Heh, I watched that game :3
However, KataGo did give a solution to this thread, since it’s also trained at playing with variable komi (is it, or was that some other AI?). Giving the pro a huge reverse komi will show how much the margin currently is.
2 Likes
Handicap stones, reverse komi, winning percentage – these are three legitimate and different ways of measuring strength difference.
Extra starting stones shouldn’t be considered as just a sloppy form of reverse komi – actually they are a test of how well Black can attack: a stone on the side hoshi is not much good if you don’t kick!
Perfect play may be weaker against handicap stones than we think.
Against 5 stones for example it will see that winning is impossible(against other perfect player), and will start to surround less big territory instead. Later top human opponent will make mistakes and perfect player will start to surround bigger territory - but fuseki is already not optimal - and it will lose.
So I got and idea: maybe its really possible to create bot that plays better than “perfect play” (against humans with handicap stones).
Perfect player starts with smaller fuseki because surrounding bigger territory would be impossible in 100% cases.
Artificial Bot will start with big fuseki anyway - and this may compensate strength difference. It will try to win even if probability of it is 10%.
For example: Perfect player will win in 1% cases against 5 stones(when top human has bad day). Artificial Bot will win in 50% cases against 5 stones and in 10% cases against 6.
Artificial Bot will use trick moves - it has access to human games database.
I already talked about it in this topic. What I’m saying now - that its probably will be possible in some years later. If human games database is not enough, then it may use data from neuro implant like Neuralink - some more years later. And if that’s not enough - then it may scan brain of opponent and make a simulation - even more years later.
Bot that plays better than perfect play against handicap seems totally possible. But when it happens, we will not know about it - since perfect play on 19x19 is probably not possible in our galaxy.
2 Likes
This is probably feasible, but I would consider such a bot to be “cheating” and as such, exclude it from being used for human rank measurements.
2 Likes
Hmmm. I think performing a remote brain scan during the game could be considered against the rules, but a true AI could make use of body language and other tells gained through a normal visual perception that humans also have access to.
1 Like
In summary: Help, the AIs have become strong beyond our understanding! We need to link them to a human brain to get a genuine source of greed and overplay. 
2 Likes
So then playing go against an AI would have some aspects of playing poker, ability of keeping a poker face would come into play. 
And what if it is an online game where the human player is invisible?
One interesting difference between chess and Go AIs is that a chess engine can rely on an endgame tablebase, since all endgames with (eight?) chessmen or less on the board have been strongly solved.
Thus a chess engine knows that if it can reach a winning tablebase position, it’ll win the game, and it can happily initiate mass exchanges in the midgame to arrive at that goal.
There’s no equivalent concept in Go, since every endgame is different – it’s impossible to construct a global tablebase, so Go bots must consider the endgame much more heavily.
1 Like
Correction: currently, the largest complete chess tablebase is seven-man, but the size of the eight-man tablebase has been discovered.
The nomenclature is that a seven-man tablebase has a maximum of five pieces / pawns in entirety on the board, in addition to the kings.
Note that a two-man tablebase is therefore only bare-king positions, which are obviously drawn.
I found this table on Wikipedia.
This second table shows the growth (calculations at three places, results at two).
Unless I’ve calculated wrongly, it seems that the more pieces added, the less each additional piece adds to the complexity of the position.
| Piece add. | Pos. inc. | Estimation | Δ |
|---|---|---|---|
| 3 --> 4 | 340 x | 125M / 368K | |
| 4 --> 5 | 210 x | 25.9B / 125M | 1.62 |
| 5 --> 6 | 150 x | 3.79T / 25.9B | 1.40 |
| 6 --> 7 | 110 x | 424T / 3.79T | 1.36 |
| 7 --> 8 | 90 x | 38.2Q / 424T | 1.22 |
In comparison, see Number of Possible Go Games at Sensei's Library
That page claims that there are about 10^38 legal Go positions on the 9x9 board. I think that would be approximately 2.5 x 10^15, or 2,500,000,000,000,000, or 2.5 pentillion times more positions than the eight-man chess tablebase.
3 Likes
Until AI is strong enough to take over the world, I don’t care. I’d like to forestall that possibility, but doing so is probably futile.
After AI is strong enough to take over the world, you probably won’t care either. Total annihilation is swift!
But why would an AI care about annihilating us? I mean, it’s a likely side-effect of whatever else they’ll be doing, like we’re annihilating ecosystems. However, we don’t annihilate ecosystems because we want to annihilate them (we do it because our collective human brain is terrible at processing the long-term harm we’re doing to ourselves).
Any AI strong enough to take over the world has no real threat to fear from humans. They might as well ignore us.
1 Like
As you said, it doesn’t have to fear us to destroy us.
My hope is that instead of paperclips, bots choose to deplete earth’s resources in search of the perfect game of Go. 
6 Likes
There’s a reasonable chance that we can stay alive on the periphery… Let’s hope!
2 Likes
See the classic SF novel about living on the periphery: The Genocides by Thomas Disch. Warning: as my best friend has said, this is a novel you can never forget, but wish you could.
1 Like
I’m not so worried about the Hollywood scenarios of killer AI and robots violently taking over the world. I think the more worrying impact of AI on society is in the economic consequences. This is far more gradual and insidious, rather than an obvious, abrupt transition.
The AI revolution has already begun and has long been underway with how more and more human labor is replaced with automation. This should inherently be a good thing, since it frees up human labor to do something else productive, or at least should give us more leisure time.
The problem is not with automation, but rather how society and our economies adapt to changes in the requirements for human labor. The problem is the inequality in how society may benefit from these changes.
Stephen Hawking shared his thoughts on how our concerns should really be focused on the problems of capitalism rather than robots:
I’m rather late to the question-asking party, but I’ll ask anyway and hope. Have you thought about the possibility of technological unemployment, where we develop automated processes that ultimately cause large unemployment by performing jobs faster and/or cheaper than people can perform them? Some compare this thought to the thoughts of the Luddites, whose revolt was caused in part by perceived technological unemployment over 100 years ago. In particular, do you foresee a world where people work less because so much work is automated? Do you think people will always either find work or manufacture more work to be done? Thank you for your time and your contributions. I’ve found research to be a largely social endeavor, and you’ve been an inspiration to so many.
Answer:
If machines produce everything we need, the outcome will depend on how things are distributed. Everyone can enjoy a life of luxurious leisure if the machine-produced wealth is shared, or most people can end up miserably poor if the machine-owners successfully lobby against wealth redistribution. So far, the trend seems to be toward the second option, with technology driving ever-increasing inequality.
— Stephen Hawking, from a reddit AMA
6 Likes
Preach, brother!
E. M. Forster addressed one dark vision of future leisure in his classic story, “The Machine Stops” (amazingly published in 1909 IIRC).
4 Likes
This is turning into a nice forum game: propose the AI apocalypse and Conrad will give us the reference 
7 Likes