A quick thought on the state of technology. In many areas, technical sophistication of machines and software is already better than the corresponding human systems. But the performance is considerably less, which leaves lots of room for improvement.
Try this experiment: Find an average computer (as of this writing lets say something like a Pentium III 1.4 MHz Windows-based machine as baseline) and install professionl audio software (e.g. Cubase) on it. You have just created as system that has thousands of times the processing power and resolution of the human ear. Now, get the computer to recognize what someone is saying in a German accent, while driving a car in the rain, with no constraints on vocabulary or syntax. I'll guarantee you that the ear wins everytime.
This begs the question: What are we doing wrong? For the tasks that they are designed for (mathematical calculation, sorting and searching of lists and storage and retrieval of information*, maybe even chess) they have surpassed human ability ten-fold, and doing these things by brainpower rather than processing power is no longer reasonable. But some of the simplest human tasks are still impossible for computers to perform.
*This advantage is only valuable for information that is not stored in the brain's memory (for example, finding a name in the phone book). Computer's excel at locating items contained in large volumes of information which they were previously unfamiliar with. However, for information stored within the brain (your uncle Bill's phone number, maybe) the brain is still far ahead, especially since it can retrieve it automatically by looking at a picture of uncle Bill, or thinking about the $5 you owe uncle Bill, or noticing uncle Bill's copy of "Stranger in a Strange Land" on the table and realizing you need to call him.
Very young children learn very easily to identify their mother or father from a group of very similar looking people, even though their brains are not yet fully developed. Moreover, most children could pass my "rainy German" test above. But even with the best computers in the world, these problems present a challenge to the computer.
It's pretty well accepted in the study of the brain, that one of the brain's biggest strengths is pattern recognition (which describes both of the problems above). Work has been done with computers in the study of "neural networks" in an attempt to duplicate the behavior of the brain. But even with enormous networks of neurons, it is difficult to match the brain in these problems. Current estimates put the processing power of the brain on the order of 10^12 - 10^15 operations per second. The world's fastest supercomputer runs at around 40 Teraflops or 4 * 10^13. So, from a pure power standpoint, there is no reason computers should not be quickly outpacing the human brain, but these calculations don't seem to tell the whole story, and more importantly, floating point operations do not appear to be the foundation of the brain's thinking power, but rather relationships between things.
To me its pretty clear that brute force is not the path to understanding the power of the brain or harnessing it. Some current research in psychobiology suggests that the structure of the brain itself (protein structures called microtubules) play a part in brain operation and are the mechanism for consciousness. In any case, I see enormous potential for great advances in computer science, but it will require us to stop thinking of the brain like a computer, and starting making computers like the brain.
Notes:
I got some information from David J. C. MacKay's book, "Information Theory, Inference, and Learning Algorithms" which is available on-line for free. (Although if you think you'd use it, you should buy it and support the author).
It contains this really neat passage (which I think I can reprint out of a 640 page book by fair use):
Biological memory recall is error-tolerant and robust.
Errors in the cues for memory recall can be corrected. An example
asks you to recall `An American politician who was very intelligent
and whose politician father did not like broccoli'. Many people
think of president Bush (even though one of the cues contains an
error.)
Hardware faults can also be tolerated. Our brains are noisy lumps
of meat that are in a continual state of change, with cells being
damaged by natural processes, alcohol, and boxing. While the cells
in our brains and the proteins in our cells are continually changing,
many of our memories persist unaffected.
A paper on estimating human memory and another on the computation limits of the brain. (Both are quite old, but still hold up to our current understanding.)
Stuart Hameroff's webpage, a leading researcher into consciousness.
Posted by ktismael at March 15, 2004 11:26 PM