AI and Us (a 3-part story about the complicated emerging relationship between humanity and AGI).
Part 1: Beyond comprehension.
Some of you may know that I used to play chess competitively in my childhood. Represented my state(s) in the Nationals, trained with the masters of the age, my contemporaries went on to become India’s first set of Grandmasters, I can still manage 30-40 moves of blind chess (where you play without looking at the board) etc. Basically I am humble-bragging to let you know that I was fairly good at the game, and even though I haven’t engaged in the game actively for the past 2 decades, I am a fairly ‘advanced observer’ who can understand games of GMs and connect the dots fairly well. Even though I can’t play anywhere close to their levels, I can understand their strategies, I can appreciate the tactics and the traps, and I can directionally predict what’s going to happen.
In 1996-96 when Garry Kasparov (the then reigning world champion and all-round chess genius) took on IBM’s Deep Blue, I watched with rapt attention as the machine out-muscled and out-calculated Kasparov (it could calculate deeper and wider). But the human vs machine debate continued for another nine years and the ratings were comparable.
In 2005-2006, Hydra and Deep Fritz took the first step jump vs humans and crushed their opponents in man vs machine tournaments. They were followed by Stockfish and many other programs that basically cemented the machine superiority of the game, and humans gracefully accepted defeat and stopped doing man vs machine tournaments. We were summarily defeated.
Humanity was humbled (albeit in a very narrow use of intelligence), I was in awe, and I appreciated the beauty and brute power of the algos.
But I was not scared. Yet.
Then Deepmind launched AlphaZero in 2016. It was basically a ML program that was given no database; it was given the rules of the game and it trained by playing against itself (hence the Zero in its name). After all of just 9 hours of training against itself, it demolished the best programs of its age (which were already playing at a superhuman level). Completely destroyed.
That’s not the part that scared / amazed me.
I saw the games it played against Stockfish. And for the first time in my life, I could not UNDERSTAND the game.
It played its middle games as if it had an intuitive, almost visceral understanding of the game where it made sacrifices and gave away material ‘just because’, that would eventually lead to a positional advantage, that would cripple some part of the opponent’s board, that would eventually lead to the demise of the opponent.
It was magical to watch the game, that had become so mechanical and so overprepared, suddenly become creative, unpredictable and a real treat. The gameplay had transcended to a level where I started finding human vs human games boring.
All this, from 9 hours of training against itself. That too 6-7 years ago.
We have no idea what AI is capable of. Anyone who claims to have an idea is unable to fathom what lies ahead. We are at the inception of an age that is beyond our comprehension.
[Part 2 coming soon].