This article also illustrates how important it is as a researcher to publicize yourself well. Mogo really only had a very small part in the breakthroughs that w

This article also illustrates how important it is as a researcher to publicize yourself well. Mogo really only had a very small part in the breakthroughs that were made. Yet it is heralded in every such article as the bot that made it all happen. For those that are curious what happened, there are some seminal references: Monte Carlo Go. Bernd Brügmann - This introduced the basic idea. Monte-Carlo Go Developments. Bruno Bouzy - Bouzy is the guy that kept hammering at the original Brügmann idea while everyone else thought it was a dead end. Efficient Selectivity and Backup Operators in Monte-Carlo Tree Search. Remi Coulom - This was really the breakthrough paper, that showed how to combine Monte Carlo simulations with a tree search. Was also the first MC bot to win the gold medals. Mogo replaced the ad-hock operators in the last paper with an algorithm with more mathematical grounding (UCT). Even though the name is now very well known, most top Go programs, including the last research published about Mogo, actually already dropped the idea and went back to more ad-hock formulas again. The article also makes the common mistake that the position is evaluated using random games. Calling them random is quite misleading. It seems to be necessary to include some knowledge in them, but too much knowledge hurts the strength again - regardless of any speed impact. It's an open problem to fundamentally understand exactly what kind of knowledge helps the program to actually play stronger. What happened in Go is also a nice illustration that if you want to make a computer good at an AI task, the key is finding an algorithm that approximately-linearly improves with the computing power you throw at it. If you do this, you crack the wall. Algorithmic and hardware improvements will do the rest in the years to come.