kmguru
Staff member
Opinion: Time for another AI reality check?
Bill Schweber
(04/02/2008 2:58 PM EDT)
URL: http://www.eetimes.com/showArticle.jhtml?articleID=207001349
I love to read stories about how software and hardware that will emulate the brain is just around the corner. Of course, this "corner" always seems to be 2, 5, 10, or even 20 years away. A recent example was presented at the International Solid-State Circuits Conference (ISSCC), "Keynoter says chips will someday mimic human brain," where computer expert Jeff Hawkins said great progress has been made, and gave some specifics.
I love these stories because they actually prove that we really haven't got a clue about how the brain works. Look at research papers from brain and artificial intelligence experts: Despite the progress, there are embarrassing gaps in understanding.
Consider the Defense Advanced Research Projects Agency competition, which brings together very innovative and proficient teams to develop a self-driving vehicle using megaflops of processor power, kilowatts of dissipation, millions of lines of code and an array of sensors to do what just about any person can do, under a wider set of circumstances and environments: drive a car.
All we need is a three-pound, 75-watt processor supplemented by a few peripheral actuators with position and motion feedback. And unlike Darpa vehicles, our input sensor is just a visible-light array without sonar, thermal, laser, radar or other supplemental imaging support.
We don't know if the brain actually translates and stores information, and, if it does, in what format. Does it digitize all the images, sounds and experiences? If so, at what resolution? (Highly unlikely: There's no MSPS, high-resolution A/D and D/A subsystem.)
How does the brain organize, sort, search, pattern-match and retrieve memories and information? How does it do this so quickly? How does it implement background searches so the name of that movie you are trying to recall finally "pops" into your memory an hour later?
How does the brain figure out how to do what has to be done? How does it learn to learn, and dynamically reconfigure itself for different tasks and priorities? How is it able to recall and regenerate sounds, images, movies or songs with so little apparent effort?
Why can we drive, run and catch a ball with relative ease? There is no megaflop processor and no organized code listing. Even if we postulate, for simplicity, that the brain is a parallel processor with near-infinite memory, how does it do what it does?
Clearly, the brain is configured and works in ways we don't grasp. All those live-action brain scans, which we now see everywhere, obscure that reality. Perhaps someday, artificial intelligence plus radically new computer architectures will allow us to build brain-like systems.
But until we acknowledge our gaps in understanding—and they are huge—more powerful processors, slightly different architectures and better software won't do it.
--------------------------------------------
Very interesting...
Bill Schweber
(04/02/2008 2:58 PM EDT)
URL: http://www.eetimes.com/showArticle.jhtml?articleID=207001349
I love to read stories about how software and hardware that will emulate the brain is just around the corner. Of course, this "corner" always seems to be 2, 5, 10, or even 20 years away. A recent example was presented at the International Solid-State Circuits Conference (ISSCC), "Keynoter says chips will someday mimic human brain," where computer expert Jeff Hawkins said great progress has been made, and gave some specifics.
I love these stories because they actually prove that we really haven't got a clue about how the brain works. Look at research papers from brain and artificial intelligence experts: Despite the progress, there are embarrassing gaps in understanding.
Consider the Defense Advanced Research Projects Agency competition, which brings together very innovative and proficient teams to develop a self-driving vehicle using megaflops of processor power, kilowatts of dissipation, millions of lines of code and an array of sensors to do what just about any person can do, under a wider set of circumstances and environments: drive a car.
All we need is a three-pound, 75-watt processor supplemented by a few peripheral actuators with position and motion feedback. And unlike Darpa vehicles, our input sensor is just a visible-light array without sonar, thermal, laser, radar or other supplemental imaging support.
We don't know if the brain actually translates and stores information, and, if it does, in what format. Does it digitize all the images, sounds and experiences? If so, at what resolution? (Highly unlikely: There's no MSPS, high-resolution A/D and D/A subsystem.)
How does the brain organize, sort, search, pattern-match and retrieve memories and information? How does it do this so quickly? How does it implement background searches so the name of that movie you are trying to recall finally "pops" into your memory an hour later?
How does the brain figure out how to do what has to be done? How does it learn to learn, and dynamically reconfigure itself for different tasks and priorities? How is it able to recall and regenerate sounds, images, movies or songs with so little apparent effort?
Why can we drive, run and catch a ball with relative ease? There is no megaflop processor and no organized code listing. Even if we postulate, for simplicity, that the brain is a parallel processor with near-infinite memory, how does it do what it does?
Clearly, the brain is configured and works in ways we don't grasp. All those live-action brain scans, which we now see everywhere, obscure that reality. Perhaps someday, artificial intelligence plus radically new computer architectures will allow us to build brain-like systems.
But until we acknowledge our gaps in understanding—and they are huge—more powerful processors, slightly different architectures and better software won't do it.
--------------------------------------------
Very interesting...