AlphaGo Presentation Review

Rob van Zeijst



Aiming to provide information about Go and artificial intelligence, last Saturday the EGCC showed the documentary film AlphaGo, released by DeepMind, which can now be rented through the usual channels, such as Google Movies.
A 50-person audience of interested parties came to watch the film and attended the presentations by Rob van Zeijst and Leo Dorst.

Rob van Zeijst gave a presentation in which he illustrated the development of the playing strength of the world champions since 1971 and how the various versions of AlphaGo first hesitantly defeated the best humans, only to take a huge leap to dwarf human strength. Rob then showed how Lee Sedol's "genius move that saved the honor of humanity" did not actually work. In short, it was a lecture that both highlighted the great strengths of AlphaGo as well as its small blind spots.






Leo Dorst, scientific expert and lecturer at the University of Amsterdam and a commendable go player as well, gave a presentation on the technical aspects of AlphaGo and AI, such as the working of neural networks, what the various layers of evaluation entailed, and how the Monte Carlo Tree Search fits into the picture.
Leo explained how the Lee Sedol version of AlphaGo had been built with some fancy footwork, because ladders, for example, had to be preprogrammed. The reason was that the AI cannot "discover" long ladders easily precisely because there is a limited horizon. It is also difficult for AI to keep several (seemingly) opposing goals in mind at the same time. Leo also mentioned AlphaGo does not (yet) have a platform incorporating a meta-evaluation function, which means that it has no "awareness" and is unable to evaluate what it does and does not know.

Unfortunately, the audience abstained from the expected heated discussion about the positive possibilities and the dangers of AI. 

The documentary is certainly worth watching and we can highly recommend it to anyone! 
Do you have an opinion about the film or about the computer program AlphaGo itself? Please don't hesitate and share it with us. You can do so on our Facebook page.

Check out the Full News Listing.

facebook twitter