Didn’t get around to actually doing it due to too many things competing for my attention (why, I just finished a presentation on basic statistics for role engineers, which was really hard for me being no mathematician at all). I only really picked up the task again, today, after I watched a replay of a documentary on 3sat about this spring’s complexity conference and this supported an issue I was having with some of Mr. Brook’s reasoning.
First, it is worth noting that the book as such is very interesting, well written, and thought provoking. It was fascinating to read how Mr. Brooks begged to differ from the schools of thought prevalent at the time by creating robots of simple components. At a time when other people were building robots (and I’m not talking about the ones used in manufacturing) as primarily intellectual entities who would build a model of reality, then sit down and make an action plan, and finally do something purely based on a mathematical model, he was building a robot (Ghengis) made of very simple individual components where each had very limited alternatives for action but which were interacting with each other and with the real world directly (what he defines as being situated.) Through this interaction those very simple building blocks formed a behaviour that appeared like it possessed intelligent problem solution strategies where they, in fact, completely abandoned what some equate with intelligence, abstract thought.
I was a bit shocked when I realized that early scientist of the field of robotics had actually tried to create entities which/who would even try to build an internal model of reality to work on, where (a bit unfair 25 years later) it seemed so obvious to me that the larger part of life (intelligent or not) just does not work like that. Yes, a human being can close his eyes and form a plan purely on the memory of reality. However, that model is so coarse and the plan so rough, that even then the majority of detail is decided on the fly not based on a model but direct feedback from reality. One could not plan how to carry a cup that has been filled too full. When you even look at it, you will spill some of its content. But if you look straight ahead and let your hand do the work, you will get much better results. Or consider how we manage to focus on the same spot even if we turn our head, or how we pull up our leg if we trip on a snag to avoid a fall. A lot (I’d say the vast majority) of our cleverness works without any abstract model. Of course, you may argue, there is a difference between cleverness and intelligence. However, that may be smaller than we all think. Who can say if the behaviour of Ghengis was not a simulation of intelligence but a germ of the real thing? Connect a billion more neurons, and who knows what will happen?
Mr. Brooks continues with a (to me personally) charming side-kick at Creationism. He observes, though, that as yet, robots have failed to achieve the consistent long-term independence we know from real live beings. He then takes a look at other simulations of life, for example simulations of evolution, where given similar evolutionary pressures we observe in the real world, the simulations come up with fascinating virtual beings evolving from much more simple elements. Such simulations have even produced social beings with different sexes. Such simulations have not, however, produced intelligent virtual beings. There were limits to the improvement of species in them. Mr. Brooks discusses various possible explanations for those:
- In all of our simulation systems there are some parameters incorrect.
- Our systems are build in too simple environments. Everything would develop as expected if we would exceed a certain level of complexity.
- We are just lacking computing power.
- There is something missing from our biological models, perhaps we do need some “new matter”
To my very great disappointment, Mr. Brooks concludes it must be (4). This, I do not believe. And I find it hard to see why somebody who has given such a good practical example of how a complex arrangement of simple things produces something beyond the sum of the parts, would regard non-linear complexity as an insufficient explanation of a lack of organization we see in a simulation when the non-linear system “The World” displays it. And of course, given that the starting parameters for a simulation matching the the world in complexity would be quite numerous (like the state of every quark in the world, where each state may well be defined by a number of properties), the probability of errors in the starting parameters is rather high. Then, slight variations of starting parameters may make huge differences for systems of non-linear complexity.
So, reason 1 & 2 are more than enough for me to explain that we have not been able to simulate the evolution of life on earth to the same level it has in fact occurred. I would even venture so far as to claim that that is as such impossible. I believe, to accurately simulate a complex system, you need a simulation system that is at least as complex. So to simulate the earth, you’d need another earth. (In other words, you cannot accurately simulate at all, if you are interested not just in the object but the object within its environment). For practical purposes, you may not aim at an “accurate” simulation but an approximation and then statistics/probability may help to reduce the data. But even for simulating problems simple when compared to “life on earth”, like climate prediction, our current simulations are so far from doing the real complexity any justice.
What’s especially fascinating here is that this kind of thought seems to be moving out of the fractal, weather prediction, chaos theory weirdo space into mainstream science, where it evolves to an interdisciplinary form of research and brings about progress in seemingly unrelated fields as physics, biology, neuropsychology. It is probably well worth keeping an eye on the Complex Systems Society.