World-class AI experts recently predicted it would take ten years for a machine to beat a human at Go. A month later it was done
BY VICTOR WALLACE HUGHES JR.
In the “Terminator” movies, Arnold Schwarzenegger plays a robot that acts like a human (versus a human political candidate who acts like a robot).
The plot: machines become sentient and seek to extinguish all human life. An important date in the series is Aug. 29, 1997, the date that the artificial intelligence known as Skynet becomes “self-aware.”
Although that date was subsquently shifted around in later Terminator films (apparently, time travel will do that), an anniversary we might want to mark on our calendars is Jan. 27, 2016.
That’s my pick for the date that the light bulb – or the self-awareness CPU of an Artificial Intelligence – may have switched on in real life. On that day, Google’s subsidiary, DeepMind, announced it had used deep machine learning, or AI, to beat the European champion in the ancient Chinese game of “Go.”
Go is a board game played on a 19×19 grid. Players place white and black stones on the grid intersection points to try to surround and capture their opponents. The player with the most stones wins (pretty much like any activity).
While it may seem Go is an easy child’s game, in reality it is extremely complex, requiring a high degree of strategic and intuitive skill. Great Go players say that they not only focus on Go’s strategy, they ‘feel” a good board that “just looks right” – using intuition, a skill that is uniquely human. Until now, that is.
While machines like IBM’s Deep Blue have beaten world chess champions by “brute force,” – computing every possible future move, chess is child’s play compared to Go. Even the world’s largest supercomputer currently can’t crunch all possible Go outcomes. So the DeepMind’s people took a different approach.
Without getting too technical, they created a machine that analyzes patterns and then let it loose on a data set of 30 million Go moves by humans. I don’t know how many different people and games that is, but using a rough back-of-the-envelope calculation (sorry deep learning), that works out to one individual playing Go for eight hours a day for their entire working life. That’s a lot of Go and a lot of actual experience.
The machine then created some rules, with human help, and then created a buddy machine so it could have someone to play (the lonely life of machines). After playing several million games, the equivalent of many human lifetimes, it evaluated what worked and modified itself so it could play better. It learned. Then it smoked the European Go champion 5-0. The skills that win at Go will work on lots of other problems.
While the machine analyzed 30 million moves and several million games, there will be no reason it couldn’t ultimately process, or intuit, its way through 30 billion moves and several billion games – so far beyond human experience as to be frightening. What insights, I mean, patterns, will it find?
The critical thing to remember about machine learning is, “The more data they get, the better they get; the better they get, the more data they get.” Forever.
But what about human intuition? Well, what about it? What if it turns out that intuition is just human pattern recognition from our past experiences – our human shortcut in analyzing 30 billion moves in Go? Intuition worked well enough through humankind’s evolution, but in an experience arms race, we are but a grain of sand to a machine’s Rocky Mountains. And the machines will pull away faster and faster. The more data you get …
So where does that leave us? Machines will solve human problems, like cancer, in ways still beyond our comprehension because of their ability to see patterns we can’t.
Machines will also evolve in ways no one can predict. World-class AI experts recently predicted it would take ten years for a machine to beat a human at Go. A month later it was done. Nobody knows where this is Going.
So mark your calendars for a 21st birthday celebration on Jan. 27, 2037. Let’s hope it’s the good Terminator that shows up for the party.
Victor Wallace Hughes Jr. is the CFO of OAG Analytics of Houston, a company that uses Big Data to improve the efficiency of energy exploration.