Why are people so fired up about a computer winning yet-another-boardgame?
Hint: it’s not about winning a board game
Why are people so fired up about a computer winning yet-another-board-game?
We saw similar media mania around Deep Blue beating Kasparov in 1996–97 but that didn’t lead to general #AI breakthroughs
But here are some reasons AI researchers are fired up about AlphaGo beating Lee Se-dol and how it might presage much more general
#AI breakthroughs with applications far beyond board games
First, Go is much more complicated than chess. Specifically, as Demis Hassabis of Google’s DeepMind points out, Go has a googol times more possible positions compared with chess.
Because you can’t brute force search through the possible moves or just use a catalog of pre-defined openings and endgames, a whole new set of algorithms were needed to win Go
Second, it shows how an ensemble of different learning techniques (deep learning + reinforcement learning + Monte Carlo tree search) can yield remarkable results, much better than results from a single technique
Monte Carlo tree search, for example, is an unsung hero of AI techniques, (since deep learning gets all the glory),
as it’s a great way of exploring a large tree without brute force visiting every possible node
Third, it perfectly illustrates the pendulum swing in AI research from expert systems—programs designed to capture human expertise in algorithms—to algorithms which learn themselves.
Another class of continuous self-learning called genetic algorithms led to this delightful system
written by Seth Bling called MarI/O that learned to play and ace Super Mario World:
Finally, as Demis Hassabis of Google’s DeepMind writes on their blog,
The game is played primarily through intuition and feel, and because of its beauty, subtlety and intellectual depth it has captured the human imagination for centuries.
That is, the oh-so-tantalizing prospect is that with this AlphaGo win, we are the dawn of an era where we can program computers to mimic human intuition and flow—or demonstrate novel inutition and flow of their own
Rather than just solve problems with brute force using lots of processors and large, reliable memories
So is this #AlphaGo win the beginning of real progress towards generalized intelligence compared to the false start of winning chess?