Sunday, December 16, 2012

Artificial Intelligence Is Stupid

Deep Blue was stupid.
Yes, I have heard all about Deep Blue, the chess-playing machine that kicked Garry Kasparov's ass (sort of - only by the skin of its chips, but chickened out of a rematch and never even had a look at the Great Bobby Fisher).

Yes, thank you, I also know all about Watson, the hands-down all-time Jeopardy! champion of the universe.

And yet I assert that today we are as far from true Artificial Intelligence (AI) as we were in the days of mechanical adding machines. Deep Blue and Watson are enormous feats of computer science, but they are not intelligent.  They do not think, do not have any awareness, have no understanding of the data that they process or the rules they follow, and they do not have any level of consciousness whatsoever.  They are, in their essence, just algorithms.  Programs.  Deterministic mechanical data processors.

But if beating every single person in the world at Jeopardy! isn't intelligence, then what is?  A correct understanding of intelligence actually leads me to see playing games like Jeopardy! and chess as not intelligence itself, but as a kind of suspension of intelligence.   Computers and machines are far better than us at things like calculating, data processing and large sorting tasks.  In order for humans to do these tasks well, we actually have to suspend our real organic intelligence while we discipline the mind to follow the rules of an algorithm.  And we have to have a damn good reason for it.  Part of intelligence is instinctively or analytically knowing when to suspend consciousness and which algorithm to run.

Deep Blue and Watson did not have to consider those problems at all. Their programmers decided for them what algorithm to run and when to run it.  All they had to do was follow the rules laid down.  And yes, you do not need to remind me that the fine-tuning of those myriad rules was done by "playing back" thousands of human moves in these games, which implicitly contain the parameters of the most effective, creative human strategies.  At its core this detail is just another deterministic algorithm and, in fact, only further emphasizes the point that human intelligence cannot be synthesized by a machine.

Why is that?  Because the foundation of intelligence is not logic or facts (sorry, all my scientist friends!)  Intelligence is not being able to associate one set of stimuli with the memory of another.  Even flatworms can do that.  That mechanistic activity does not necessarily accompany an understanding of what those stimuli mean.  The core of intelligence is a consciousness of the meaning of facts or events. So, how do humans create meaning and gain consciousness of it?

To begin with, the human mind is integrated into a living organism which has at its very core the instinct for survival.  It wants to survive, needs to survive, and fears annihilation with absolute physical, organic terror. Being integrally connected to living flesh, blood and bone allows the primitive brain to be aware of the internal state of the organism and its prospects for ongoing operations.  It can control the organism and correct for situations that threaten the organism.  It does this not because it is artificially programmed to do so, but because it is terrified of death and therefore of anything that takes it in that direction.

With eyes, ears and other external sensory organs the core brain can assess the organisms environment and its prospects for nourishment, safety, sex and health. The brain is conscious of the meaning it assigns to any set of stimuli because it associates them with either survival or death. These primitive emotions are at the core of all our more complex human emotional states, and that same mind makes assessments of the meaning of all events relative to the organism's health, survival, and genetic success.

Our intelligence derives entirely from our ability to be conscious of and to make meaning of events, situations, and objects in our environment.  It should be noted that these meanings are usually subjective to the conditioning of the mind; hence we have optimists and pessimists making distinctly different meanings for the same events.

Please do not dismiss this as that "emotional intelligence" artsy-fartsy nonsense.  Let's take Garry Kasparov's brand of kick-your-ass braniac intelligence, for example.  How does his brain so aggressively, cunningly and accurately prosecute a game of chess?  Where does the will to do so come from?  Because the primitive brain wants something.  Garry's brain associates winning at chess with organic survival, and so his vast resources of memory, cogitation, creativity and gut feel combine in a highly trained mind to be virtually unbeatable in an extremely logical, cerebral activity that is unquestionably "intelligence" in action.

Once the will, drive and organic resources are in motion, then the champion (in whatever game - chess, Jeopardy!, quantum physics, stock trading, or jazz) must suspend and master his emotions to keep the required outcome firmly in consciousness.

Here is the one issue that AI proponents are not able to solve:  How do you make a program WANT something?  I mean really, desperately, deeply and irrationally WANT something?  Or to be afraid of something?  Or to feel anything at all?

Without that, a machine can never understand the context of information and be conscious of its meaning.   Until machines can be in any way concerned for their organic existence and be conscious of the emotional context and meaning of events relative to its well-being, then machines will never be intelligent.

My Prediction:

Artificial Intelligence will always be stupid.

1 comment:

  1. Very well written. A computer is fed its data by people. Those who worship machines should remember that.