How can computational language help decode the mysteries of nature and the universe? What is ChatGPT doing and why does it work? How will AI affect education, the arts and society?
Stephen Wolfram is a computer scientist, mathematician, and theoretical physicist. He is the founder and CEO of Wolfram Research, the creator of Mathematica, Wolfram|Alpha, and the Wolfram Language. He received his PhD in theoretical physics at Caltech by the age of 20 and in 1981, became the youngest recipient of a MacArthur Fellowship. Wolfram authored A New Kind of Science and launched the Wolfram Physics Project. He has pioneered computational thinking and has been responsible for many discoveries, inventions and innovations in science, technology and business.
STEPHEN WOLFRAM
In modern times, it's popular to talk about artificial intelligence. The term has been around since the mid-1950s, and it's gone through a variety of different meanings and levels of popularity.
Much of what I've done in building computational language and tools can be seen as related to an earlier version of what people thought artificial intelligence was and would be. However, in modern times, the real point is about letting computers automate more things for us. There's a particular twist on the notion of artificial intelligence that has to do with tasks that are very human-like, such as generating human language. These tasks have only recently become really accessible to computation. As we move forward, we can expect to see more and more things in the world get automated.
Looking across human history, many things are just the same now as they ever were. The human condition and the way people are have been very much the same for a very long time. However, the one thing that's changed in the course of human history is the developments in technology and the increasing automation that we can do. This is something we can expect to continue. Automation shines a brighter light on the question of what you actually want to do. As a human, you determine the direction things should go, and automation can support that progress. However, the AI or computational system has no built-in sense of what it should do. There are an infinite set of possibilities. There's something we kind of learn from my efforts in studying the computational universe: there's a lot out there. Most of it are things that we humans, at least so far, have not cared about, and sort of the story of the development of things tends to be what do humans decide that they care about? In what direction do they want to go? What kind of art do they want to make? What kinds of things do they want to think about? What kinds of objects do they want to produce? There's an infinite collection of possibilities, but it's something that's a matter of human choice, which of these infinite things do we actually choose to pursue?
And, then our AIs, our computational systems can help us to be as effective as possible in pursuing those directions. But I think as there is more automation, there is more kind of emphasis on this question of our choice. What do we actually want to do? In what direction do we want to go? And in a sense, there is in the computational universe of all possibilities, there is sort of infinite creativity. There's all these different possibilities out there. But our kind of challenge is to decide in which direction we want to go and then to let our automated systems pursue those particular directions.
Early Years and Path to Physics
I've been fortunate to be able to sort of alternate between doing basic science and building technological tools, so far about five times in my life. And it's sort of an interesting thing because you build up kind of intellectual ideas in basic science, then you apply them in technology. Then that technology lets you come up with new ideas in basic science. And it's a cycle that can continue. But in terms of my own trajectory, you know, I got interested in physics when I was pretty young, maybe 10, 11, 12 years old, and started studying it intently, discovering this amazing thing that you can just read books and figure things out. It was a time when physics was in a very energetic phase in the mid-1970s, with lots of things that were being discovered in physics.
I had the chance to kind of turn computers in the direction of the computational universe and see what was out there. And what I discovered was a lot of things I certainly didn't expect. In particular, that even when you had a very simple program, very simple setup for the computer, the actual behavior of the program, the actual patterns that you generated, could be very complicated. And I sort of realized that's something that seems to be the secret that nature is using to make all this complexity that we see and so that got me very excited about kind of applying the idea of computation to things like basic science. And somehow in the middle of all of that, knowing more about the essence of computation, let me kind of understand more about how one could build tools based on computation to kind of work with things in the world.
Philosophy and Computational Language
My mother was a philosophy professor in Oxford, and when I was a kid, I would always say, if there's one thing I'll never do when I'm grown up, it's philosophy, because how can one be serious about a field where people are still arguing about the same things that they were arguing about 2,000 years ago, and there's no kind of apparent progress. But actually, the exciting thing has been that both in my kind of work in building computational language, and in my work in understanding the computational foundations of physics, that it turns out that a bunch of those things that people have been arguing about for a couple of thousand years, we can actually say some real things about.
Creating Wolfram Research, Mathematica, Wolfram|Alpha, Wolfram Language
It's a funny thing because I've spent my life sort of building this big tower of science and technology and, every so often, something comes out of that tower that people say, "This is a cool thing, we're really going to be excited about this particular thing." For me, the whole tower is the thing that's really important. And in the future, that's what the tower that I've tried to build is certainly the most significant thing I've been able to do. And it's something that, you know, I've been able to see now over the course of half a century or so, kind of how various ideas I've had and directions I've gone have actually played out.
What is ChatGPT doing and how does it work?
ChatGPT is something that was not of my creation that I didn't really expect was going to happen at any particular time. It was being able to get an AI neural network system to be a fluent producer of human language. And that happened in late 2022 with ChatGPT.
Nobody, including people who worked on it, really sort of expected this to work. It's something that we just didn't know scientifically what it would take to make something that was a fluent producer of human language. I think the big discovery is that this thing that has been sort of a proud achievement of our species, human language, is perhaps not as complicated as we thought it was.
It's something that is more accessible to sort of simpler automation than we expected. And so, people have been asking me, when ChatGPT had come out, we were doing a bunch of things technologically around ChatGPT because kind of what, when ChatGPT is kind of stringing words together to make sentences, what does it do when it has to actually solve a computational problem? That's not what it does itself. It's a thing for stringing words together to make text. And so, how does it solve a computational problem? Well, like humans, the best way for it to do it is to use tools, and the best tool for many kinds of computational problems is tools that we've built. And so very early in kind of the story of ChatGPT and so on, we were figuring out how to have it be able to use the tools that we built, just like humans can use the tools that we built, to solve computational problems, to actually get sort of accurate knowledge about the world and so on.
Exploring Computational Irreducibility
I think one very big example of this phenomenon is the computational irreducibility. This idea that even though you know the rules by which something operates, that doesn't immediately tell you everything about what the system will do. You might have to follow a billion steps in the actual operation of those rules to find out what the system does.
There's no way to jump ahead and just say, "the answer will be such and such." Well, computational irreducibility, in a sense, goes against the hope, at least, of, for example, mathematical science. A lot of the hope of mathematical science is that we'll just work out a formula for how something is going to operate. We don't have to kind of go through the steps and watch it operate. We can just kind of jump to the end and apply the formula. Well, computational irreducibility says that that isn't something you can generally do. It says that there are plenty of things in the world where you have to kind of go through the steps to see what will happen.
In a sense, even though that's kind of a bad thing for science, it says that there's sort of limitations on the extent to which we can use science to predict things. It's sort of a good thing, I think, for leading one's life because it means that as we experience the passage of time, in a sense, that corresponds to the sort of irreducible computation of what we will do.
It's something where that sort of tells one that the passage of time has a meaningful effect. There's something that where you can't just jump to the end and say, "I don't need to live all the years of my life. I can just go and say, and the result will be such and such." No, actually, there's something sort of irreducible about that actual progression of time and the actual living of those years of life, so to speak. So that's kind of one of the enriching aspects of this concept of computational irreducibility. It's a pretty important concept. It's something which I think, for example, in the future of human society, will be something where people right now will think of it as this kind of geeky scientific idea, but in the future, it's going to be a pivotal kind of thing for the understanding of how one should conduct the future of human society.
Reflections on Science and Spirituality
I have grown up in the kind of Western scientific tradition, so to speak. And what's interesting to see is that some of the questions that we get to ask now have sort of grown out of the Western scientific tradition, are things that have also been asked in quite different traditions. So, a thing that I didn't think we would ever have anything scientific to say about is a question like, "Why does the universe exist?" But that's something that now we can make some scientific statements about. And those statements have a lot of resonance, I think, with things that people have come to from very different kinds of ways of thinking about the world.
It's sort of interesting that, when you describe the convergence of science and other kinds of things, when I was a kid, people would talk about sort of at a religious level, they would talk about souls and so on. And one would say, "Well, that just can't be anything scientific." I mean, you know, what does a soul weigh? Anything that exists must have a weight; that sounded reasonable from the point of view of the narrow way of thinking about science at the time. Now that we understand this idea of computation, we understand that there can be a thing that is real and meaningful, but it doesn't have a weight. It is merely an abstract thing, a computational thing. And when we think about souls, that's, I think, the idea that what is going for is this kind of computational representation, this computational engram of what's in a brain, for example. And we now have a much better understanding of what that sort of engram, what that abstract, it has no physical weight or anything like that. It's just an abstract thing that can be rendered in a brain.
It's sort of interesting to me that there are things that people have an intuitive sense of and have for a long, long time had an intuitive sense of that sometimes in science, there's been a tendency to say, "Oh, no, no, no. We have a particular way of thinking about things in science and that doesn't fit with it. So let's lock it out," so to speak. So an example of that, well, for example, animism; you mentioned this question of where are their minds? Is it reasonable to think of the weather as having a mind of its own? Is it reasonable to think of the forest as having a mind, so to speak? Well, in these kind of computational terms, yes, it does become reasonable to think about those things. Now if you say then, one comes to that idea from a place of formalized science, but nevertheless, it relates to sort of intuitions that people have had for a long time about that come from that didn't come from that particular kind of branch formalized thinking.
