The European: Last month, Alan Turing would have turned 100 – and in recent years, he has been elevated from relative obscurity to computer science fame. What is so interesting and significant about his work?
Wolfram: Turing was almost unknown when I first started hearing about him in the 1970s. After Andrew Hodges published his biography of Turing in 1992, he became both a cause and a famous scientist. In modern times, much has been converging to what Turing thought about decades ago, and I have discovered that I am interested in a sequence of things that are almost parallel to what he was interested in fifty years ago. The biggest thing he did was to figure out the abstract idea of computation. He took that idea a certain distance, but got distracted by other things after a few years and never tried to simulate actual computation. That was basically left to me. I’ve spent much of my own life trying to understand at a fundamental level what computation is, why it is significant, et cetera. One interesting discovery is that computation does not require a lot of complicated engineering, but is naturally occurring. We’ve discovered the simplest universal Turing machine – a model that can simulate any conceivable computational operation on a step-by-step basis – a few years ago. It is simple enough that it occurs in nature!
The European: Another concept that Turing talked about before World War II – and long before technology would evolve in that direction – was the idea of artificial intelligence.
Wolfram: Yes, and we have found little phrases in his writings which suggest that he even thought about something like WolframAlpha, our computational knowledge engine. There are so many instances when we look at computational problems today and realize: Turing was here before us! We were making a poster with the Rieman-Zeta Function a few years ago and plotted the function’s behavior on a long piece of paper. The mathematical problem behind the function has been unsolved for 150 years, but suddenly I realized that Turing had once held the record for having worked out the longest sequence of wiggles. He had built a bizarre mechanical computer with cogs to figure this stuff out. What really interests me is why Turing didn’t really follow the path that his work seemed to foreshadow. He was the kind of person who worked on one project, and then got interested in something else. In some ways, that’s an enviable trait.
The European: I want to go back to the idea of the Turing machine. One of your arguments is that very simple programs can produce very complicated results – programs so simple that they could be encoded in a cell’s genome, for example.
Wolfram: Our intuition tends to be that we have to go through a lot of effort to build something that is complicated. But nature is very complex without going through a lot of effort: Evolution seems very complicated on a large time-scale, but the actual processes are not. They simply work and unfold and lead to new species. That was always a mystery to me, so a few years ago I began to experiment to see whether simple programs could produce very complex patterns of behavior. The question is: Is that how nature does it? I got a lot of evidence that in many cases, that is how nature works.
The European: This seems to contradict the general trend to drive technological innovation by packing more computational power into a single computer chip.
Wolfram: A couple of points. There is a phenomenon which I call computational irreducibility. When you have a process where the behavior is quite simple – like a planet orbiting around a star – we are smart enough to use math to figure out what will happen in the future without having to wait for the planet to move around. We can compute the outcome by plugging the right numbers into a formula. But many systems are irreducible after a number of steps – you really have to simulate each step to see what will happen. We need a lot of computational effort for that. But it’s a fallacy to believe that our current technology is the only possible computational technology. The fact is, we can make computers from a lot of materials, not just transistors. The reason that’s exciting is because it opens up the possibility of making a computer out of molecules. It hasn’t been done yet, and there’s a lot of ambient technology that is required to make a molecular computer possible. But it reminds us that we must not shrink transistors – we can use much simpler components.
The European: Maybe it is helpful to talk a little bit more about what you call “computation in nature.” Our common sense tells us that there’s a big difference between animate and inanimate life, or between human technology and natural organisms.
Wolfram: Computation describes a system that starts somewhere, goes crunch-crunch-crunch, and produces a result. The question is whether all computations are like those that we program into computers with our current software engineering. The answer is no. But when you start enumerating programs at random, a lot of them look remarkably like the kinds of processes we see in nature. Today, we are using active algorithm discovery in our research, where we mine the computational universe for programs that might be useful for doing computer processing; and it’s becoming much more obvious that naturally occurring computations are not unlike the processes inside a computer: We start at one state, and end at another state.
The European: This equation of nature and technology raises a few very hard questions about the definition of life, about free will, and about intelligence. What are the mathematician’s answers to those questions?
Wolfram: Let’s talk about free will first. The problem with human free will is that deterministic stuff is going on underneath, like the chemical processes in our brain, but that we don’t seem to act in a deterministic way. People used to think that deterministic processes must result in deterministic behavior, and that belief has underpinned much of the debate about free will. It’s the reason why the science fiction robots of the 1950s often speak very logically and behave very stupidly. The main scientific discovery is that it must not be like that. We can have simple deterministic underpinnings that result in very complex and seemingly random behavior. Computational irreducibility is a key feature of life: We cannot grasp life through a formula, but must really simulate and observe it to see what happens. That’s how we as humans end up freeing ourselves from the deterministic rules. I tend to think that the concept of computational irreducibility is probably the answer to the philosophical debates of the past two thousand years about the relationship between free will and determinism. Philosophy is always at a certain distance to human behavior, so a lot of questions really get answered by science.
The European: And intelligence?
Wolfram: It’s an interesting issue. The question is how we can define intelligence in the abstract. What are the criteria we use to judge whether something is intelligent or not? That’s roughly analogous to the question of whether something is alive or not: People have come up with different definitions over time. The Greeks used to say that something is alive when it can move itself. Well, we know that all kinds of machines can move themselves, so that’s not a good definition. For every definition that relies on chemical or reproductive properties, we can come up with examples that satisfy the definition but probably wouldn’t be considered “alive” by most people. We don’t usually think about life or intelligence in the abstract.
The European: And we probably base our definitions on our own experiences. In almost all science fiction films, even the most bizarre aliens look rather humanoid.
Wolfram: When the first landers were sent to Mars, the big question was whether they would discover life forms. And from today’s perspective, some of the criteria that were discussed were rather absurd: “Feed it sugar and see whether it does something.” Here on earth, it’s relatively easy to see whether something is alive because everything has a shared history, a shared RNA, et cetera. Our definition of life is based on history and we don’t know what the abstract definition might be. It’s the same with intelligence. You might say, “this object is not intelligent because it only goes ‘click, click, click, click’.” So what is the threshold for complicated and intelligent behavior? This brings us to a second principle, the principle of computational equivalence. It says that as soon as the behavior of a system is not obviously simple, it will probably be as complicated as anything else. That means that many systems are equally complex. If one of these systems counts as intelligent in abstract terms, so do most others! So the only way we might be able to talk about intelligence is to pursue a definition that is based on history and experience.
The European: How do we figure out whether there’s intelligent life in the universe that doesn’t share our terrestrial history?
Wolfram: Well, in the past we assumed that complicated signals from the cosmos must have been produced by civilizations that were about as advanced as we are. What the principle of computational equivalence tells us is that this isn’t true. You can get complicated sequences and patterns from simple rules that don’t require billions of years of evolution. A famous example is when Marconi and Tesla detected weird radio emissions from space and said, “the Martians must be signaling us!” The radio had just been invented, so they assumed that we had reached a technological threshold and were suddenly able to communicate with extraterrestrial life forms. It turns out that the signals came from the magneto-hydrodynamics of the outer atmosphere. When pulsars were discovered much later, scientists were also really excited because the signal was so periodic and seemed too intentional. In both cases, there was a confusion about the underlying cause. For me, the realization that we cannot really talk about abstract intelligence had an important personal consequence: I realized that artificial intelligence doesn’t require us to build a brain-like thing that we can later program, but that we can start with simple computation.
The European: The New York Times Magazine recently published a profile of Craig Venter, who led the team that decoded the human genome. Two things about it struck me as very interesting: One, he argued that the main challenge for innovation is not to do more, but to spread the benefits of innovation around the globe. Two, the best way to do that is through private enterprises and not through academic research. What’s your take on that?
Wolfram: I was an academic for a while, but I really like energetically doing projects. What I tried to do is build a very efficient mechanism to turn ideas into things. Right now, entrepreneurial companies seem to be the best way to do that. I look at my friends in academia and think: “Wow, things moved so slowly there in the last 25 years!” When we hire academics to work on WolframAlpha or Mathematica, the biggest shock for them is always how quick everything moves. We sit down, and an hour later we have decided what we are going to do and moved on. We can do crazy projects! If you want an immediate impact on the world, that’s what you need.
The European: Is that what drives you – to have an impact on the world?
Wolfram: I find it really interesting to take these very complicated projects and make them accessible to everyone. That’s neat, and it gives me a warm fuzzy feeling. Our intention is: If a question can be answered by an expert, we want to automate it so that anyone can access that knowledge from anywhere. When I was little, I had to bicycle to the local university library and look information up in a book. I had to write a computation down by hand, or find someone who could tell me how to do it. Now, it’s automated and instantaneous. One lesson of the past decade is that we tend to do something a lot more when it is automated. When Google started, some people questioned whether we would want to spend our time looking up information in an online catalogue. They said: “There’s a reason that we have few reference librarians in the world. There’s not much demand for them.” It turns out that’s completely wrong. When you automate the work of the reference librarian, everyone starts using it. Gradually, we are making different decisions because we actually have access to much more information. That’s good progress in the world.
The European: What’s the difference between a search engine like Google, and a computational knowledge engine like WolframAlpha?
Wolfram: With a search engine, you type in a keyword and try to find the best matches. It’s like walking into a library and being handed the ten best books about a topic. What we are trying to do with WolframAlpha is to create custom-created reports to answer specific questions. We are computing answers – even if nobody has ever asked that question before, maybe we can work out a report that answers it. It takes human experts to do that, and that is something that the search engine crowd is often skeptical about. They say that something is only good when it is based on a good algorithm and infinitely scalable. But we are interested in encapsulating the world’s knowledge, not in scalability. Wikipedia is basically a container for random texts written by random people at random times. We can surely do better than that, especially if we want to build something that has different layers and relies on good information. The actual data that we have inside of Wolfram Alpha is now roughly comparable to the textual content of the internet, and much of it comes from primary data sources that are not available online. I find it quite interesting that Google’s search division recently changed its name to “knowledge division.” Sergey Brin used to be an intern with us before he co-founded Google. We have had many good discussions, and I like to think that the name change came out of those.
The European: In contrast to Google or Facebook, you are not dealing with personal data at the moment, but the thrust seems to be going in that direction. How might that change WolframAlpha, and what new challenges does it raise?
Wolfram: We know that we can do scarily well at computing things about people. When you look up someone on the web, it’s noisy and messy. But by combining different data sources – particularly in the United States, where many sources are publicly available – we can pretty much nail a precise report about a particular person. For the time being, we have decided not to do it because the bad appears to outweigh the good. I feel somewhat squeamish about it. My guess is that it will eventually happen, but it’s not a trail I want to blaze.
The European: This seems to be a common reaction. It’s technologically possible, and probably very economically profitable, but we often have a gut reaction against that kind of exposure of private data.
Wolfram: We expose lots of information about ourselves on Facebook, but at least we have a degree of choice. When you are able to see who has bought what piece of real estate for what price, that’s data which is published without your consent. I don’t want to fight that particular battle; I don’t want to say that nobody should be allowed to have an air of mystery about them. It’s quite a different story when people upload data and allow us to analyze it. That’s very interesting and we should actively pursue it. If you know your complete email history, for example, it becomes possible to ask questions like “when did I last visit this area?” If you combine that with an augmented reality system, you can get immediate feedback. Maybe there’s a restaurant nearby that you liked or that a friend recommended in the past. I believe that this is a very powerful development, because we can really base our decisions on our own life history and not just on publicly available data.
The European: So there are limits to the intrusiveness of data searches that violate personal privacy. Is there a similar limit where we might say: Even if it were technologically possible to automate most everyday processes, we should not do so for the sake of intuition or creativity.
Wolfram: It’s interesting that you mention creativity. We did an experiment a few years ago where we randomly plucked music from the universe of possible musical arrangements, and it actually sounded quite decent. I have been hearing from composers that they use that website for inspiration, which is the exact opposite of what I had expected. But the question remains what we humans should do if everything became automated. The answer, I think, is that we figure out what we should do. Let’s assume that everything is automated and wonderful. What do you choose to do in that case? As humans and as individuals, we have certain purposes that we are trying to achieve and which cannot be automated. Highly advanced artificial intelligence can be programmed to have a particular purpose but it cannot answer the question of what’s the right purpose to have. I find it highly interesting to figure out how human purposes evolved and how technology might affect them. At different times in history, we have said that our purpose is religion, or maximizing pleasure, or maximizing money. Some of the purposes we have today would seem rather bizarre from a historical perspective. Imagine a paleolithic ancestor trying to figure out why someone would walk on a treadmill indoors! So when lots of things are automated and possible, what purposes will we value? My personal and rather bizarre answer is that future generations will return to the wisdom of the ancients. The times we live in right now mark the first time in human history that data is permanently recorded on a large scale, so future generations can study us and say: “These people lived finite lives and had to make tough choices. So maybe those choices can tell us something about what it means to be human, and about what endpoints our idea of progress should aspire to.”
The European: Is that your purpose, to think about human progress?
Wolfram: I suppose my crazy way of expressing a purpose is to say that I like to build alien artifacts. I like building things that nobody expected to be built, and I’m not really excited by the idea of taking something that already exists and making it slightly better. That’s somewhat egotistical because I can say, ‘this would not have happened without me.’ It often starts with a very broad idea or project and then leads me to drill down to the essential point, to the golden nugget that might be at the core of an idea. That’s what I like.