Human moves tend to fit a narrative and be explainable, although often for very concrete reasons. For example:

"I am sketching out territory while attacking an opponent group."

"I am making my group safe so that I will not have to worry about its life while I accomplish other strategic goals."

"I am making a very strong group so that I can use it to make it harder for my opponent to accomplish anything."

Master, on the other hand, will sometimes seemingly just plop stones down in the middle of the board in a way that is hard to assign a narrative to. One can list a bunch of ways that the stone might come in handy in various futures, but it's too "vague" a move to be played with confidence by a human professional go player.

Of course, human strategies may change in response...

When playing against a much stronger player, it's always very hard to figure out why they play tenuki (make a move somewhere else on the board, apparently uncorrelated to the current fight).

When AI gets strong enough (and it seems like it has already), it will just tenuki everyone all the time, while winning. Sounds like exactly what's happening already. It's past the event horizon for human understanding.

> It's past the event horizon for human understanding.

AlphaGo creators could "rewind" the whole program state to that move and inspect the tree search probabilities according to the board states it looks through to find a list of board states that generate a cumulative highest probability of wining by doing a move in that exact odd spot.

My guess would be that while humans tend to put stones with a single, double, or sometimes triple "reason", or in AI words, "high probability of local effectiveness in upcoming several turns", AlphaGo, with his ability to see further, can see past the local effectiveness into more global effectiveness and higher probability of winning further down the road.

In other words, those tenuki moves might actually be past the event horizon of human understanding, but only if inspected by looking at them and thinking ourselves. If we use AlphaGo itself, it should be possible to find out the reason for every single tenuki it will ever do.

> past the event horizon for human understanding

I think this phrase is going to pop up more and more frequently.

This thread is literally the only Google search result for this phrase (for me)...

I searched for it in incognito mode and saw the same. Pretty interesting; it's such a nice, catchy phrase.

"Outside the light cone" is more on point -- it's far enough away or moving fast enough that we'll never be able catch up.

To be fair, humans haven't had a lot of time to analyze AlphaGo's play. It's possible that we could understand these moves after some study.

"past event horizon for human understanding" I think you are too pessimistic and that "centaur Go" will evolve in the same way that "centaur chess" is, where a player (or team of players) with access to the software can outperform the stand-alone software. I guess my vote is for "intelligence augmentation" over AI, if only because opaque models/algorithms cannot benefit from human creativity.

Centaur chess players do not outperform the best chess engines. That is based on a misrepresentation of one of the first centaur chess tournaments. Human modified moves only reduce the strength of the best chess engines now. The benefit that was reported was one of tactics one centaur chess competitor had over other centaur chess competitors, not centaur chess vs stand alone chess engine.

It is a misunderstanding that has been breathlessly repeated by futurists and transhumanists for years.

And it's such a strange claim, you'd think it would be obvious that it makes no sense. Nobody would ever try to claim "I think that a team of a top pro and a mediocre amateur would be stronger than the top pro alone", but that's basically what they're doing with centaur chess.

Why do startups staffed by amateurs sometimes outperform teams of experts at established firms? Creativity and new perspectives can yield surprising results.

> Human moves tend to fit a narrative and be explainable, although often for very concrete reasons.

I'm not a Go player, but could it not be that if the AI played its moved in a more human order you could see what was going on and assign a narrative to the moves, but the AI can see the order of the moves doesn't matter sometimes so it seems to play more vaguely/randomly to observers? For example, say the AI played a set of moves at the top of the board that you could give an attacking narrative to and then it plays a set of moves at the bottom of the board that you could give an defensive narrative to. If you intermixed the move order for both narratives you, it seems like the AI is playing in a nonhuman way when really its moves have a narrative but humans are too fixated on the move ordering.

More likely, playing it in order commits to a certain approach too early, and makes that approach predictable. Starting from the middle leaves other options open.

Given that it trains against a copy of itself, with equivalent predictive powers, it makes sense that it would pick moves that have lots of branching possibilities because that would increase its effectiveness against itself.

I've heard a similar approach described in military strategy at all levels: rather than looking for a single dominant tactic, you try with each "move" to create so many potentially-viable future positionings at once that your opponent cannot predict you in order to effectively concentrate their effort.

It'd be very scary to watch a "sibling" to AlphaGo play a 4X game.

DeepMind has set their sights on StarCraft II next: https://deepmind.com/blog/deepmind-and-blizzard-release-star...

Directly thought of StarCraft while reading this comment: never revealing your strategy and constantly attacking while being defensing and investing in economy to keep your opponent on the defensive. Like another commenter said, StarCraft 2 is DeepMind's next bet.

I am not too familiar with StarCraft but from what I can gather, it relies deeply on micromanagement. It seems to me that any half competent AI strategy wise would destroy any human player just with raw clicks per seconds during fights.

That's true, with StarCraft you have to basically limit the actions per minute the AI can perform to human levels, otherwise it can effortlessly out-control a human player. The game was only balanced to be fair for players with human capabilities, if you take that assumption away the unit and race balance goes totally off the rails.

There is the question of what your AI takes as its input though. If you feed it the rendered frames of the games there are a lot of open challenges in comparison to feeding it a friendlier representation of the game state.

There is also the question of teaching the AI a strategy vs only giving it the rules of the game and letting it develop strategies by itself.

In comparison to Chess or Go, StarCraft has a lot more game elements to understand and conceptualize. StarCraft is played with incomplete information, because of the fog of war - so for an AI to play it properly it has to have persistent knowledge of what it has seen of the game state and be able to draw conclusions from that about the probable overall state of the game.

As far as I'm aware they force the AI to only interact at a certain frequency, limiting this kind of advantage

You call it narrative when you understand what's going on.

You call it chaos when it's far above your head.

Good insight, I think there are implications for the Cynefin framework model that offers four possible state spaces: simple, complicated, complex, and chaotic. This classification may have as much to do with our state of information (or depth of understanding) as an objectively "chaotic state."

Made me think of this: [https://www.youtube.com/watch?v=fU7hZ3smj0g&t=4m22s](https:/...

This is usually called the metagame and evolves over time. My guess is that due to these unseen tactics the metagame will change and human playstyle will become more AI-like and adapt to the changing circumstances. The real question is whether people would be able to keep pace with rapidly changing AI playstyles?

Depends on if people can figure out how to evaluate the more random seeming 'AI style' moves. With the complexity of Go it's possible to only way to really tackle the complexity is to wrap groups of moves together in a local narrative.

> The real question is whether people would be able to keep pace with rapidly changing AI playstyles?

If we're far enough from "perfect play", then the answer is no.

AlphaGo often plays surprising moves. Typically high-level commentary describes the horizon of "surprising" here as playing wider, faster, and more influence-oriented than current professionals think is appropriate. AlphaGo is repeatedly praised for identifying moves which seem too unconcerned with the opponents threats and instead take more power on the board in a leisurely fashion. The first big surprising move it played was a "4th line shoulder hit" which was thought to give too much territory to the opponent (a "3rd line shoulder hit" is considered a very fine and popular move). AlphaGo showed that at the right time even the 4th line variant is good since is gave AlphaGo large influence at the right time and position.

A lot of go comes down to timing. AlphaGo doesn't seem to play moves which are inhuman in that they just make zero sense as much as plays moves which are more daring than humans would like to try. Then, infallibly, it turns out that AlphaGo's daring move had the mark of being amazingly well-timed.