At 12, Stephen Wolfram started writing a dictionary about physics, and the California Institute of Technology awarded him a Ph.D. at age 20. The brilliant, enigmatic creator of Mathematica and founder of Wolfram Alpha now runs a massive company remotely, spending his days on the phone and with screen share as he works through problem after problem with his staff. His latest development is the Image Identification project, which will enable your computer to tell you what picture it’s looking at. He talked to Pacific Standard about computing the world’s knowledge, how his kids got him to start traveling, and why he’s not scared of artificial intelligence.
How young were you when you first realized that you were smarter than almost anyone else in the room?
I had the good fortune to go to good schools in England where I had the perception that there were a lot of people who were much smarter than me. After, I went on a search for places where there would be a large collection of people much smarter than me. When I went to college at Oxford, that’s what I thought. When I went to graduate school at CalTech, that’s what I thought. I have to say that I was a little disappointed that at each of these places I thought that everybody would be much smarter than me and that didn’t happen. Eventually, I realized: “Gosh, it’s pretty scary. I may be pretty smart compared to people out there.” After that, I thought I should do something that makes use of being decently smart.
You once said, “Ever since I was a kid, I’d been thinking about systematizing knowledge and somehow making it computable.” Why were you drawn to that goal?
I probably wouldn’t have characterized it precisely that way, but when I look back to what I was doing when I was 12, it was collecting knowledge, organizing it, and trying to set things up so there would be ways to work things out. At that time, it was typewriters and diagrams rather than much with computers. I started using computers about when I was 13, but at the time the computers were primitive enough that there wasn’t a lot that could be done in terms of storing knowledge with them.
I got much more seriously interested in organizing knowledge on a computer probably around the time I was 16 or so. I started using computers to do mathematical calculations for physics problems, and I started to think seriously about how you would store technical knowledge in a way that could be used automatically. I started writing serious code for that kind of thing. I built a ground floor for the next 40 years of my work.
You gave a commencement address to one of your children’s high school classes and said that a reason for developing a computer program that could do computations was because you were bad at them. Is that true?
Definitely. It’s not my skill or interest. I guess I just found it kind of boring. I’m quite a meticulous person, so as I think about it now, I’m not sure why I would make the type of simple mistakes that I made over time. I guess it was because I wasn’t very interested in the calculations. I don’t know what would happen if I tried to do them today. I haven’t done calculations for 30-something years. When I see my kids doing calculations, I tell them I didn’t know humans still did that kind of thing.
You’ve talked about how the philosopher Gottfried Wilhelm von Leibniz was thinking about organizing knowledge in the same way you want to, only he was doing it 300 years earlier. You live in a time where doing so has become more technologically possible. Do you think you came about at the right time?
I’ve been very lucky. I think that different people have different things that they are good at that resonate with their personalities and skills. Understanding computation and its significances happens to be a good match for my particular skills and personality. The time I’ve lived happens to be the time when that stuff has first emerged as possible. That’s a break that was lucky.
What’s a typical day like for you? Are there any typical days?
I actually have a very regimented routine. I’ve been collecting all this personal analytics data on myself for 25 years, so I actually know that I have a deeply habitual life. I was a little disappointed to find that out actually. I get up at about 10:30 in the morning. I have my day absolutely full of meetings with people at my company. I have spent many years of my life trying to build this engine for turning ideas that I have into real things. That’s the company that I’ve built. I’ve been a remote CEO for 22 years, so my day is very virtualized. It’s phone plus screen sharing all day. Particularly in the last decade, my basic approach to things has been that I spend the day figuring stuff out. But I like to think in public as much as possible because I’m trying to have other people and the team understand why we are doing something and maybe learn from the process. My day is booked into half-hour or hour blocks: think about this project, think about the next step in figuring out this thing. I have a great time. On any given day, I’ll have a bunch of ideas and most of them will go through the pipeline to turn into real things. I feel quite useful doing that sort of thing.
Yesterday, we launched a website called ImageIdentify.com, which is a thing for doing the sort of AI-ish thing of identifying possible images and telling you what they are. That had a certain amount of visibility today, and we’ve been trying to start to understand the feedback we’ve gotten and from the various things people have put into the site. It’s pretty funny. It’s like, “When does one identify a person?” versus “When does one identify a superhero?”-type things.
One of the things that is a characteristic of a lot of stuff that I do is that we’re trying to figure stuff out and some judgment call will have to be made somewhere. I guess that I kind of like trying to get as clear as possible in terms of how I think about things so that I’m able to make the best judgment call as quickly as possible. That’s what I spend much of my time doing.
I usually work until six, then I do things with my kids and so on for a couple of hours, then I work again and go on until about two in the morning. I have tried to optimize the work that I do to try to be the stuff that I like to do. This works pretty well for me.
You’re talking to a lot of people during the day. Is it difficult to communicate what you’re getting at because you’re so smart?
I work with lots of smart people. This is the very nice thing that I have been able to do over the last 28 years, building up this company of mine. There are people who maybe didn’t originally think along the same lines as I did but have gradually learned to develop in that direction. For some things I’m interested in doing, like computer language design, it took me fully 10 years to get to the point where I could work with people and be comfortable that they really got what I was saying. Now, they can take an idea and run with it.
I’m always coming up with a lot of ideas that are a little bit further out than most of the rest of the team is coming up with. Maybe it’s insanity or something, but they are used to the “Why don’t we do this?”-type of idea from me.
Communicating is something I put lots of effort into, and I got lucky to be able to find smart people to work with. It would be a very different experience for me if I was embedded in some organization that I hadn’t built myself and it was just, “OK, here are some people you need to work with.” Maybe that wouldn’t resonate nearly as well.
Do you read for pleasure?
I’m really bad at reading books. It’s terrible. I have about 8,000 books that I’ve collected over the years, and my wife told me I had to stop buying books because I’ve overfilled any possible space for them completely. I have rarely, if ever, read a fiction book. I read a lot of stuff, but it’s mostly scanning it for content rather than absorbing it for pleasure. There are a few things that I’ve said to myself I’m reserving for later in life. I have a few of those. Reading is one of those things.
What are some of the others?
One of them for a long time was travel. I traveled when I was quite young, but I haven’t traveled much for 20 years. What broke that was my children got to an age where they said I got these invitations to places all around the world and I would say no or do video conferences. They started to pick the exciting places so they could come along on vacation with me. The last two or three years we’ve traveled to all kinds of exotic places, and I have to say I’ve learned a lot from it.
Image identification is another step toward AI. What’s your interest in AI?
I think the overarching story of civilization is trying to automate as much as possible. AI is just another step in that trajectory toward automation. I find it worthwhile to contribute to that trajectory of technology. For me, what I’ve tried to do in building Wolfram Language and stuff like that is encapsulate as much knowledge as possible in a computable way.
Why do I want to do this? Good question. I’ll make a statement: When you think about AI, you think about what AI makes possible and what the role of humans is in a world where AI has been completely successful. What one realizes is that AI does not in and of itself create goals. The AI can execute, but it has to be told the goal. Where do goals come from? For people, they come from personal history, culture, trajectory of civilization, and things like that. When you ask why someone would be interested in automating as much as possible, I would ask where goals come from in any situation. I can’t answer why I’m interested in AI in a meaningful way. It’s just something that I have always been interested in. I feel like it’s a central issue in the history of civilization. But why that’s what I choose to do rather than choosing to sit silently and do nothing with my life, I don’t think I can answer that in a meaningful way.
Elon Musk thinks AI is more dangerous than nukes. You aren’t worried about the potentially dystopian side of AI?
I think that’s a complicated issue. I think that depends on what you think the future of history should be like. There are questions like, “When so much is automated, what will humans do?” Yes, those are real questions. Americans watch an average of two hours of television per day. Gosh, if there’s really nothing that needs to be done, what do people choose to do? For a person like myself, I have many lifetimes worth of stuff that I’d like to do. Much of what I’m interested in doing, at least in my theory of AI, is a lot of goal-setting-type stuff, which is precisely the kind of thing that by definition is not automatable because it makes no sense to make an AI that invents goals for itself.
The thing to understand about this—and it’s a bit philosophical at some level—is that AI and humans are all doing various types of computations. Nature is also doing computations. When we look at the computations that we do in our brains, we think of them as having purposes, having goals. Nature is doing lots of elaborate computations, probably similar to the ones that go on in our brain, but we don’t know how to assign purposes to those computations. They are just things that happen. That’s what one has in a situation where there is a human connection.
I have sort of been thinking about this for ages, but this happens to be an area that I’m thinking about a lot right now. It’s always terrible when I talk about it before I’m at a point where I fully understand it because it sounds more complicated than it does when I’ve actually understood it all. This is one of those cases where I’m a certain distance from having what I consider to be a really crisp understanding.
In my life, I’ve chosen to work on a lot of different things, and it’s worked out pretty well. I’m always in a situation where I’m trying to climb the hill of the next thing that I’m trying to figure out. What I’ve found is that I like that. It feels fulfilling when one figures the things out. I’ve also discovered that the more one knows, the easier it is to figure the next thing out. There’s an amazing network effect of knowledge in different areas. I am lucky that I have a fairly good memory, and I’m spoiled in that I have all these computers systems that try to enhance my memory by automating things. That means that when I have learnt stuff in the past, I end up remembering it and having a decent chance to apply it to some new thing.
I suppose another feature of my life and times is that I have been building things for a long time. Some people do one thing, then throw that one away, and go on to the next thing. For better or worse, I have tended to be someone who builds a taller and taller tower on things that they’ve done before. I’ve done that from the technology stacks that I have tried to build and from the development of the things that I know about. It’s amusing to talk about all kinds of weird stuff.
Who should I talk to next?
Terry Sejnowski, a professor and head of the Computational Neurobiology Laboratory at the Salk Institute.
What Makes You So Smart? is an ongoing Q&A series.
