Monday, October 10, 2016

Intelligence and the Limits of Machine Logic

Among the opinion makers and wealthy scions of Silicon Valley, it is treated as an article of faith that Strong AI is inevitable. And like all faiths, this is rooted in sentiment and a limited experience of the world rather than any serious philosophical investigation.

To demonstrate how wrong these assumptions of the techie class are, imagine a chess match between yourself and a super computer. This is not some imaginary You who is an experienced Chess Grand Master, but the actual you reading this. The one who likely has no more chess experience than average, maybe so little that you don't even know the horsey ends its move on a turn. However, you can defeat the supercomputer in your very first move. White or black, you can win the match with only one move and there is no way it can be countered by your hypertech opponent:

Unplug the computer.

You win by default! A computer, no matter how complex or advanced, cannot function without a steady supply of electricity. If it cannot function, it cannot calculate chess moves. So it cannot play and forfeits the match.

"But that's cheating!" you say. Exactly. The entire point of this thought experiment is to demonstrate something intelligent life can do that simply cannot be recreated in a computational system: cheating!

People too often view computers as mystical and irreducibly complex. The truth is that modern computation is indeed perfectly reducible to its component parts, all of which must work as planned for the system to perform its function. And this necessitates a machine logic that is strictly linear.

Anyone with coding experience knows this in their bones, though surprisingly few will admit the reductionist nature of computing. Even though they frequently employ reductionism in their engagement with society... But to elaborate, every program you've ever used, no matter how complex and seemingly reactive, springs from a code sequence that has been laid out along a strict A to B route. It may contain variables from A.1.j to B.7.&, but if it encounters anything not covered in these variables the whole system will freeze up or throw a fault.

Another example illustrating this quality of computation: video games. Currently the manner in which the most people directly engage with AI, so much more pertinent to this discussion than the latest half-baked idea to emanate from a tech billionaire, video games demonstrate both how computational systems can appear adaptive and dynamic while strictly enforcing their own linear logic.

Ever felt like the computer is a cheating bastard? A boss fight that violates established rules on health and damage or, for you youngsters, the enemy's bullets always hit the mark while yours veer off course for the sake of "realism." While aggravating, this is not indicative of conscious cruelty on the part of the game but rather crummy design. Every pixel that flies across the screen to knock pixels out of your pixel is as predetermined as Spinoza's cosmos, and whether the determined events are pleasing or frustrating to you depends on how many of "you" are upset enough to create a marketing problem for the game's publisher. Otherwise you're likely to hear that old saw, "It's not a bug, it's a feature."

The linearality of machine logic is indeed its biggest feature. A graphing calculator can do your high school algebra test better than you, but that's all it can do. Because that's all it was designed to do. And, returning to the heart of our argument, it cannot cheat.

Intelligent life cheats all the time. You, your pets, and the squirrels in your front yard - all can and do cheat. Frequently. Because survival in this harsh and complicated universe of ours requires constant and ruthless adaptability. Real intelligence does not have the luxury of working within a defined system and so must seek any possible advantage.

Example: A sociopathic carnivore with knife-hands.

Further, mere quantitative power is itself no indication of intelligence. Returning to the example of chess, Garry Kasparov famously went win-loss-draw with Microsoft's Deep Blue. Much better than You would have done in the same circumstances but still Kasparov operated within the rules of the computer system, rather than employing the full range of options available to him as a sentient lifeform. He lacked the simple wisdom of Alexander to pull the plug and declare Man superior to Machine.

This is not to call Kasparov dumb. No, he is dumb because of the smorgasbord of conspiracy theories he embraces: NATO invaded Afghanistan to destabilize Russia, Anatoly Fomenko's assertion that human history only begins in the 11th Century AD, and of course that Microsoft cheated in the match that granted Deep Blue a victory. More damning to quantitative logic as the litmus for intelligence, this range of crankery does not make Kasparov an outlier among chess "geniuses." Bobby Fischer was famously anti-semitic, despite being of Jewish descent, and praised Osama bin Laden for the 9/11 attacks.

These cases of chess players - brilliant in their field, bonkers everywhere else - is indicative of the "intelligence" one can expect from a machine system. The system may perform excellently within the bounds of its programming but the unbending logic of programming itself means it cannot achieve intelligence even on par with a cunning adolescent. That the myth of Strong AI has so occupied tech culture is not because it is close to being realized but simply because tech culture itself is a dog wagged by its own tail.