So… what is it you do again?

Just had an interesting afternoon with a friend in London, who works for the government evaluating research funding requests. Her area is health care, the whole curing diseases bit, which is concerned more with cancer than with ALife. It was great to have a conversation with a skeptical biologist about what exactly ALife is / can do, in the research group no one generally wants to begin the really interesting conversations starting with “So, what’s the point of your work again?” I hope (for her sake) that I did a decent job explaining it in semi-layman’s terms, but the entire train ride home I was thinking of how I could have done it better, and what I’d say if I had another chance.

Artificial Life:

(my) Definition:

A scientific field that abstracts processes of living systems into algorithms, and uses these algorithms to better understand the limitations and capabilities of the systems. The engineering branch of ALife then applies the algorithms to solve computational problems in a variety of areas. A third philosophical branch tries to identify a core group of “living processes.”

Artificial Life is related to Systems Biology, but differentiated by the philosophical focus on discovering and implementing life algorithms in traditionally non-lifelike systems.

Artificial Intelligence intersects Artificial Life as well, sometimes leading to confusion even among the researchers. Many sectors of AI are focused on learning as a tool to make complex decisions, but learning has traditionally been seen as a process only present in living things. Generally, if the definition of learning expands into adaptation and search (the three are often difficult to distinguish, especially in simulations), learning and therefore Artificial Intelligence encompasses a huge variety of life-like processes, and the distinction of AI and ALife becomes difficult, though largely pointless. The main examples of these “muddied” areas are in evolutionary algorithms and neural networks. However, the AI researcher would see his “science” as extending the capabilities of these algorithms, while the ALife researcher would see his “engineering” as extending the capabilities of these algorithms, sometimes creating a “What’s the point?” mismatch.

Artificial Life is a controversial field for two reasons:

A) Notions people have about the inherent unknowability of life, and inability to identify or recreate similar behavior in non-biological systems.

All of these arguments eventually boil down to the idea that “life is special.” Ignoring any religious or non-materialist notions, this is a pretty untenable position. Two hundred years of relentless scientific progress have mapped pretty much every process in an organism to universal physical and chemical laws. To think that these laws, based on mathematics, cannot be understood through computational means is in my view, very naive. I want to make the strong point that these physical and chemical systems are the most complicated systems in the universe, and I DO NOT assume that translating them to computation will be easy by any measure, but in the 21st century we do have the tools to start trying (computers). I also want to make the point that I also believe that “life is special,” (otherwise I wouldn’t be interested in it!) just not that it is unknowable.

B) The idea that abstracting a biological system is a useless, unless you want to better understand the biology.

This is the basis of a lot of criticism from well-respected parts of the biological community, and is often very valid. If, for example, we’re trying to study the evolution of life on earth, why would someone create an artificial world with intentionally unreal “living” things.

To address this criticism, I just want to quickly jump into early physics, for a comparison with biology. Historically, the science of physics was built from the simple upwards: “toy” systems of point masses, frictionless surfaces, pulley systems, to name only a few. In the very early Newton times, it was hypothesized that maybe all the complicated behavior we see in the real world (apple trees bending in the wind, bicycles, the motion of the planets) could all be explained as just really complicated applications of these simple laws, acting in a variety of ways. And the hypothesis was correct, this small set of equations DO explain everything physical, except for the really big and small (for that Einstein had to make the laws a bit more complicated).

Biology was built a different way. Anything we observe biologically is horrendously complex, so the way early biologists tried to “understand” biological things was to basically observe in minute detail every aspect of an organism, and write them down in a big book somewhere. This type of approach continues to this day (not in all areas of biology), though at smaller and smaller scales, and has worked extremely well in particular applications. These really complicated systems are important for keeping people alive, so it makes sense to know all their parts. However, while writing all the details down in a book is helpful, it isn’t true understanding. Biology is fundamentally a science of organization and processes, the particular atoms that make up “me” will change drastically from day to day, but the processes keep me alive (some might say define me as alive, but that’s another argument for another day). The “point” of Artificial Life is to find the simple, “toy” building blocks of these processes, so we can say “this organism / species / ecosystem is incredibly complicated, but every part uses basic organizational laws that we can understand.” The correct place to look for “basic” building blocks of processes is now in research on algorithms and computing (which, though the general public has yet to learn it, has NOTHING to do with computers, and EVERYTHING to do with processes). If biology is organized processes, then computer science is the basis of biology.

Artificial Life “toy” problems can then be understood as attempts to understand the building blocks of biology. By abstracting a process down to its mathematical core, its actions become predictable in a useful sense, and that knowledge can also be linked to the rest of science. While simulating a realistic system starting from these mathematical relationships may be out of reach for generations, or forever (Newtonian dynamics generally can only be approximated in realistic systems, even now), the algorithmic understanding of these processes allows for valid approximation, as well as distinctly artificial engineering. The internal combustion engine is NOT something I would have built after observing apples fall from a tree, even if they were described in minute detail. Observation is important, but so is the mathematical abstraction.

I had more to write on the nature of simulation and computer science in general, but it isn’t coherent enough to bother with yet… perhaps later.