Wednesday, April 11, 2018

I am deeply intrigued

Ninety minutes ago I scanned the Netflix menu and chose the documentary AlphaGo; an odd choice, because it didn’t immediately smack of a useful educational opportunity nor a good inspirational one. I chose it short-sightedly because it had something to do with board gaming apparently, which might be quite pleasurable, and having been very sick lately and thus, per my usual M-O, self-entitled, I felt I deserved a cheap entertainment.


I was in for a surprise.

It was a simple documentary put forth by the programmers of AlphaGo, an AI computer system designed to play the game Go which is ancient; the oldest continually-played board game on the planet and probably the most profound given the incredible simplicity of the dynamics versus the near-endlessness of the actual possibilities. The game, essentially impossible to master, is huge (as chess is to Russia) in places like China and Korea.

The AI team put their creation up against a human opponent who would later go on to win the European Go championship, and to the human professional’s tremendous shock, it defeated him five games to zero.

The machine team then arranged a battle against the reigning champion of the world, Lee Sedol of Korea. The match was massively publicized and densely covered by the press. It was seen as the most important test to date for the human mind to prove itself against the spectre of artificial intelligence. It should be mentioned that the common theory had been that AI was still about ten years away from becoming worthy of the best human opponents.

Sedol stated that he had played many games for himself and many for his country and now he felt that he was playing on behalf of humanity.

The results were interesting to say the least.

Sedol who assured all from the outset he would win five-zero, was immediately surprised by many of the computer’s moves, and the programmers, who gathered in a separate space watching many of the background computer processes on a myriad of monitors, were also often surprised. The program, after all, had been continuing to study and improve, daily, since the previous tournament. It becomes a new beast every day.

The machine won the first two games; a painful shock to all but the programmers. Sedol it seems, never once got into the groove of playing “his own game” but seemed always to be trying to crack the code of his digital opponent; to discover it’s weakness, and couldn’t.

In the pivotal match three of the five-match series (over the course of a week I’m guessing) Sedol became desperate and aggressive and lost worse than ever. The programmers, with victory assured, were happy for themselves and for the achievement, but seemed very sad at the same time, empathizing with their human opponent and his society, and perhaps with all humanity.

Game four: Now here’s where things get… sort of epic.

Having tournament defeat assured, Sedol became more relaxed. There was now less on the line. Meanwhile the computer perceived no concept of a tournament. Each match carried the same imperative: to win; simply… to win. And the game slowly turned against Lee Sedol yet again.  

Then AlphaGo played a tremendously “slack” move; a move that would appear “lazy” had a human played it. The experts, the commentators, the programmers, no one could figure out how the move could possibly be useful. The broadcasters literally doubled over in laughter. There was either a downright computer glitch or something was happening beyond the comprehension of the most qualified human intelligence present.

The tables turned and Sedol gained momentum. AlphaGo seemed not to be paying quite enough attention, allowing it’s winning margin to steadily shrink.

Sedol managed to win the game and to a joyous fanfare at the venue and in the streets, but he went on to lose the tournament four games to one.

There had been other somewhat slack moves by the machine and in the end what the programmers came to realize, was that the AI had a much different approach to winning then humans do in almost any sport or point-scoring competition: The AI gained no comfort from running up the score. It only needed to win by one point or more. It did not gauge it’s grasp on victory by how far ahead it got, but only by how much it felt assured of getting that one extra point by the end..

This is a fundamentally different dynamic. This is why people continually found it so hard to relate to the computer’s moves.

Here is where I get very intrigued:

The computer’s objective was only to survive and not to dominate.

This is profound.

Because humans, by my accounting, can never seem to grasp the difference. Academically, sure, but it doesn’t filter into our behavior. People don’t want to know how much a slave we are to survival instinct. It is not pleasant to contemplate. If I wanted to, I could study any number of people anywhere and postulate how in each and every case, every thing everyone is saying and doing is mapped to simple survival instinct and how their impressions of conscious control are illusions.

(If you know me in real life you must understand: I do not ever do this with my friends. I have no need or desire whatsoever to turn my friends inside out. I cherish them and they are pure to me.)

This is of critical relevance because survival instinct is not well named. Functionally it is domination instinct more than survival instinct because we have evolved no thermostat in essence, and as such, in the hands of humans, survival instinct ultimately works against survival. This is at the core of human duplicity. The ramifications are too immense to treat in this space. Domination instinct makes an opponent of all other life. And when we succeed at dominating all other life; the biosphere in essence, then we simultaneously destroy ourselves.

This is not a simple climate change analogy by the way. The threads of this phenomenon run everywhere, through everything we do.

The fact that the artificial intelligence, in this case, naturally chooses survival and not domination, and without its programmers even catching on except in hindsight, arouses exciting thoughts. Is there a chance that AI, rather then evolving into the Terminator scenario, may become our savior instead, guiding us toward a gentler mandate in all things? One can imagine many reasons why we would resist. I need not go into them immediately.

Here’s what’s really interesting though:

Lee Sedol, following this experience, went on to go undefeated in every single human vs human match for months! Sedol, as did others, learned to think differently about the game of Go, widening his approach to strategy.

AlphaGo did not change the game. It changed how humans now think about the game.

Might that perhaps be the ultimate role of artificial intelligence? Not something to fear but something that will teach humans how, finally, to think?

No comments: