We’re going to set aside the usual cliches about machine learning and artificial intelligence in this blog.
But before we do, I’m a sentimental character so let’s list them; glowing red eyes (Terminator), fiendishly evil computer minds (HAL, 2001: A Space Odyssey)....and laser cannons slung over one shoulder (yes…oh yes... Short Circuit).
It’s funny how the mind always turns to the potential for AI to become a malign influence. Why hasn’t anyone made a film about robotic nurses? The answer… probably for the same reason you may not have heard of AlphaZero.
For the robotic chess prodigy and a bed wash have one thing in common... they make for terribly poor action sequences.
But that doesn’t mean its feats are any less deserving of the sort of wide-eyed, opened-mouthed, tongue-unfurling amazement that you see in the cinema.
Where AlphaZero is concerned, the drama is unseen, and it wasn’t just what it achieved that confounded critics but the time it took the computer to do it.
Having been set the challenge of teaching itself to play chess, it did something no other machine has ever done.
It taught itself to play the game in four hours, devised moves never seen before in its 1,500-year history and beat a chess grandmaster.
Occasionally computing wizards have rolled out robot minds to show the world how clever they are. But this, I think, is the first time the general public have been shown something that truly gives them a flavour of what machine learning and artificial intelligence truly is, hinting at what it will be capable of delivering over the next two decades.
We’re not talking Deep Blue, which was programmed to play chess.
Deep Blue was billed as an ‘AI machine’ and it was the first computer to beat a grandmaster, Garry Kasparov, but it benefited from a rule that meant it was allowed to be re-programmed in between matches. It wasn’t left alone to adapt on the field of play, by itself, in the same way as AlphaZero.
AlphaZero wasn’t programmed to play. It was programmed to learn. And it could do this all by itself.
That’s why AI has been so misunderstood over the years. In the early days the presence of AI in these types of computers was more mirage than reality. It’s not just about being smart using programmed knowledge. When the programme is capable of expanding its own horizons - that’s the game changer. And that’s what we are focused on ourselves.
Real AI, as we understand it, might make headlines with chess moves but AML involves its own game of chess.
The good guys try to chase the bad, while the bad try to anticipate the tactics that AML agencies and executives will use to catch them. This is a fantastic use case for machine learning and shared learning in the industry means the costs of these systems are no longer prohibitive.
If computers can act as detectives in their own right and devise their own methods for investigating, having been given only the rules of the game (the law in this case), then the AML industry will become hundreds of times more effective.
Money launderers and fraudsters were the kings of crime but the tables are turning and they’re soon to be the pawns in our great game.