2008 : WHAT HAVE YOU CHANGED YOUR MIND ABOUT? WHY?

rodney_a_brooks's picture
Panasonic Professor of Robotics (emeritus); Former Director, MIT Computer Science and Artificial Intelligence Lab (1997-2007); Founder, CTO, Robust.AI; Author, Flesh and Machines
Computation as the Ultimate Metaphor

Our science, including mine, treats living systems as mechanisms at multiple levels of abstraction.  As we talk about how one bio-molecule docks with another our explanations are purely mechanistic and our science never invokes "and then the soul intercedes and gets them to link up". The underlying assumption of molecular biologists is that their level of mechanistic explanation is ultimately adequate for high level mechanistic descriptions such as physiology and neuroscience to build on as a foundation.

Those of us who are computer scientists by training, and I'm afraid many collaterally damaged scientists of other stripes, tend to use computation as the mechanistic level of explanation for how living systems behave and "think".  I originally gleefully embraced the computational metaphor

If we look back over recent centuries we will see the brain described as a hydrodynamic machine, clockwork, and as a steam engine.  When I was a child in the 1950's I read that the human brain was a telephone switching network.  Later it became a digital computer, and then a massively parallel digital computer.  A few years ago someone put up their hand after a talk I had given at the University of Utah and asked a question I had been waiting for for a couple of years: "Isn't the human brain just like the world wide web?".  The brain always seems to be one of the most advanced technologies that we humans currently have.

The metaphors we have used in the past for the brain have not stood the test of time.  I doubt that our current metaphor of the brain as a network of computers doing computations is going to stand for all eternity either.

Note that I do not doubt that there are mechanistic explanations for how we think, and I certainly proceed with my work of trying to build intelligent robots using computation as a primary tool for expressing mechanisms within those robots.

But I have relatively recently come to question computation as the ultimate metaphor to be used in both the understanding of living systems and as the only important design tool for engineering intelligent artifacts.

Some of my colleagues have managed to recast Pluto's orbital behavior as the body itself carrying out computations on forces that apply to it.  I think we are perhaps better off using Newtonian mechanics (with a little Einstein thrown in) to understand and predict the orbits of planets and others.  It is so much simpler.

Likewise we can think about spike trains as codes and worry about neural coding.  We can think about human memory as data storage and retrieval.  And we can think about walking over rough terrain as computing the optimal place to put down each of our feet.  But I suspect that somewhere down the line we are going to come up with better, less computational metaphors.  The entities we use for metaphors may be more complex but the useful ones will lead to simpler explanations.

Just as the notion of computation is only a short step beyond discrete mathematics, but opens up vast new territories of questions and technologies, these new metaphors might well be just a few steps beyond where we are now in understanding organizational dynamics, but they may have rich and far reaching implications in our abilities to understand the natural world and to engineer new creations.