Re:Ray Kurzweil 2019 Predictions
by jcranmer on Wednesday January 02, 2019 @10:41AM
(#18806845)
A few things went wrong.
The first is the evolution of microprocessors under Moore's Law. Since about 2005, we've come to realize that we can't afford to cool chips anymore if we crank up the frequency. What that means is that the advancements in building silicon is much less efficient at being turned into sustained computation rates. So building the exponential growth curves predicted required (and still requires) revolutionary thinking about how to build computers, rather than just easy gains each generation.
The second thing that went wrong was what Derek Lowe refers to as the "Andy Grove Fallacy." This fallacy is that every hard problem (particularly in biological systems) can be solved by just shoveling computational power at it. In the modern day, we still can't model the neurological system of a nematode that has just 302 neural cells, let alone understanding how the entire organism works. Multiply that to the scale [1] of a human being, and we are truly ignorant of how things work except in the most general sweeping statements possible.
The final thing is, well, people tend to underestimate the difficulty of problems. Things such as bipedal motion or turning a doorknob are problems that seem utterly trivial to us human beings (because we learn them before we can remember the process of learning!) but are absolutely maddening to robots.
[1] Recall that complexity tends to grow exponentially with scale, not linearly--each pairwise, three-way, four-way, etc. interaction contributes to complexity.
The first is the evolution of microprocessors under Moore's Law. Since about 2005, we've come to realize that we can't afford to cool chips anymore if we crank up the frequency. What that means is that the advancements in building silicon is much less efficient at being turned into sustained computation rates. So building the exponential growth curves predicted required (and still requires) revolutionary thinking about how to build computers, rather than just easy gains each generation.
The second thing that went wrong was what Derek Lowe refers to as the "Andy Grove Fallacy." This fallacy is that every hard problem (particularly in biological systems) can be solved by just shoveling computational power at it. In the modern day, we still can't model the neurological system of a nematode that has just 302 neural cells, let alone understanding how the entire organism works. Multiply that to the scale [1] of a human being, and we are truly ignorant of how things work except in the most general sweeping statements possible.
The final thing is, well, people tend to underestimate the difficulty of problems. Things such as bipedal motion or turning a doorknob are problems that seem utterly trivial to us human beings (because we learn them before we can remember the process of learning!) but are absolutely maddening to robots.
[1] Recall that complexity tends to grow exponentially with scale, not linearly--each pairwise, three-way, four-way, etc. interaction contributes to complexity.