The Machines Are Coming!!

The AI revolution is upon us. Most of you would have listened to the telephone conversations Google Duplex had with humans. Most of you must also have marvelled at its astonishing ability to hold a conversation so nicely. It doesn’t matter that people pose moral questions ranging from the blatantly ridiculous “Who gave Google the right to treat telephone operators at small businesses as guinea pigs?” to the more nuanced “To make AI sound more human, would we teach it cuss words?” It really doesn’t matter what questions people ask. And if you think that AI is only going to remain being your personal assistant in the phone and can talk like Morgan Freeman or Amitabh Bachchan, think again.

It wasn’t too long ago that people who follow the game of chess saw something that blew their minds. Google’s AI Alpha Zero learned to play chess. They just fed it the rules of the game. Then they gave it four (yes, FOUR) hours to practice by playing against itself. No “greatest games by GMs” stored in its memory, no “endgame table base” or anything like that. Just the rules and practice. Then it played a 100-game match against a very highly rated chess engine called Stockfish.

People swear by Stockfish. It has an ELO rating of upwards of 3400. By comparison, the current World Chess Champion (human, heh) Magnus Carlsen currently has an ELO rating of 2853. His best was (I think) 2882, which happens to be the highest rating a human has ever achieved. Considering that these days you need to be around 2400 to achieve the title of Grand Master (an estimate), the gap between the best rated human and the best rated engine is clear. There are many engines in the 3000+ club, make no mistake.

So what happened in the match between Stockfish and Alpha Zero? Nothing at all. I mean, if you can call 28-0-72 score line (wins, losses, draws) in favor of Alpha Zero “nothing” then that’s what happened. I read somewhere that Alpha Zero could be rated over 4000 looking at its performance. Not sure about that. Google only released 10 games out of the 100 that took place. I also heard that the Stockfish version wasn’t the latest one. Die-hard Stockfish fans also say that if Alpha Zero played against the “real” Stockfish, it wouldn’t stand a chance. From the games and analyses that I saw on YouTube for the released games, I really don’t think so. Any Stockfish version wouldn’t stand a chance. But that’s just my opinion, and I’m less than 1400 ELO, never really checked. Nothing to write home about.

The point is, it doesn’t matter that it wasn’t the “real” Stockfish or that Alpha Zero isn’t 4000 ELO. The point is that we now have the AI that can teach itself and in a short time can compete with the best in that field. Why is that so bad? It excites me to no end. AI-assisted research areas are just opening up. Asimov’s words ring in my ears, “Don’t let your sense of morals prevent you from doing what is right.” His character Salvor Hardin says this in the first book of the Foundation trilogy.

So, AI. Let it learn, teach it to help us. This is quite an opportunity. I’m sure that people at the tech companies will go forward with their projects silently. I’m also sure that protests will mount as and when AI starts making inroads into our daily lives. Google has already conceded a little bit of ground by agreeing that the AI will identify itself as a robot at the start of any interaction over the phone with a human. Again, I’m sure that this is a step to curb at least some of the protests and criticisms.

However, I do think that the march of the AI will continue and will thrive in the not-too-distant future. There was some news recently that Facebook technicians closed down some of the AI systems that they were taking help of. The reason given was that the different AI systems developed their own language and started conversing. I think this phenomenon could be further explored under stringent lab conditions. How they do it, how the new language evolves, how the other AIs pick it up, how different is the new language from the language(s) they already know... I’m very excited about this too. Are you?