Life In 19x19

In fact, it is a weakness of many human players that their local tactical reading, which should be functionally complete, is incomplete. With greater effort, they can make it complete.

RobertJasiek wrote:

yoyoma wrote:it's very unlikely that simply thinking harder on move 79 would have helped, the position was too unclear too deep into the tree for AlphaGo to read.

AlphaGo does not read and this is the reason why it lost at this moment: it did not verify by reading. The AlphaGo team is too focused on neural nets.

They kind of sort of did OK, some people would say. Must try harder, naturally.

It is true that MCTS is not the same thing as reading, at least in the same sense as we humans typically read.
However, AlphaGo DOES read similarly to how we do it, going down a tree of moves and then evaluating whether the resulting position is favorable/unfavorable. It does that using the value network, while humans rely on judgement.

The Nature paper explained that the results of the MCTS and of the value network were combined 50%/50% in order to assess the quality of a move. I think it is likely that some part of this changed in the past 6 months, but as far as we know, AlphaGo DOES read in a very similar way that humans do.

RobertJasiek wrote:In fact, it is a weakness of many human players that their local tactical reading, which should be functionally complete, is incomplete. With greater effort, they can make it complete.

Arguably no one has made it complete by your assumptions.

oren wrote:Arguably no one has made it complete by your assumptions.

Understatement of the year...(emphasis mine)

I didn't see this posted yet, but Haylee explains the move pretty well in her review starting at about 1:09

https://youtu.be/482sitMhspo?t=1h9m