5/01/2005

Body & Soul III

Before I leave Neuroscience behind, I should mention that there is certainly a lot that computer science can learn from the brain. I wasn't trying to suggest in my previous posts that computational modeling of the brain is not a useful endeavor, only that it is incapable of what most neuroscientists are asking it to do. In fact, some procedures which can only be described as algorithms have been discovered in the brain which rival or surpass in efficiency all previously known methods of accomplishing the same end.

One thing humans are very good at that computers are not is distinguishing signal from noise – i.e. recognizing imperfect patterns. This shows up in speech recognition: the best speech-recognition programs can usually identify (which is not to say understand) speech as long as it is spoken clearly and with the same intonation every time. But humans can understand speech in a range of accents, speeds, and tones. Another example is vision. Humans can (with difficulty) make out lane divisions while driving on the grapevine in the middle of the night in the pooring rain. But in Darpa’s grand challenge last year, computers couldn’t keep a vehicle on the road in the desert at mid-day. The problem is the imperfection of real life. If the road was one solid color and the lane markers another, it would be trivial to teach a computer to drive within the lines... but it’s not.

I don’t think there are any compelling reasons to attribute the current superiority of humans to computers in distinguishing signal from noise to an immaterial cause. Granting the existence of the soul does not change the fact that the brain is incredibly complicated, nor would the discovery of a physical mechanism make this ability of the brain’s (which we perform so effortlessly) any less amazing. Every day the cells of the body perform feats of engineering (such as self-replication) so far unequaled by human endeavor, and they are no less amazing for being physical. I suspect that, as we learn more about the brain, we will discover new techniques for sense processing which will make computers much more able to exist in the uncertainty of the real world.1


1Unless, of course, it turns out that the ability of the brain to distinguish signal from noise is so irreducibly dependent on the chemistry of life that it is difficult to see how to implement it any other way. This is true of many of the cell’s known functions. Michael Denton suggests something like this in Nature’s Destiny: "If the anthropocentric thesis is correct, then the human brain should be the most powerful thinking machine – biological or artificial – that can be built out of the atoms of our world. It should be peerless."

No comments: