This week’s BEACON Researchers at Work blog post is by MSU postdoc Arend Hintze.
When I am asked what I do, I normally smile apologetically and say something like “Theoretical Biology” or “Computational Biology,” and with a wink of my eye “like biological evolution … only in the computer” followed by a hand waving gesture that looks like me typing. At least that is my way of coping with the problem of explaining what it is that gets me excited, and most people slowly nod their head and respond with an encouraging voice: “In the computer, I see!” and in most cases we change the topic.
But the question remains: What do I do? And even though this question is absolutely clear to me, and I never ask this myself, it remains hard to communicate. So let’s give it a shot.
I am fascinated by artificial intelligence or consciousness and the very idea that we could build a computer that thinks! So much that I imagine myself having a dialog with the computer, chatting about the meaning of life, and making a lame joke about how it comes that the very idea of consciousness is apparently only an illusion created by a very complex machine.
When I was a kid I had one of the first home computers, and immediately tried to code an AI. Over many years I made many different attempts at that, only to learn at last that programming such a computer brain is impossible, and will most likely remain impossible to do for any human. And at the same time I see intelligent beings around me, and I know that they came about without help, and without a master plan, only by evolution. Kind of disappointed about computer science, I started to study Biology, apparently the only science dealing with evolution, the one driving force that ever achieved what I was looking for. I got my PhD in genetics and developmental biology studying the complex system by which a nematode egg becomes a tiny worm… only to decide afterwards that evolution can, in my opinion, not be sufficiently studied in real life, but only in computer simulations (sorry, Rich, what you are doing is great!). So I turned to artificial life and artificial evolution, where I developed the “artificial cell model”: a computer simulation that is capable of evolving the metabolism of cells, building complex molecules from simple precursors using a metabolic network of interacting catalytic enzymes. One of the main concepts that I developed was how to translate a genome into a complex network, and how to analyze these networks.
At the same time I was intrigued by how brains make decisions and what might influence their outcome. So I also worked on evolutionary game theory, where I let virtual agents play simple games like the well-known Prisoner’s Dilemma, for example. The key is to use a set of probabilities that determine the strategy of a player and let these probabilities evolve. I learned a lot about evolutionary principles, population dynamics, and cooperation. But the key concept here is that if agents have information about each other, for example by communication, they can evolve strategies that are better in the long run than those that thrive for short term benefits. I have to point that even though you might have read something about evolutionary game theory, most people actually screwed up the evolutionary part of it. The take home message here is: If you read evolution, make sure it means: inheritance, variation, and natural selection. Everything else is just not it! I had a very disappointing month reading myself through 30 years of “evolutionary” game theory literature.
Anyway, here is an interesting question: “Why are there no robotic doctors, self-driving cars, or computers that write books yet?” After all, we are more than 40 years into micro-electronics, and the founding fathers of computer science like Alan Turing were already interested in intelligence. Apparently we are doing something wrong. The two major concepts in AI either try to build an insanely large database on top of a complex algorithm that “knows” how to deal with it (IBM’s Watson comes to mind) or we use some form of neuronal network that is great when it comes to discriminating between different situations, but has to be taught from the outside to explicitly solve one and only one task.
So we are currently developing NEUERA, an evolvable system that uses lots and lots of small probabilistic logic gates, like a computer only that it uses probabilities instead of ones and zeros. These brains are designed to be evolvable, and at the same time are known to be able to solve any computational task. This system fundamentally shifted my work from designing an AI into building worlds that are conducive to evolving intelligent behavior. We have been quite successful with this project – we have robots that navigate something like a maze, learn the difference between large and small objects just to catch the small ones and avoid the large ones, or robots that cooperate in pushing blocks around and help each other find the right places for that. We are embedding these brains in mobile rovers that should seek moving objects and flee from larger “predators.”
All of this is just a beginning, and we are very thankful to have ICER as our partner for doing all these very computationally intensive things!
I also work with Prof. Titus Brown developing a web portal that should allow biologists to rent cloud computation resources to solve their bioinformatic problems mainly in nextGen sequencing and genomics. In brief: I learn how to use more and more computers at the same time.
I doubt that the above sufficiently explains what I am doing, but at least you should understand why I use “computational biology” as a weak abbreviation for “learning everything about evolution and computer science to evolve conscious machines, while at the same time understanding evolutionary game theory, network theory, complex systems, and systems biology, because all of that is necessary to achieve my goals.”
For more information about Arend’s work, you can contact him at ahintze at mac dot com.