Do you worry about robots taking over the world? Are you concerned that the Terminator movies are prophetic and that the robot apocalypse is just around the corner.
Not going to happen.
Why? Because robots are increadably stupid and they don't have any mechanisms to become smarter. Basically, current artificial intelligence is not. Intelligent that is -- we are no closer to replicating the thought processes in even the smallest ant than we were in 1968 when Arthur C. Clarke conceived of HAL.
As part of my job I get to work with, see, play with, and go to conferences to talk about the smartest, most capable robots and AI systems in the world. I also have a lot of exeperience with supercomputers. I've built AI systems and spent years as an AI researcher. And I'm convinced that we are all wet when it comes to computers and robots behaving intelligently. We are no only not on the right road to building a self-aware computer, we have not even found the right concepts to discuss it.
You may have seen a lot of new articles like this one, that say that soon computers will be built that surpass the human brain in compute power.
That is complete and utter BS.
While it may be possible to equal a brain's worth of number crunching, we are nowhere near creating a system that equals the brain's total bandwidth and memory capacity - because we are using the wrong materials. Let's look at a computer. We store information in a computer in binary code - 1's and 0's, on or off states. It takes millions of those ones and zeros to just encode a picture, and that's just the visual information.
Our brains, on the other hand, are completely analog systems. A single channel, rather than being on or off, can convey a complete range of values with millions of possibilities. (NOTE: some neurlogical SME will jump in at this point and say, "Wait! Neurons are either on or off" -- that's the wrong way to say it. Neurons either transmit or don't (true) but when they transmit, they provide a range of data, not just a single value.) So rather than having a memory cell that has a single bit of data that is either on or off, we have a single value that has a whole range - and can be a color, a smell, a sound, a touch, a muscle memory, a texture, a weight, and so on. So the total BANDWIDTH of the neural system is enormous. So while the computer can process the data - you can't get that much data in or out as fast as a brain.
To avoid getting too long on this post, let me summarize in a simple statement.
Untill we grow our computers from analog components
and teach them rather than program them
and a robot can learn to identify and manipulate objects just like a baby
we are not doing Artificial Intelligence
and you have nothing to worry about.
6/18/2009
Subscribe to:
Post Comments (Atom)
Dear,
ReplyDeleteI'm the author of a big book on Robots, "Robots genese d'un peuple artificiel" (Robots, genesis of an artificial people) which has been translated in US and german and I must say I totally 100% agree with your view ! I also consider such theories of robots dominating man childish; stupid with a main offender : science fiction books and movies which accustomed people to this inane theory.
Daniel Ichbiah
http://ichbiah.com
I would agree with you, but for the research of Numenta? It seems as though they are on the right track with their cortex-modeled self-learning software.
ReplyDeleteRegarding your criticism of traditional AI as all about binary "on" or "off" I would further point out that Numenta's software uses a hierarchy of learning nodes based on the columns in the cortex that uses bayesian belief propagation to assist in making its decisions. In other words, the software, as a whole, is using probabilities rather than "on" "off" just like the human brain.
ReplyDelete