Paul Penfield, "God, the Scientist” - God and Computers: Minds, Machines, and Metaphysics (A.I. Lab Lecture Series)

[MUSIC PLAYING]

PROFESSOR: Welcome to all of you. And I'm very glad. Thank you all for coming today to the first lecture of this fall's lecture series-- God and Computers-- Mind, Machines, and Metaphysics. I would like to draw people's attention to coming series. This is the first off of in total 10 different serious. 10 different of cognitive scientists and computer scientists will talk about their way to address existential questions within and outside of their research.

We have printed a little brochure in which you can find the actual program. I heard today that TedTalk has actually not quite the latest version of that series. So this is the last and ultimate series schedule. I also would like to draw people's attention to something you can also find in that brochure, which is a discussion group, a noncredit discussion group Harry Cox and me will held at Harvard Divinity School four times this fall to give people the opportunity to discuss questions which come up here in the series in different talks and where they don't have the opportunity to address them within the talk, within the forum on Wednesday evenings.

And it is now my pleasure to introduce our speaker for today. Professor Paul Penfield is currently head of the Department of Electrical Engineering and Computer Science here at MIT, which is by far the biggest department here at MIT. And Paul has held this position since 1989. But at MIT itself, he has been far longer than that. He got his doctorate of science here in 1960, and entered afterwards in the faculty. And today he is a member of the National Academy of Engineering and was also president of the National Electrical Engineering Department Association.

In addition to numerous other words, he received the Darlington Award and the Centennial Medal of the Institute of Electrical and Electronic Engineering. Paul has widely published technical issues, but also increasingly about questions on pedagogy and problems in computer science and electrical engineering education. In his time as head, the electrical engineering and computer science department introduced a special bachelor program in which students after five years receive a master of engineering degree.

Paul has impressed me since I came here with his humor and also with his genuine interest for his students. It is apparent that he cares greatly about graduating students who are unsurpassed engineers and computer scientists and yet who have also a strong sense of responsibility as individuals and in their role in society. It is therefore my great pleasure to have you tonight here and to share your thoughts with us on God the scientist. Paul.

PENFIELD: Well, thank you, Ann. I was wondering-- actually I don't know that we need the houselights down all that much. Thank you. Why? Let there be light. Isn't that a Biblical response? I didn't know whether Ann was talking about me or someone else. But I am honored to be able to be invited-- thank you, Ann-- to give us first lecture in the series on God and computers. It is being held in conjunction with an MIT EESC course which Ann is offering to students this fall.

What I hope to do is to say a few things about the effectiveness with which God is invoked as a way of explaining things. That's my title, God, the scientist. But I have to start with a disclaimer. I'm not educated in philosophy or religion, as I hope you could tell from Ann's introduction. I really know nothing about God. I have-- somebody thinks that's funny. I have no particular religious training, and I regard myself more as a scientist or an engineer than I do as a devout person.

Nevertheless, I am active in the mainline Protestant Christian church in the town where I live. And I come from a family in which my grandfather and uncle were both Presbyterian ministers. And one of my cousins today is a UCC minister. Their love of God and things religious, though, never did rub off on me as much as it did on other members of my family.

But one thing that did rub off on me was my family's love of good times and singing. Picture me, if you can, at the age of 10 with my family sitting around the campfire after dinner. Somebody breaks out a guitar, and we all sing along. One old favorite that was sure to be sung with much amateur harmonizing is called "Tell Me Why. "

RECORDING: (SINGING) Tell me why the stars do shine. Tell me why the ivy twines. Tell me why the sky's so blue. And I will tell you just why I love you.

PENFIELD: This is a song that many of you perhaps have song. It's got a bunch of good questions. Why the stars shine? Why is the sky blue? One answer to these questions is contained in the second verse to the song.

RECORDING: (SINGING) Because God made the stars to shine.

PENFIELD: See, I got God in here somewhere.

RECORDING: (SINGING) Because God made the ivy twice. Because God made the sky y so blue. Because God made you. That's why I love you.

PENFIELD: This is what might be considered the romantic answer to those questions. It appealed to my teenage sisters at the time. Perhaps it still does to our grandchildren. I'm not sure. However, we do small boys in the family sometimes came back with an alternate verse, which might be considered the smart aleck answer to these same questions.

RECORDING: (SINGING) Nuclear fusion makes stars to shine. Adhesive tendrils make ivy shine. Rayleigh diffraction makes skies so blue. Chemical hormones, that's why I love you.

PENFIELD: Whatever you may think of the way that this verse destroys the peaceful campfire mood, at least it does offer answers to these questions and answers that are sort of based in science rather than on God. Why do the stars shine? Is it because of nuclear fusion as astrophysicist and little boys would tell you? Or is it because God made them to shine? Or might both explanations be valid and perhaps useful in different circumstances?

This is a caricature of the kind of question that I want to try to address today. It's a small example, perhaps a trivial one, of our tendency to invoke God to explain what we humans may not understand. We're appealing, therefore, to God the scientist, and this is how I got the title for this talk.

So the question is, is God a good scientist. How would the answer-- because God made the IV twine-- how would that stack up as a scientific explanation? Well, let me explain that, in science, explanations are provided by theories. I'll have to say a few things about how scientists go about distinguishing good theories from bad theories partly because many non-scientists-- and I gather there are some non-scientists in the audience here-- may not appreciate what a scientific theory really is.

For example, some people if you know a scientist you might ask him or her, do you believe Einstein's theory of relativity? Do you believe Einstein's theory of relativity? Or another question, are Newton's laws true? Are Newton's laws true? The word true and believe are what I'm getting at. To a scientist, these questions are really not appropriate. They're not appropriate kinds of questions. And I'd like to explain that as a way of going on.

To a scientist, a scientific theory is really nothing more than working hypothesis, one that's consistent with what the available evidence. A scientist may use a theory, compare its predictions with observations. But the concept of believing a theory, the way a religious person may believe in God, is not necessary. And if it's not necessary, it's not done in science.

A scientist or engineer might have an opinion about how useful or how accurate a theory is, but a theory after all, is just a tool. If it works in your circumstances, fine. If not, got a better tool. A hammer is a tool. You don't believe in a hammer. You use it when you think it does the job.

Now another common misconception is that a scientific theory is either true or false. Remember the questions-- are Newton's laws true or are they valid? The misconception is that the scientific theory is either true or false and that a single experiment can invalidate that theory. However, a scientist would not go along with that. A scientist would only say whether a theory is accurate enough for the purposes at hand. It's not a matter of being right or wrong.

For example, Newton's laws of motion, as usually formulated, fail when the motion is at very high speed because of effects of special relativity. This does not mean that Newton's laws are wrong. What it means is that they are approximate. They have their own domain over which they are very effective and can be used. All theories and laws in science are approximate. The important question to a scientist is not whether the law is valid or not, but whether it's good enough for the purpose at hand.

Now don't get me wrong. I'm not saying that all scientific theories are good, and I'm not saying that some of them are wrong. I'm saying that some are good, but some are bad. Scientists and engineers and others who use science in their work do judge theories and render judgments about how good this theory is, and the scientific community is going to reject bad theories by simply not using them.

And there's another point that has to be made for the benefit of the non-scientist in the audience. You may not appreciate that there can be and usually is more than one theory for a scientist or engineer to choose from. There's a sort of a free enterprise competition among theories. Those that are used are those that are deemed most useful. For example, in mechanics, a scientist would have at least three different theories which could be used to predict the motion of electrons or other objects.

There's is Netwon's second law, which is usually used in cases where speed is not approaching the speed of light. It's very accurate in everyday use, and it's very simple. Einstein's special theory of relativity is useful when the coordinates systems are in what's called the inertial frame, and Einstein's general theory of relativity is useful when there are accelerated frames and gravity present. Each of these is successively more complex and yet successively covers a wider domain. So somebody who wants to do something in mechanics would have at least these three theories to work with.

Now think for a moment about the campfire song that we just heard. There seem to be two proposed theories about why this sky is blue-- one of them, Rayleigh diffraction, the other, because God made it blue. This is not unlike the case of mechanics. A scientist is faced with the need to explain the color of sky, can choose from those two theories just as another scientist calculating the motion of an electron might choose from among these three theories.

So how does a scientist tell a good theory from a bad one? And I'm going to tell you what I believe are the three most important metrics by which scientists judge theories to see whether they are any good or not. A good scientific theory is one that is accurate, simple, and suggestive. And I'll explain what I mean by each of these attributes.

Accurate-- a good theory should allow any scientist to predict the results of experiments not yet performed or observations not yet made with a reasonable degree of accuracy over a reasonable range of conditions. Reasonable domain-- the larger its range of the validity and the more accurate the predictions, the better. For example, since Einstein's theory of general relativity can describe systems undergoing acceleration, it is more accurate than the special theory of relativity.

One aspect of accuracy that we have to point out is the ability of a good theory to work from first principles without the need of ad hoc assumptions or the need to estimate missing parameters. This feature of completeness, while it's desirable, is really not all that common. Many scientific theories involve parameters, such as the gravitational constant, which have to be measured experimentally, cannot be predicted from first principles or the theory which does not involve such missing parameters is going to be deemed better by scientists.

Simple-- obviously, a good theory should be based on concepts which are itself simpler than the phenomenon that you're talking about. It should be easy to use, easy to teach, easy to explain to others. It should not require extensive computation. By this measure, the theory of general relativity-- this bottom one-- gets low marks. Theory of general relativity gets low marks because it uses curvilinear coordinates systems, which are very difficult to work with.

Suggestive-- this is something that scientists don't always think about. But a good theory should lead a perceptive scientist to think of interesting new experiments or extensions to the theory, connections to other theories, explanations of other phenomenon not originally considered. For example, the general theory of relativity, despite its complexity-- the general theory of relativity predicted the existence of gravitational waves. And now scientists are busy trying to find them. They haven't found them yet. And the special theory of relativity suggested the equivalence of mass and energy through one of it's one of the branches from it. And that of course was discovered in the forms the foundation for atomic energy.

So these are the three attributes that a good scientific theory has. It's accurate, it's simple, and it's suggestive. So what this as a background, let's go back to the campfire song and ask how God the scientist is doing. The theory, because God made the sky so blue-- this is a scientific theory. It purports to explain why the sky is blue-- because God made the skies blue. Now that gets low marks on accuracy. Think about it. Sometimes the sky is blue, but sometimes it's not. This theory does not account for clouds. It does not account for observations made in outer space. On the other hand, Rayleigh diffraction-- which, I should explain, Rayleigh diffraction is the tendency of blue light to bend as it comes through the atmosphere more than red light does. All light then somewhat coming through the atmosphere, but blue light bends more and that tendency is what Rayleigh diffraction predicts.

That provides a more accurate explanation. For example, it doesn't fall into the trap of predicting that the sky is going to be blue at night. And the explanation, because God made the sky so blue, you go out at night time-- what do you expect? Do you expect a blue sky? Well, so the God the scientist is not doing too well on accuracy.

Now what about simplicity? It's true that the statement, because God made the sky so blue, is itself very simple. There was no question about that. But the theory is not itself based on simple concepts. Let me point out. It seems that, in doing so, God had to make certain choices. For example, why didn't God make the sky green? Well, Rayleigh diffraction would answer that question. But because God made the sky so blue makes you wondering, why not green.

Either the color of the sky is an arbitrary choice made by God or it's related to some property of God which hasn't been explained yet or it's an unpredicted parameter of the theory, a missing parameter. Anyway, God the scientist is in trouble with this explanation. If God has some property that sets the sky color, it must be one of thousands of such properties. And what are we to make of any concept which has thousands of seemingly ad hoc or arbitrary choices connected with it? Any model of God with so many ad hoc properties must be extremely complex.

Alternatively, if the sky color is an unpredicted parameter, then the theory really fails to explain anything. Because if you ask what color is the sky and say, it's an predicted, you have to measure it, it doesn't do very well. So, there's a problem there.

Now, how about suggestiveness? Rayleigh diffraction really is a good explanation of why this guy is blue. It not only explains why this guy is blue, but-- think about it-- it explains why a sunset is red-- for the very same reason because, as the light comes in, the blue sky gets bent more and the red sky goes and continues through the place where people are observing sunrises and sunsets.

On the other hand, invoking God in this context is not really explaining anything at all. We have to come down and say that. It's just a way of closing off discussion or saying that the explanation is really so arcane that it's not worthwhile to pursue it. And this, I submit, is the opposite of opening up new questions. So it seems that God the scientist isn't doing too well in explaining why the sky is blue. Of course, in God's defense, it has to be said that the campfire song really has nothing to do about science. The more romantic mood is, in fact, enhanced rather well by using God to avoid any scientific explanations.

And besides, aren't we stretching a point too far? Surely nobody would ever try to use God to explain why the stars do shine, or would they? The campfire song asks relatively simple, perhaps trivial questions. But there are other deeper questions that science has not yet solved, and God is often called upon to explain those. Let me mention three of these.

First one is, the creation-- how did the world start? Second one-- what is life, and what is cognition? Cognition is very germane to the topic of this whole series because, in a sense, understanding cognition is the holy grail of artificial intelligence. So let me mention these three. Although physicists have some pretty good theories about the Big Bang, they're still only speculation about what might have been there before. On the other hand, there are many religious myths about the creation. In my own Judeo-Christian tradition, we learn in Genesis that God created of the world in six days. It was explained to me when I was in high school that that was because God was a good Jew. And Jews observe six days and then have the Sabbath. And the theory was that God was created in man's image rather than the other way around. Well, I don't want to necessarily take sides on that.

Christian fundamentalists, on the other hand, accept literally, as a matter of faith, this biblical explanation. Many of them will even tell you why that happened-- sometime within the past 10,000 years, 5,000 some number of years. And they might also quote Genesis 1:16-- Notice the last line. What were we saying? About nobody seriously using God to explain why the stars do shine? In fact, Christian fundamentalists would say that that's exactly what they do.

Now let me go on. Life is the second of those deeper questions. Biologists are now beginning to understand the physical and chemical processes that go on in biological systems. Gradually, the theory of evolution is being flushed out with more details, experimental evidence is accumulating. One of the most interesting of these happened last May. There was a report of some very rapid adaptation of a species of lizard on the Caribbean island. It seems that, within 15 years, after introduction to an island that had smaller branches for these lizards to perch on, within 15 years, the lizards that were growing up in those islands-- not, this was just only a few generations-- those lizards were found to have slightly shorter legs.

This effect had been predicted before the experiment was done. These lizards were actually transported to the new island, a series of 11 or 12 islands actually. And a prediction was made, and the correlation between the size of the branches available for the lizards to perch on and the length of the legs was quite striking. And this was in only 15 years. The effect had been predicted. It's not clear whether that adaptation was genetic or not. And this is one of the open questions about that particular experiment.

Now fundamentalists will scoff at the theory of evolution saying that it's only a theory. In fact, I've personally heard that very statement. It's only a theory. Well, of course it is. Yes, most scientists would agree it is a theory. But then they would point out that it's a better theory than creationism when judged in terms of its accuracy, simplicity, and suggestiveness. And these are the criteria that scientists cherish.

Now let me go to the third of those deep questions-- cognition. This question seems to be a very difficult one for science. The object of the study can be phrased in many different ways. One of them is to imitate nature by building a machine that seems human. Another way of doing it is to understand how and why the mind differs from the brain. Another way of explaining it is to say that science would like to understand what the soul is.

So far the best efforts of biologists, neurophysiologists, computer scientists, philosophers, and others have not really lead to any success. Computers, for sure, have been able to beat humans at some games such as tic-tac-toe or even chess for that matter. But to my knowledge, the Turing test has never been met.

For those of you who don't know what the Turing test is, this is a test of machine intelligence. It was defined by the late Alan Turing, who was one of the legendary computer science pioneers. In this test, a person sits and ask questions and tries to tell whether the answers that come back come from another person or come from a computer. A computer that is good enough to be able to trick people into thinking it's human will pass the test. And so far, that has never happened. Maybe it will someday, but it hasn't yet. So cognition is a case where science really has made, what I would say, is very little progress on the heart of the problem.

So here we have some examples where science has something to offer and examples where science does not. And among the questions for which the tools of science don't seem to work-- and remember, that's the important thing-- do they work or not? Not, are they true, but do they work. Among the questions for which the tools of science don't work, I think it is useful to identify three different categories. The first category is, doesn't work now but it's only a matter of time and everybody believes that sooner or later science will crack that problem. There are plenty of examples like that.

The second category I would have is a category of scientific theory in which science will never work, and it can be proven that it will never work for demonstrable reasons. And the third category, of course, goes in between where reasonable people can differ from each other as to whether or not science will ever work. And I'll have examples of the last two.

First how about an example in which science will never work? And my friends in the audience who are concerned with computer complexity will jump for joy at this one. Consider a gas such as air. There is already a fairly good theory about the behavior of a gas and the scale of a few millimeters up. This is a so-called macroscopic scale. Concepts such as pressure, velocity, temperature, density, and so on are used in most theories. We know how to design airplane wings, and we know that we can predict with good accuracy how gases flow around various kinds of obstacles. The macroscopic theory works. To be sure, it has a bunch of missing parameters. But once those are measured, it's a reasonably good theory.

But there's also a microscopic picture. And think of little gas molecules bouncing against each other. They're little particles bouncing around, hitting each other. One theory which would be useful and which will never happen is to predict, through simulation of the microscopic molecules, what the macroscopic motion is and thereby to deduce the properties of gases.

And think of it-- you have a little container that has maybe a lot of gas molecules in it. If you know each gas molecule's position, you know each gas molecules direction, the velocity, the property of each one. You ought to be able to calculate what the configuration is anytime time in the future. And therefore you could predict the results of experiments that were conceived of at the macroscopic scale. There is a finite number of molecules. And so why don't we just go ahead and calculate it.

Simulation, which is what this is, is a very effective scientific and engineering tool used all the time, and it would be extremely useful if it would work in this case. Let me explain several reasons why it won't. First, think of the number of molecules in question. This number is approximately what's known as Avogadro's number, which you may remember from your high school chemistry. That's the number of molecules in a mole of any chemical substance, and it's really a pretty big number. It's about 6 times 10 raised to the 23rd power. This is really an enormous number much larger than any number of which is encountered in everyday life.

And I tried to think of a way in which the size of that number could be explained to the audience, and try this. Consider the best laser printer which you can buy for your personal computer. Laser printer puts little dots on a sheet. And if the dots are small enough, it looks like continuous writing. And these are very successful products, and the best laser printers you can buy have a resolution of about 600 dots per inch. That is, in any inch, you get 600 and then square inch you get 600 times 600 such dots. On a single piece of paper, there could be 30 million separate dots. Now 30 million is a pretty large number, but it's a lot less than Avogadro's number.

Now think of the whole surface of the Earth, all the land and all the oceans all covered with dots of this size. And the number of dots that you could get, just 600 dots per inch, would be about Avogadro's number covering the whole surface of the Earth. And that's such a large number that it's impossible to really conceive of in everyday terms. Now consider the practical problem of calculating with this many particles. Suppose each air molecule is characterized by its position, its velocity, its orientation, and its angular velocity.

Then, any computer that simply looks at the data for all the molecules without even doing any calculations but just looks at the data, it would take a long time to do that. In fact, if you did that on a very fast modern computer at the rate of 50 million of these per second, it would take about five billion years. Now for those of you who don't recognize large numbers, 5 billion years is the age of the Earth. So if you wanted just to read all the data about this system, it would take the estimated age of the Earth just to do that without even doing any calculations.

The conclusion from all this is that a detailed microscopic simulation is impractical because of the large number of molecules that are involved. Remember the criterion that a good scientific theory has to be simple and easy to apply? This one really fails big time. The most that can be done from a microscopic simulation is to predict the general nature of a few microscopic variables, and such simulation doesn't even do that very well. We cannot predict any detailed motion.

Now it's interesting to note that there are other reasons why this approach will never work either. One of them is the huge size of the computer memory needed to store the data. Another is the possibility of measuring simultaneously the position and velocity of small particles, which comes from the so-called Heisenberg's uncertainty principle. But we ought to be able to get around that by using quantum mechanics instead of Newton's laws. And another one has to do with the extreme sensitivity of the calculations to very small changes in the assumed initial conditions, initial velocities and positions.

Now this is what's known as a chaotic system. The term chaotic maybe gives the wrong connotation. But systems in science where there's an extraordinarily, what's called, exponential dependence on the initial conditions is known as chaotic. And this is one of them. This means that it's almost impossible, for practical reasons, to do microscopic simulation.

So what does this example imply about science? What it says is that different areas of science may be separated by a gap that cannot be breached even using the fastest computers imaginable. Different approaches, different variables, different styles, different theories are needed. Fluid dynamics, the macroscopic, is really different from any molecular theory that you can imagine. It uses a different approach, different styles. In a similar but less extreme way, chemistry is different from physics, even though chemical systems surely obey all the basic laws of physics. And biology is different from chemistry and different from biology, even though biological systems have to obey all the laws of chemistry and physics along the way. And I will argue, in a minute, the mind is different from a brain in a similar way.

So let me come to the third open question, which is cognition. And I'm going to assert that cognition-- I just gave an example of this category. I'm going to assert that cognition-- there's a difference of opinion as to whether cognition will ever be cracked by science or not. One way of understanding the thinking process might be to perform simulation based on a model of the brain. Now a model of the brain doesn't have to use 10 to the 23rd elements. What does it need to do? Well, the brain has about a trillion neurons-- that's 10 to 12th the last time I checked. Each of them is connected to 1,000 other neurons on average.

So a simulation using a neural model requires consideration of a trillion neurons and a trillion times 1,000 other connections. And this is getting to be a pretty large number even for modern computers. Simulation using the neural model is not going to work because of the large number of neurons, the complexity of the models for each neuron, the number of connections among the neurons, the sensitivity of the calculated results to the details of the initial conditions, and of course the speed of even the fastest computers that are likely to be available. The most we can hope for from neural simulation is the identification of a few averages, like in fluid flow, like the pressure and density.

To explain cognition, though, it'll be necessary to deal with many more cognitive quantities than you deal with in fluid mechanics. For example, you'd have to predict at least as many results as a typical person would have thoughts or feelings. And how many thoughts or feelings does an individual have? You'd have to make at least that many predictions from any simulated theory. So I assert that neural simulation will not work.

What about other approaches? Well, the mission of computer science, at the end of the day, when all is said and done, is to manage complexity. All the things that computer scientists do in one way or another are limited by complexity of the artifacts that they deal with. If cognition is also complex, perhaps the tools of computer science can help. Now I may be wrong, but I tend to be pessimistic about this approach. So far, there's really very little evidence that the architecture of a brain is similar in any deep way to the architecture of a computer. Without some degree of fundamental similarity, I assert there is no reason to suppose that the underlying concepts would be the same and that the tools of computer science would work.

In my opinion, a more likely outcome of study of this and a very useful one would be to create new computer architectures that mimic the brain. That's much more likely and would be very useful. Now there is a field that's received a lot of attention recently-- so I should say something about it-- that goes under the name neural networks. The structure of a neural network is motivated by the assumed structure of the brain with its interconnected neurons. There is, at this point, to my understanding, no general theory of how such systems work from a theoretical point of view. But they have been made to do useful engineering tasks.

But as far as I know, there hasn't been any helpful insights from these activities that have been applied back to help formulate a theory of cognition. I would be happy to be informed otherwise. I'm not aware of any. So we've seen some approaches which don't seem to offer much promise. Are there other approaches that will work? I assert that there is no clear evidence one way or another. This is a debatable case.

A lot of people have put a lot of effort into working on this without success. And I'll tell you what my opinion is here, and this will not be the opinion that's shared by everyone in this room. But my opinion is that there will never be a good scientific theory that explains cognition-- a theory that, like other good scientific theories, is accurate, simple, and suggestive. Remember, a theory that stands up to judgment and criticism of scientists.

Now, I'm not trying to denigrate research in this field. On the contrary, some progress has been and will be made in this area. There will be lots of results obtained along the way, which will be very useful, for example, in guiding the design of made systems, such as computers, robots, and so on. But in my opinion-- and this is debatable, and therefore other people can have other opinions-- the lofty goal of explaining cognition scientifically will not be achieved.

Now, if we accept that assertion-- or even if we don't accept that assertion for that matter, and if we only say the cognition is not yet understood. The people who believe that it will someday be understood are in this category. If cognition is not currently understood-- and I say, it never will be understood by science-- some things are therefore excluded from the realm of science, and those are things that are naturally related to self-awareness, things that are felt internally, personally. And I'm going to lump these together under the name human nature, although that isn't quite the right concept. I think it's pretty close. These include thoughts, aesthetics, desires, feelings, emotions-- like love, grief, jealousy, hatred-- intellectual capabilities-- like instinct, judgment, intuition, intelligence. These will not be adequately explained by any scientific theory.

So what alternative do we have? If we need to reason about one of these topics-- or for that matter about any of a number of other topics that are not now understood by science-- if we need to reason about them, it will have to be done by nonscientific means. One such approach is to make assumptions that are not justified by science and then to proceed. These assumptions are sometimes called leaps of faith. And science in general tries to avoid such assumptions, whereas most religions and many other human activities seem to require them. But of course, faith is an intensely personal matter. What one person assumes will generally differ from what another assumes. And science, by its very nature, produces results that are supposed to be able to be reproduced by others. So the leaps of faith are inherently nonscientific.

I want to call this approach faith-based reasoning. Whether or not the concept of God or religious concepts are involved, assumptions that are not justified scientifically, I want to call faith-based reasoning. Faith-based reasoning inevitably lacks the scientific criteria of accuracy simplicity and suggestiveness. It's tempting to say that it's therefore bad science. Well, maybe it is. But maybe it's the best we have. And maybe if we keep in mind the underlying assumptions, keep in mind the leaps of faith that are made at the beginning of the analysis, maybe we can still have some reasonably useful theories.

Lots of people we all admire-- artists, architects, doctors, managers, musicians, social scientists, and engineers for that matter-- rely in part on science and in part on understanding of human nature and social systems. And if, as I assert, human nature cannot be understood by science and if we get too smug about discounting anything nonscientific, we may be left with no tools at all to work with.

Another point we scientists and engineers should not forget that science as an endeavor is itself based on a set of assumptions. We use mathematics, logic, deductive reasoning, and a host of other tools that, in the final analysis, are not themselves provable by any rational means. We all assume that the behavior of nature is not changing with time. How can we be sure? When you really get down to it, we use all these assumptions underlying science for what is really a very simple pragmatic reason, and that is that they work.

Why shouldn't we be prepared to be equally pragmatic in considering whether to use faith based reasoning? The important question is, does it work. Now, there is another reason we should not look down upon nonscientific reasoning too much. Science itself is a human activity. We scientists are ourselves humans after all. We have our own personal thoughts, hopes, dreams, feelings, and emotions. And I assert that those cannot be explained by science. From time to time, we all pursue activities that are not based on rational thought I'm sure sometimes when doing science, but at least off the job-- enjoying fine music or beautiful art, even though there's no scientific explanation of aesthetics.

Even on the job, science deals with artifacts and theories that scientists themselves consider to be things of beauty-- so aesthetics, even as applied in the judgment of scientific theories themselves. As for another example, let's consider the emotion of grief. Many of us in this room know what it's like to experience grief resulting from the death of a close friend or relative. Right now the entire MIT community is going through a period of grief over the tragic loss of a freshman, Scott Kruger, to alcohol poisoning a few days ago.

Now this emotion of grief is a very real one. Someday, science may have something to say about the neural patterns associated with grief. Science may even develop a specific mood changing drug that inhibits those patterns in some ways. However, I assert that grief also has a deeper emotional side to it. And if you need to cope with this aspect of grief, you might find helpful support in societal mechanisms, including religious ones. Many people have found that a faith based on a belief in God is effective in providing solace and comfort. Not many people would receive much consolation from knowing that their grief is somehow related to chemical imbalances in the brain or patterns of firing of neurons. In other words, in handling grief, science doesn't help, but a belief in God does. Let's be pragmatic. If it works, that's the important question.

As for another emotion-- love-- I sincerely hope that everyone in this room has or will experience the joy that accompanies total unconstrained love for another human being. It's a wonderful thing. It may be true that sex, the physical side of love, is somehow explained by chemical hormones as a smart aleck verse of the campfire song says. Chemical hormones, that's why I love you. But I assert that it's surely true that the deeper emotional and spiritual aspects and ultimately the more satisfying and pleasurable ones are better described by a romantic answer such as, because God made you, that's why I love you. And I'd like to wish for everyone in this room a chance to say to your loved one something like-- what to tell your loved one, chemical hormones that's why I love you? Or wouldn't you rather say to your loved one, something like, because God made you, that's why I love you. Thank you for your attention.

PROFESSOR: We have 20 minutes time for questions. Thank you, Paul. It was wonderful talk.

PENFIELD: Thank you. Thank you.

PROFESSOR: We have 20 minutes time for questions. So there are any questions, comments, or--

AUDIENCE: You mentioned that cognition and human nature will never been understood.

PENFIELD: Well, that's my opinion. Other people differs. And I was careful to say that that's in that third category.

AUDIENCE: Well, why do you say that? I think that you have an ax to grind because you are a believer and you're putting a lid on it. And it's very unscientific.

PENFIELD: Well, it's true that that's an opinion, and therefore it's not subject to the same kind of scientific analysis that I would like to apply to calculate the trajectory of electrons. That is true. I'm basing it on the following observations-- that a lot of very smart people have worked for a long time and made very little progress. Computer science has this open question called is p equal to non-p, equal to np. There are certain problems in computer science that can in principle be solved by running a computer in a time which is known as polynomial time, that is not too fast. And the other ones for which nobody has yet discovered a procedure or an algorithm that works that way. It's never been proven.

I think most every computer scientist, if you ask them-- is p equal to np-- that is, is it possible, eventually to find a polynomial time solution-- would say no. But it's never been proved. And the reason they would say no is that a lot of very smart people have worked a long time and never made any progress at all really on that. Now some of my colleagues are going to differ with me. But I think that most of the computer scientists you'd approach would express an opinion that p is not equal to np.

Now, I'm expressing an opinion that's based on the same kind of evidence, and I would be happy to listen to other people who can explain why they have a different opinion.

AUDIENCE: Well, I would like to hear some atheists making those assertions. That will have more credibility.

PROFESSOR: If you come back in three weeks, Rodney Brooks will talk about artificial humanity. That is exactly on-- October 27nd. And he's atheist, and he will talk about this very same issue.

PENFIELD: That's right. Ann has done a marvelous job of getting people with different fundamental beliefs here, and I will have the wind knocked out of my sails completely by Rod Brooks who knows much more about the subject that I ever will.

PROFESSOR: I don't know. Do you want to handle the question or--

PENFIELD: OK, question?

AUDIENCE: Isn't the crux of the matter--

PENFIELD: Now, be sure to speak loud enough so that everybody back in can hear.

AUDIENCE: It seems to me that the crux of the matter is you have to first acknowledge there are different ways of knowing.

PENFIELD: I don't know what it means to know.

[LAUGHS]

As a scientist, I will take a tool and use it where it's appropriate. As a carpenter, I will use a hammer or a screwdriver and I know when to use what tool, and I've learned what tools work. What's it mean to know?

AUDIENCE: Paul, I like it all the way up to a certain point. And then you took a leap that I think leaves a lot, which is you've split it into this--

PENFIELD: You mean, you got through the first minutes?

AUDIENCE: This bifurcation of a choice as to whether we have a scientific theory by which-- to you, you mean scientific theory like physics.

PENFIELD: Yeah.

AUDIENCE: Or on the other side, you have things like faith in God. And there's this huge middle that I think is an enormous part of human experience, including all of the social sciences, which have theories. Which although not tested in the same way or coming up to the same level, in some sense or abstract scientific perfection as, say, physics, nevertheless are very useful theories and yet still are not based on faith in, say, God or religious values and yet which still are very useful. And there's a whole range--

PENFIELD: When are you going to start disagreeing with me?

AUDIENCE: So what I'm saying is it that you pose it as this choice between two things that, when you skipped over this whole middle thing, which may be an even more adequate explanation for the various things you described here.

PENFIELD: No, Charles. remember what I did. I said, when you don't have a good scientific theory, you have to make some assumptions. And I call them a leap of faith, and that doesn't necessarily involve religion. It doesn't necessarily involve God. It doesn't necessarily involve-- but it does mean an assumption. But if you can keep a number of assumptions small, you can have a better structure that's built on it. Maybe the social scientists do that reasonably well. There certainly is a big middle ground, and it isn't as though you have on the one extreme the scientists who don't have any assumptions. In fact, I tried to do say that science is itself built upon unprovable assertions and on the other side religion.

I don't think it's like that at all. I fully agree with you that, in the middle, there is a complete gradation of problems.

AUDIENCE: I have kind of a sneaky praise to the pragmatism of religious thinking. And one, I have a hypothesis that people may have a religious instinct from birth.

PENFIELD: A hypothesis that someone may have a religious instinct.

AUDIENCE: It's very useful for an infant to believe that the environment will respond to it when it cries for example and then that the environment watches it. It's kind of true that the environment is the parents. And then it's nice to give it that feeling. But whatever else you have like love and all those wonderful theories you cannot explain, it's nice to you to just believe that are wonderful and that they heal the whole universe instead of going and studying cognitive science. And you come up with all those useful metaphors, and you can, by the way, increase those things, just thinking as the [INAUDIBLE] whatever.

But the thing is, when you start believing that there is some unknowable entity whose structure you don't even postulate, who makes those things work in a way you don't care to explain, then you make some leaps that are just sloppy thinking.