog of war. And he said, consider the Tet Offensive, US intelligence could see that the Viet Cong were mobilizing for an attack. US intelligence concluded that if the Viet Cong attacked, they would lose. US intelligence concluded that the Viet Cong knew that if they attacked, they would lose. Therefore, the Viet Cong won't attack. And at that point, they promptly attacked.
Five years later, the Yom Kippur War, Israeli intelligence could see that the Egyptians were mobilizing. Israeli intelligence concluded that if the Egyptians attacked, the Egyptians would lose. Israeli intelligence concluded that the Egyptians knew that. Israeli intelligence concluded that they therefore would not attack, whereupon, of course, they promptly attacked. In both cases, the forgotten element was that there was a political, rather than a military motive for the attack.
But given you had one, why couldn't you see the other in terms of the one. Why couldn't you read the precedent, the history, onto the new situation and understand it from a tutored perspective taught by history. I was telling this story to John Reed just a few days ago. And he said that the recent financial meltdown is a better example, that there were ample precedents that should have informed us that what was coming was coming, just as Israelis should have know that the Egyptians would attack.
So it's possible that we can use this kind of idea, if we focus on applications, in important ways. And not just on using precedents, and history, and stories to understand new situations, but using it to understand cultures. Understand how a play in Peoria, is a phrase I often hear. I happen to be from East Peoria, so I could never figure out why this was such a hard problem. But on the other hand, I would be completely at sea if you asked me how something would play in Baghdad, because I've grown up in a European culture. So I've been tutored with the JudeoChristian Bible. I've been tutored with Homeric myths. I've been tutored by the national heroes. And I've been tutored by modern cinema and entertainment, which is probably not much more than a translation of the Homeric myths.
But these folks over here would have a different set of stories on which they would base their interpretation of events. And if we're to understand other cultures, we're going to have to understand how stories define a culture. Is it 10 stories that matter, or 100, or 1,000? We don't know. But what we'd like to have in the future is a system that would be as indispensable to a political or intelligence analysts as a spreadsheet is to a financial analyst, something that will allow you to turn the dials and say, well, this isn't going to play very well in Baghdad.
So how do we think? We think with our hands, our eyes. We think with our mouth. We think with our imagination. And we think with our culture's stories. And thinking about the creation of intelligent programs from this perspective is a radical departure from what we've done in the past 50 years, as we've tried to understand and build a computational theory of intelligence. For the first 25 or 30 years, we were almost entirely symbolic in the way we thought about things.
We thought if we could only make a program that could do symbolic integration, we'd be on the way. If we can only make a program that can understand a children's story, we would be almost there. Well, we figured out how to make programs do calculus by 1960. And a great deal progress was made on understanding children's stories. But computers remained as dumb as stones. So for the next 20 or 30 years, we built robots, perceptual apparatus, things that move. No language though. So it's almost that we forgot that the distinguishing characteristic of human intelligence is that linguistic veneer that stands above our perceptual apparatus.
So what we have today, as we sit at that round table, is a growing understanding that we humans are intelligent, not because we have a language system, and not because we have a perceptual system, but because we have a language system that interacts with a perceptual system. And to use a metaphor I could only use at MIT, when our language apparatus asks our visual apparatus to answer a question, it's like a Fourier transform. It's putting the question into another space, where the answer can be calculated more readily. So without that ability to communicate with a vision system, without that ability to imagine, we would have systems that could answer questions, do calculus, but never be anything like-- never have anything like the intelligence of a person.
So this brings us to the question of why computers make us stupid. And one answer to that is that they've made it unnecessary to take notes. It's all there, after all. And you can look it up. Everything that happens in the classroom's online. Any question you might want to answer is over there somewhere on the web. And it's not that taking notes has any value in recording information. It has value only in that it forces the engagement of our language apparatus, and our visual apparatus, and even our hands.
So it's that forced engagement that is so valuable in taking notes. And when you don't take notes, that engagement is not forced. The mind wanders. It's like watching a movie and thinking about something else. There's another side of this, of course, and that's the instructional side. The other reason that computers make us stupid is because we, on the teaching side, confuse the presentation of information with the delivery of information.
I'm sure you've all seen slides that are just so packed with words it would take you 20 minutes to read one. I was-- one of my students was critiquing one of her fellow student's presentation toward the end of class this past semester. And she said, well, interesting talk. I've taught them always to say that first.
But, you know, your slides had so many words, and you talked so fast, you jammed my language processor. It's like that Spelke experiment, where you reduce the professor to the rat, when you cram so much information into the language channel, there's nothing left to understand the content of what you're trying to convey. So this is the second of the big reasons why we computer scientists owe you all an apology. It's all our fault. We made it possible to cram a lot of information on a slide. And if that's not enough, you just use a smaller font and get even more on there.
So being more optimistic, what do we do if we want to be smarter? Well, now you know the answer. You take notes, because that forces the engagement of your linguistic, and motor, and visual apparatus, and you can look at the picture after you've drawn it. So you draw pictures too. You talk and imagine. And if there's no one around to talk to, you can talk to yourself.
I do I advise some caution in this dimension, however.
Because people who don't understand exactly what you're doing may reach an incorrect conclusion. A neighbor of mine for a time was a prominent Boston area psychiatrist, who I met once at a community party. And not knowing he was a psychiatrist at the time, I attempted to strike up a conversation. I said, what do you do? And he said, I'm a psychiatrist. Silence. So groping for something to say, I asked him what kind of patients he had. And he said I only take patients that other psychiatrists have given up on. Oh, I said, where do they come from? Mostly MIT and Harvard.
And, of course, he went on to explain that there's a narrow line between, or perhaps an overlap, between seeing the world in a creative way, seeing what ought to be there but isn't, and seeing things in the world that aren't actually there. Talking to Eleanor Roosevelt and believing that she's that would be an example of that sort of thing, I suppose.
So finally, tell stories. Two of my undergraduate students conducted an experiment in the past month in which 100 of their friends were divided into two groups. 50 were told to read some information about computer networks. Another 50 were instructed to do the same thing, but then explain it to the experimenter. They were tested a week later. Guess who did better. At a very significant statistical level, the ones who had explained it, who had told the story right after reading it, did substantially better than those who just read it passively.
So this is the final slide I wanted to show today. And I'd like to conclude just by answering one of the questions that I'm sure is very much on your mind. You're probably saying to yourself, well, I do feel a lot smarter now, having seen this presentation. And in consequence of that, it's possible that downstream, I might make a better business decision, or perhaps invent something in consequence of using one of these techniques, heuristics. But what if I should wake up in the middle of the night agonizing over how I can express my appreciation?
Well, remember, this is the age of the web, and websites are standing by.
MEAD WYMAN: Here's where we find out whether, in fact, we did get smarter. And, of course, as usual, we're running quite late on time. So I'm going to try to confine this to a couple. There were several questions dealing with abstracts-- abstractions. I'm going to read these-- and whether you could get computers to handle these kinds of things. Can computers understand giving and receiving abstractions, such as Paul gave Cindy the flu. Would the computer think that Paul touched Cindy?
In a related question, a little bit more direct, do you think it's possible to engineer an artificial consciousness? It may or may not be the same, but you'll be able to deal with it.
PATRICK WINSTON: Well, I find myself thinking of techniques that we learned in freshman humanities classes. I won't exactly answer that question, I'll answer another question and hope for partial credit.
And, of course, another technique that politicians use when you ask questions with multiple parts, I'll answer one part and leave the other one alone. But when we use words like give and take, of course, there's a context surrounding them. And one of the challenges in understanding language, making computers understand languages, is the challenge of understanding how to confine the scope of a word use, to those situations in which it actually applies. And it's a long, technical discussion, a one-hour lecture in my artificial intelligence course, to discuss mechanisms that we are now inventing that restrict the application of an interpretation to exactly the right part of the world to which it would apply.
So for example, if we say what's fury? And we say, well, dogs are furry and rugs are furry. Cars are not furry. We've now built, and discovered, and defined, and refined mechanisms that shrink our knowledge of furriness around exactly the kinds of concepts that should be considered furry, without letting it leak off and cause us to misinterpret something else. So bringing that back to the question about give and take, certainly we know that most gives involve the movement of a physical object from one person to another. But if you give someone a cold I still got it that's a different kind of give it's a different interpretation of the word it's syntactically the same But from the point of view of the meaning that it stimulates-- the meaning, interpretation that it stimulates in the other person, it's quite different, and we believe we can make that happen.
There was another question about abstractions. And abstractions are something that we worry about all the time. Because that kind of abstraction is the great differentiator of human intelligence. So it's very much on our agenda.
MEAD WYMAN: The last question, then we really must break. There was quite a bit of interest in your experiments with rats and MIT professors. And if I can find this--
While you're looking for it, I'll tell you about an experiment with chimpanzees that I've been much taken with. It involves a mean human being. That can be me. And a monkey, a chimpanzee, that can be you. And here's how the experiment goes. If I put the banana here, you, as the chimpanzee, will not approach it unless I put a blindfold on my eyes. Then you'll come and steal the banana.
So I mention this only because Rebecca was talking a great deal about those parts of our brain apparatus that are devoted to understanding what the other agent is thinking. It seems that that's even-- that a version of that is even present in the chimpanzee.
MEAD WYMAN: Interesting. Well, the question was--
PATRICK WINSTON: I was hoping we would run out of time.
MEAD WYMAN: It's actually pretty direct. It's not too confusing. There won't be a linguistic problem here, I think. If you paint one wall with an odor, does the rat function as well as the MIT professor?
PATRICK WINSTON: Yeah.
MEAD WYMAN: Obviously, an odor instead of a color.
PATRICK WINSTON: Yeah, as far as I know, it's not known. There are many fascinating questions, or experiments that can be suggested, many of which are still under way. There's a question of whether this works with chimpanzees or not. There's the question of whether it works with humans who don't appear to have a notion of left and right, but rather make judgments about direction on an absolute north, south, east. But many of these things are very controversial and have yet not produced any information that we can make use of.
MEAD WYMAN: Okay, thank you so much.