Planes don’t fly like a bird…

I was excited and anxiously following the matches between Google’s AlphaGo program and Lee Se-Dol, excited when AlphaGo won the first matches, and frankly even more ecstatic when Lee Se-Dol beat the program in a match. There is a another match to go, that I’ll be anxiously waiting to see the results.

The specific value of this program and it’s success is fairly minimal, but what it represents is pretty enormous. AI has been all over the news, it seems to be the hot topic in the echo chamber of some VCs and silicon valley, and recent advances have cemented wins in the space that a decade or more ago were thought “a long ways off”.

The collection and processing of massive amounts of data have led us to using more data-intensive mechanisms (like deep knowledge networks), and new insights. The whole venture of AI is still transforming, and just like we saw a shift in the 70’s away from perceptrons and neural networks towards symbolic and calculus based systems, I suspect we’ll see a swing in the opposite direction in AI research now as well – everything to “deep learning”, and quite a bit less on symbolic algebra.

What’s so interesting to me is that AlphaGo is a fusion of the two mechanisms. It’s using deep learning and intensive data mechanisms that mimic neurons in combination with symbolic algebra and search functions. Planes don’t fly like birds, but we none the less derive incredible value from them. AlphaGo represents something like that – solving the incredibly complex problem space of the game of Go. The wright brother’s first plane was just as exciting. Not at all like a bird, but it lofted into the air and powered itself. There is plenty of room for advancement and improvements and this represents (to me) a breakthrough advance that starts that process.

I wish Iain Banks was still with us, I’d love to hear what he thought of this advancement.