AlphaGo used Monte Carlo methods, which I believe qualify as "brute force"—you generate a bunch of random positions, since the full game tree is way too large.
AlphaGo uses neural networks trained on human play to generate candidate moves, and then Monte Carlo Tree Search to evaluate them deeply. The neural nets are a strong player by themselves.
AlphaGo primarily refutes this essay, rather than supports it. Prior to AlphaGo we had many Monte Carlo bots, but none of them approached professional level of play, much less world-class play. A strategy other than brute force was required to get there.
---
AlphaGo's loss in game four is interesting, by the way; I'm not sure how you would fix it in their model. It appears that it lost because it failed to predict the (creative, unusual) wedge move in the neural nets, did not spend enough time evaluating it in MCTS, didn't give a strong response to it (which was available through MCTS if it had looked), and didn't understand afterwards that it was now dramatically losing the game and played some very strange moves.
Actually AlphaGo found that move and it rated it at a 1:10000 chance of being played. Possibly because of that it didn't evaluate the possible continuations. However it also took almost another ten moves before the value network decided that it was losing the game.
I found this article[1], (linked from the original) that also mentioned this:
"In recent years, many programmers have tried to get around this problem with Monte Carlo simulation, a statistical means of finding the best first move from a vast database of the games that might begin from a given position. That method is also used a bit in AlphaGo, together with the tree-generating methods of yore. But the key improvement is AlphaGo’s use of deep neural networks to recognize patterns."
Monte Carlo for Go was itself a major step forward; it pushed the best bots from about 1-5k to about 2d (amateur). It's not brute force; it's statistics!
Hassabis declared somewhere that AlphaGo evaluates 10,000 positions per second. That's not brute force for today standards. It has two neutral networks that decides where to look for moves, how good a position is and then it does a Monte Carlo search there. I believe the positions in the MC search tree are not counted as part of those 10k positions.
