Sure and any of us could jump on a motorcycle and fly past Usain Bolt in the 100m, but that kind of misses the whole point of the competition.
> Give the world-champions each a copy of AlphaGo on equal computers. Have them play Go against each other WITH computer assistance.
Won't the winner be the one who just takes AlphaGo's recommended move every time without changing anything?
The impression I got from Lee Se-dol in the Netflix documentary was that he had a lot of his identity tied up in this (particularly being the best Go player in the world). Not a huge surprise given the time and effort required to do it, but there's probably a healthier mental framing (I think Kasparov has a healthier one).
If people give up on learning or doing something every time someone or something else can do it better then there's going to be a lot of disappointment as things continue to move forward.
The point is we don't really know what limits exist beyond our small pinhole of perception. During industrialization I'm sure people said the same thing: "What's the point of hand-made clothing if machines can do it?". Today both hand-made and factory-made clothing have their place in the market.
Throughout history, tools have disrupted humanity and we've adjusted and expanded our horizons. To me this feels like another wave, and If I could bet and could collect winnings centuries in the future, I'd say this isn't even the last one.
Some track and field records set in the '80s still stand, e.g., Marita Koch's 400 m, Jürgen Schult's discus throw, Galina Chistyakova's long jump.
Maybe he's lost motivation, but this retirement is not a surprise.
No. Consider this scheme.
I take my copy of AlphaGo and for a full year, I'll build a database of all opening positions AlphaGo is willing to explore. I'll rank these opening positions from "best" (Black-wins with most consistency) to "worst" (White-wins with most consistency).
I'll put all of this information into a 16TB database on a singular, $400 16TB Hard drive, and load it into my computer during the contest. https://www.newegg.com/p/1Z4-002P-015K6
If you dare to pick AlphaGo's best move, you will lose. Because I already know which moves AlphaGo will take, and I already checked to find all of the positions AlphaGo wants to play (but loses anyway).
The only way you can equalize the field is if you yourself ALSO build a 16TB database to consult and override AlphaGo's instincts during the game. If you see AlphaGo wanting to play "losing position #6234115", you'll tell AlphaGo to "search harder" and find another move instead.
Good luck.
Go does not have openings. It has reasonable choices to make, with an insane branching factor, with few moves making much of a difference by themselves. Therefore your database will only extend a few moves, and all of the positions that it winds up with will still be very close to even. So your database confers very little advantage.
If you have alpha-go play itself a thousand times, it is unlikely that by move 10 you will wind up with the same board position twice.
This is exactly the problem that made Go so hard for computers in the first place. Alpha-beta search is useless within practical limitations of computer hardware.
That said in a different game, such as chess, your strategy would work very well indeed. (Which is why all decent computer programs have an opening book.)
There are 381 opening moves in Go, but really only 96 because of symmetry. 96 (opening moves) x 380 responses x 379 x 378 x 377 == ~2 Trillion positions after 5 ply.
These 2-trillion positions will easily fit in a 16TB hard drive for $400. That's 8-bytes per position, so you probably can get there with more symmetries and some compression applied.
----------
You're thinking too much like a human. There's no Go-openings in the age of Human-Go. But in the age of Cyborg-Go where 16TB hard drives are allowed, we can begin to exhaustively build openings.
We even have a super-human AI that can automatically, and algorithmically, explore this opening book. We can build AWS-instances with V100 Tensor cores to use neural-nets to explore all of these positions now.
-------------
> If you have alpha-go play itself a thousand times, it is unlikely that by move 10 you will wind up with the same board position twice.
Alpha-Go doesn't seem to implement much randomness at all into the moves it plays. The source of randomness is in time-controls (AlphaGo may choose MoveX before 30 seconds of analysis, or MoveY after 30 seconds of analysis), but this is a fairly constrained number of moves.
Play alphaGo by itself a thousand times, at precisely the same time MCTS-controls (say: 1-million nodes visited in the MCTS tree), and it will probably play the same game 1000 times in a row.
This makes AlphaGo extremely prone to opening database "attacks". Which is why I am using opening books as an easy example for how to beat a particular AlphaGo network. At least, until AlphaGo updates its algorithm for more random play.
If the goal is "Beat AlphaGo" in a game, then the opening book construction is far, far simpler. Even with random elements (ex: AlphaGo picks randomly from the top 10 best positions it generates), that is far more constrained than a full 381 x 380 x 379 x ... style opening book.
No, you're thinking too much like someone who understands chess but not Go.
Suppose that we built an opening book with a trillion reasonable courses of action on it. Each one analyzed well. As you have discovered, you will only go a few ply into the game. And all of the positions that you will be directed towards will have only a small edge.
Instead put a tiny fraction of the computing power necessary to build this book into self-training. You will get a better internal model and therefore a significantly stronger computer player. (That is how alpha-go was built in the first place.) This option will produce much better results for far less effort, and again leaves no role for a human to do anything useful.
The fact that a memorized opening book is useless has nothing to do with human vs computer vs cyborg. It has to do with the characteristics of the game. In chess, it is useful to memorize openings and both computers and humans do it. In Go it is a waste of energy, and it is a waste for both computers and humans.
In my opinion, it's really strange to describe this as an expert "beating" AlphaGo, when really it's just a technique for making AlphaGo stronger than it is without a huge pre-calculated cache of moves.
10 moves into a chess game is actually pretty far. The directions the game can go are massively narrowed 10 moves in.
If the OP spent a month learning Go, I am sure that it would make sense to him as well. Work through a series like https://senseis.xmp.net/?LearnToPlayGoSeries while playing Go regularly against a variety of opponents. Before book 3 it should be obvious.
I'm not an expert by any means, but I'm well past "spend one month" learning the game.
If you like, I could have a look at several your lost games and maybe suggest how to improve. Just a 3k at kgs, but still.
This means if you enter a hypothetical "Cyborg" Go competition (computer-assistance allowed), the majority of newbies will simply be playing LeelaZero #1 plays over and over again.
You don't need to build an exhaustive opening book covering all possible moves. You only need to pick say, the top 5 moves LeelaZero ever considers. If you spend ~16-bytes per position and store 1-trillion positions on a 16TB Hard Drive, you'll be able to exhaustively map the top5 moves LeelaZero considers into 17-ply.
From there, you pick the positions that LeelaZero thinks its winning in, but in actuality is losing. You have a map towards 1-trillion positions to choose from, and your opponent (if they only pick from the top5 best moves LeelaZero ever outputs) will walk into your trap.
---------
As long as your opponent picks the top 5-moves from LeelaZero, you'll have the map towards victory. I think you're severely underestimating the abilities of a simple, dumb, opening database.
Do you believe that AlphaZero could continue to improve dramatically with another 6-months of training? Or if it can improve at another 6-months after that? At some point, the network will reach a local maximum, and it will be unable to improve beyond that.
Characterizing AlphaZero's moves through big-data analysis is innately going to become useful as self-training plateaus. Even Google wasn't able to get more than a few months of training in before the plateau.
At which point, it will be more reasonable to characterize the weaknesses of the network and build an opening book. Avoid the positions that the network was unable to learn about. Etc. etc.
Opening books, at a minimum, would grossly improve AlphaZero's play at competitive levels. Anyone with an opening book of AlphaZero's mistakes will be able to push AlphaZero into a mistaken position.
And then AlphaZero was better than AlphaGo after around a day of self-training.
Furthermore you are arguing for an opening book without considering how small an advantage an opening book would be. As I have said repeatedly, an opening book takes a tremendous amount of work to generate, will only go a few ply in, and the positions it directs you to will only be a tiny bit better.
Therefore for the foreseeable future, more training and better algorithms will produce better results than trying to create an opening book. Theoretically this could change. But that day will not be today or this year. I would be astonished if it happened during this decade. I would be surprised if it happened in a lifetime.
Your proposed approach is an excellent one for many games. But not for Go.
The plateau is real. I'm not sure if continuous self-play will lead to continuous progress for all of eternity. The system is clearly slowing down in self-learning.
AlphaZero's plateau is also well documented: https://i.imgur.com/NMNp6Kq.png
--------
I'm not trying to cast doubt upon reinforcement learning / MCTS / Neural Nets in the game of Go. It is clearly the best methodology we got today.
But anyone who has any experience with neural nets knows about the local-maxima problem. ALL neural nets reach a local maxima eventually. Once this point is reached, you have to rely upon other methodologies to improve playing strength.
Assuming Elo-growth for all time using a singular methodology is naive. We will go very, very far with Deep Learning, but are you satisfied with that limit? Other open questions remain: Go is very far away from being a solved game, even with a magical machine that plays 2000+ Elo stronger than humans.
This only works if AlphaZero never retrains on previous games.
At 5 ply, the complexity of the game hasn't started in any meaningful way. In a typical game, that's 4 corner moves, and then one of a: an approach to a corner (kakari), b: an enclosure (shimari), c: a wedge (waruichi) or d: creating a side framework (such as the Chinese fuseki or Sanrensei). There are some odd opening such as tengen, or corner-corner-corner-kakari which typically turns into a sente fight, but 99% of games will fall into the aforementioned pattern. The database you describe is about as useful as a database of amateur games, since most games, including AlphaGo's games, follow just a few basic openings that early, and even amateurs can play these first few moves "correctly".
Even if you get out to 10 ply you're still only getting partway into a single joseki sequence, often leaving three whole corners of the board which haven't even been approached, so this database still isn't very useful.
Incidentally, your numbers are also wrong. Symmetry reduces the first move to 55 possibilities, not 96, and there are 361 points on the board, not 381.
The basis of reinforcement learning algorithms is the exploratory nature of learning due to the initial application of largely random moves.
Only after some time the agent is given confidence into his learned ability and grafually moved into a more deterministic behaviour mode.
This is the exact opposite of your statement. Star Craft players have noted that the fleet of different AlphaStar instances training in ensemble observed very different behaviour due to this property of RL.
Then you wring out a tiny advantage-- a fraction of a stone. It's a small benefit compared to building other parts of understanding of the game.
Human + computer might still beat computer in Go - this was true for a few years in chess, and even now to some extent in correspondence chess - but what you describe isn't really that.
All you have done is establish a computational arms to see whose computing rig wins when you press the 'pick best move' button. You're not playing Go any more, you're playing Database Administrator.
Based on what we've seen in a couple decades of computer-aided chess, no. A good chess player using a top-rated engine to help them can pretty consistently beat the engine by itself.
There are tournaments and even a world championship in computer-aided (correspondence) chess and you don't come close to winning by just taking the program's recommended move every time.
That's only true if AlphaGo never makes a mistake or if AlphaGo will 100% always make the better or equal decision than a human + computer at any given state of the board. I know the former certainly isn't true and I assume the latter isn't true either, but I don't know enough about Go to say for sure.
Even if AlphaGo makes mistakes, and somewhere on the board a better move can be found, you would also need the human to reliably spot it.
Eg: AlphaGo makes a move. Let's say that at least 20% of AlphaGo moves can be bettered. Is this one of them? How can you tell? Most of the time, you'll mistakenly think a move can be improved and end up playing a worse one.
But, let's make AlphaGo even more fallible. Let's say that at least 50% of AlphaGo moves can be bettered. Again, is this one of those? How can you tell? And more to the point, on the times you are wrong, are you more wrong than AlphaGo is with its mistakes? Because even if you imagine you can spot a better move than AlphaGo and pick the actual better move 50% of the time, you also need your mistaken moves to be better than AlphaGo's mistaken moves or you'll still lose.
Worst of all, you can rule out a really good ability to spot AlphaGo's mistakes already. Let's say 99% of AlphaGo's moves have a better option. If you could spot them all, you'd be beating AlphaGo regularly on your own. As no human can now beat AlphaGo, this plainly isn't true.
So it's likely that:
a) No human can reliably pick a better move than AlphaGo and/or b) No human can reliably spot a move from AlphaGo can be improved, and/or c) Human mistakes are worse than AlphaGo mistakes, so even if you could fight it up to parity you'd still lose.
You can use the relative scores to decide when to overrule the AI. Eg, if move A has a 50.1% win chance with 2k branches explored, and B has a 50.2% chance with 1.9k branches explored, I would go with the opinion of an expert human, as AG thinks the moves are essentially equal.
Plus, you only have to outsmart the car AI once to 'win' - e.g. just override one 'drive into the highway barrier' or 'run over that pedestrian' AI mistake.
An accurate metaphor would be a person on a motorcycle competing against another person on a motorcycle.
I really like the way you put it. In isolation, humans today have essentially the same natural mental capacity we had 2000 years ago, but we've become part of a much bigger computer. A human brain 2000 years ago was a very isolated entity. Even in centers of learning in the classical era, the amount of knowledge one could tap into was vanishingly small compared to our capabilities today. Everything and everyone was isolated in both space and time.
A brain today is not just an entity on it's own, it is intricately wired into the common consciousness. Crucially, we have a vast database of knowledge - easily searchable and distilled for maximum learning rate - all available at our fingertips. A modern brain is a neural network linked to billions of other neural networks, and all of them are linked to a shared memory that they can use at will.
Assuming that the main limiting factor is the bandwidth between a human brain and computers (and thus by extension between individual humans), then a direct interface could ostensibly bring about a new revolution (for better or worse).
I feel computers (and the networks between them) are a natural evolution of our hive mind. When we first started taking spoken language and committing it to writing, we created a hive mind that extended beyond a generation. Prior to that point, everything was passed on through observation of other humans (observing behavior, observing speech, etc.). This meant everything had to pass through, and be mediated by, a human brain. Once we committed that to writing, we had a direct line back to the original brain that created the content unencumbered by the mushy grey bits that have consumed and regurgitated it since.
I feel we are still iterating on that. Cataloging our collective minds and building ever low latency systems to navigate those catalogues. Computers are just an extension that improved our ability to catalogue and retrieve the contents of each other's brains in an incredibly low latency way.
We are part of a hive mind. Our industry is actively building the load bearing infrastructure that supports our hive mind.
From my understanding, indigenous Australians and Polynesians had rich oral histories and cultures. Also oral culture may be more adaptable than a musty written document that never changes and must be followed. The human experience of living in an oral culture is naturally being lost, but it doesn't mean it was wrong or bad. I find it fascinating to imagine what life was like for them, how they did things differently.
"doesn't mean it was wrong or bad"
It is a dead-end, though.
While these cultures are interesting from anthropological point of view, I highly doubt you would give up all the benefits of the culture you live in today and join an Polynesean or Australian tribe.
This hive mind over-simplification is ridiculous and to me: disgusting. As if we were one
It’s not about having a queen or mindlessly following. Being a part of a hive mind doesn’t make you a lemming.
Example to demonstrate: assume you know calculus. How much of calculus did you discover yourself? How much of the corpus of math that led up to calculus did you discover yourself? How much of that corpus is your individual contribution vs. how much of it represents the brain power of hundreds of thousands (if not millions) of other humans throughout history exploring that problem domain?
For any given problem domain, humans have documented it in a shared corpus that you can “download” into your brain. You are an individual. You still get to choose what to download and how to interpret it. But you are still sharing in a commons when you do this. That commons is what I’m referring to as a hive mind, it’s a shared consciousness where our collective brain power is building a corpus that no single brain could. It extends far beyond ourselves.
We may not be a hive mind, but we may have one.
Hive mind as an emergent property still implies a common thing for which one lives. Or a queen. A peer that is one's reason to live. The closest thing I can see to a hive mind is the army.
We are all different and it is not because a person one talks with does not express a different opinion does it mean he thinks the same.
Would one say that lions have a hive mind?
I don't understand how that has any bearing on whether human society can form hive minds or hive mind like entities.
> Hive mind as an emergent property still implies a common thing for which one lives. Or a queen. A peer that is one's reason to live.
No that is absolutely not what a hive mind is. A hive mind is a collective consciousness. That's it. It says nothing about the objectives or agency of it's parts.
> We are all different and it is not because a person one talks with does not express a different opinion does it mean he thinks the same.
You seem to be laboring under the apprehension that a hive mind supplants the consciousness of it's parts. That may be the case for certain hive minds, but it is not a necessary feature. A hive mind can just be an emergent property of individual minds that are strongly connected but still retain agency.
Thanks for precising hive mind, I understand it better now.
Hive mind can be seen as uncritical conformity or collective intelligence. Because it can be understood as uncritical conformity I will use another word/concept. Like tribal mind. Because it suggests tribes (plural) that one belongs to or not, status, etc
If we define hive mind as a collective consciousness, then for sure one can argue that such a thing seems to be arising in human society. As a matter of fact in modern parlance it is fine to call a strongly unifying force, set of norms, school of thought or strong social bonds a hive mind. The hive mind can still be constructed of individuals capable of agency and independent action. It does not replace the minds of it's constituents, rather it can be thought of as an emergent property of many individual agents forming a deeply connected collective.
You're a bee, dawg.
There’s your hive
Just consider the fact that everyone thinks they are above average in compassion or intelligent and stuff that like that. Even if you scream in their face that "HUMAN OVERESTIMATE THEIR PLACE IN THE AVERAGE!!" they will still claim that they are in the 51% percentile.
If anything, technologies and networking have made human a lot more stupider in that its harder to climb against all the misinformation.
When computers became able to beat humans, the tradeoff was that computers were better tactically and humans better positionally. Therefore a human computer pair with the ultimate decision up to the human is able to be a better combination of positional and tactical than either alone. This is why Cyborg chess works so well.
But in Go, AlphaGo is simply better at everything. It is better both positionally and tactically. It can't always explain why the move is right, but adding human intuition on top of the computer only detracts from the quality of play.
But generally, humans have accepted computer supremacy in chess pretty well. Everyone realizes that a better chess-playing entity exists out there (in everyone's pocket, in fact). That doesn't make the competition between human players less exciting. It's still a match to prove that you're better than the person sitting across the board from you, and it's still incredibly impressive to see how deeply and accurately the top players can calculate, or how much knowledge they have of the game.
"Could a human beat a computer at chess?"
Like, now. 20 years after Kasparov. Could someone do it? The question feels daring somehow.
Exercise to the reader: prove that at most one color can have a winning strategy.
Since the problem is open, it could be that one side has a winning strategy. It's even theoretically possible there's a winning strategy simple enough for a human to follow. In which case, yes, a human could beat computers with 100% dominance--as long as the human is allowed to choose which color to play.
If you're asking how a human can contrive a scenario to beat Stockfish/Leela, that seems like just a conversation that devolves into what is and isn't too contrived.
Big showcase computer-human matches ended in the mid-2000s. That's about the point when people realized it was hopeless to try to beat computers. The best you could do back then was draw, by playing extremely defensive "anti-computer" chess (positions where the pawns clog up the board, making short-term tactics useless, and making long-term strategy more important). But with modern engines (particularly AlphaZero, which understands positional chess much better than previous engines), even anti-computer chess doesn't work. Now, the top players actually rave about the games played by AlphaZero. Magnus Carlsen has said he wants to play like AlphaZero.
For a more quantitative comparison, these are the Elo scores of the top chess engines: https://www.computerchess.org.uk/ccrl/404/
A difference of 400 Elo points means that the better player should score 0.9 (where 1=win, 0.5=draw, 0=loss). The top computers are about 3600 Elo nowadays, compared to 2870 for the best human (Magnus). It's difficult to directly compare human and computer Elos, but this comparison is based on computer-human games played in the late 1990s to early 2000s. Still, computers are many hundreds of Elo points above the very best human players.
If you start from a winning position, even something as stupid as a random-mover will eventually play a perfect sequence of moves and win, given enough attempts.
However humans are not very good at being random, so this may not apply :)
` It was a massively parallel, RS/6000 SP Thin P2SC-based system with 30 nodes, with each node containing a 120 MHz P2SC microprocessor, enhanced with 480 special purpose VLSI chess chips.`
I misread at first thinking it was grossly 30GHz summed but it's only 3.6GHz (I know, gross maths). So yeah the comparison is apt.. a recent SoC can outperform this old behemoth.
1. https://www.computerchess.org.uk/ccrl/404/cgi/compare_engine...
We've been cyborgs much longer than that! Writing is a very old auxiliary memory mechanism.
See also: Truth of Fact, Truth of Feeling by Ted Chiang - http://archive.is/oYo0l
Two problems with this plan: (1) the best players at human vs. human Go are likely not as good at playing on a human/computer team, and (2) it may be that the world champion human/computer team is the one that defers 100% to the computer, which seems uninteresting.
AlphaZero doesn't have any randomness built into it, does it? Which means I can build preparation against the lines of play AlphaZero would want to go down.
I think you're underestimating the human, and also overestimating Google's laboratory experiment here. No one has really played "Cyborg Go" yet.
The first casualty of "Cyborg Go" will be AlphaGo in its current form: its clearly inadequate for AlphaGo to play deterministic. Random play MUST be incorporated, lest opponent's preparation sends it down the wrong path.
If I know that AlphaGo plays slightly worse on 3-4 opening (or 4-4opening, or maybe even the 10-10-opening), then that's what I'll do. Give me a copy of AlphaGo and I'll be able to find a weakness somewhere.
I'm with the other guy; I don't see how a person could make the AI play any better.
That's not what MCTS means.
MCTS refers to the bandit problem which formulated the search parameters. MCTS always chooses the "most interesting" path to explore. (Where "most interesting" is the path that balances explore-and-exploit hyper-parameters).
AlphaZero improved upon MCTS by deferring to the neural net as the hyper-parameter. But AlphaZero, for a given network on a given board-state, will ALWAYS choose the same position as "most interesting".
Turning AlphaZero from its current deterministic form into a random form would be an easy fix. But its just one example of how AlphaZero really isn't designed for competitive use yet (despite playing the best game of Go of all time). Instead of picking the top #1 move, maybe you randomly pick from the top 3 moves... or some other scheme.
The AI selects a move. What state is the board in now? It doesn't know, because the opponent also selected a move.
MCTS models this with a probability distribution of the states, and samples from this distribution repeatedly to build an estimate of the effectiveness of each move it could make.
But what's the probability of each move made by the opponent? And after the simulation has looked as many moves ahead as it can in the time constraints, how good a position is it in?
These are the same question, really - what's the chance of winning from this board state. In Chess you can use a heuristic algorithm to figure it out. In Go, you can't. But you can use a neural network to learn an approximation that improves as it sees more games complete.
AlphaGo does this. MCTS is a random sampling technique, and the neural net informs its probability distributions, but doesn't make it deterministic.
If given White in chess, LeelaZero plays 1. e4. Each time, every time. Guess what that means?
If you're building an opening chess database vs LeelaZero (or at least, this version of LeelaZero: https://lichess.org/@/LeelaZero-UK), you only have to worry about 1. e4 openings.
Nothing.
Play AlphaGo against itself. Go rarely has draws (requires Triple-Ko, a very, very rare position).
Almost every game you play with AlphaZero vs AlphaZero will result in a winner-and-loser. You will quickly be able to characterize the positions that AlphaZero loses in.
The strongest early moves create the most potential for winning (maximizing potential winning paths, sort of); they do not push the game towards one best end state. They do not have counters. I saw elsewhere you have some understanding of the game (15kyu) so you should be able to demonstrate this to yourself by playing some of AlphaGo’s openings on a board and trying to write deterministic counters to them. You will not be able to push the AI into a situation where it has too few options to avoid loss. You will also find you need to create a book much larger than a few moves to meaningfully predict play and so will exceed the number of states that can be stored (referencing your 16TB comment elsewhere.)
Please actually try this as I think it is a key to improving your skill in addition to understanding the challenges in automating play.
For most of us, that means we'll all have the best verison of LeelaZero to grab from Github and use in our own personal studies. Which should still be super-human in terms of play.
And don't forget that AlphaZero is gonna be getting better over that year too; you're trying to beat a moving target.
It would be nice if we got a decade of centaur Go but I'm not sure we will. While Deep Blue had fairly crude heuristics for guessing which lanes of inquiry were promising AlphaGo's strength is its intuition-like neural networks that it combines with a rigorous tree search. It's worth investigating whether human/computer teams are superior at Go right now but I wouldn't count on the answer being "yes." And I especially I wouldn't count on a full 10 years of combined dominance.
Over in the "Cyborg Chess" community, people have already analyzed LeelaZero vs Stockfish. It turns out that Stockfish is far better at tactics (especially the endgame), while LeelaZero is better at opening positions (aka: Positional play).
There are numerous theories about the proper combination: perhaps using an Opening Book database for the first ~10 moves or so, using LeelaZero for moves ~10 through 30, and then using Stockfish to check for tactics (LeelaZero misses a lot of tactical options in the midgame, so double-check to make sure that LeelaZero doesn't lose a queen or something), as well as finish up the endgame.
--------
Choosing the correct combination of tools, studying these tools and coming up with a more beautiful chess game. That's cyborg chess in a nutshell. LeelaZero and Stockfish are both superhuman in terms of play, but the cyborg can choose to use both tools to play superior compared to just a singular tool.
Anyone purely using "Stockfish only" gets beaten by opening book analysis. The dude over there with 1TB of opening book databases consisting of every losing position that Stockfish tends to play will completely own you. Same thing with LeelaZero, the opening book guy will own any LeelaZero-only player. These engines have weaknesses that can be undermined with good analysis, big-data, and a bit of custom code.
That's the funny thing about these computers: they tend to play the same. So you can build opening book databases to exploit their patterns. You require a cyborg / human to play at the "level above that" to guide Stockfish / LeelaZero away from those traps.
And then program the computer to use that opening book directly. Now what is your cyborg player going to do? After you patch all these easy rules, you will have to discover harder rules, and the computer can discover them faster than you can.
Build and configure the machine better. Quick, tell me, what's better at playing Chess:
* Or is Xeon Platinum 8180 with maximum 8-way SMT memory sharing with a big-ol 1TB shared transposition table the fastest computer?
* Or will it be cheaper to rent AWS-instances with their V100 GPU in the cloud? Or is the latency for the remote-access bad?
--------
The cyborg player has to still build and program the machine to compete. It is all part of the competition. Can you run a transposition table shared between chess engines over RDMA 10Gbit SFP+ Fiber? Or is that too slow?
-----
This isn't hypothetical at all. The winner of the World Computer Chess Championship 2019 was 8 x Intel Xeon Platinum 8168 running Komodo vs 24 x Amazon AWS Intel Xeon E5 running Shredder.
Configuring and building the computer is still an incredibly difficult part of cyborg chess.
It's a stretch to categorize that as a cyborg player. You might as well categorize any chess program as a cyborg player because a human had to program and train it.
The key difference between cyborg/centaur/advanced chess and plain old computer chess is whether there is a human in the loop making the move decisions. My argument is that having a human in the loop will result in a worse player.
When you build an opening book, you need to pick-and-choose which engines will self play. Will you build an opening book vs Stockfish? Komodo? LeelaZero?
If so, how will you generate these LeelaZero games? You'll have to build a computer (or rent one from AWS) to play these LeelaZero games. What are the time-controls of matches?
Self-play at 40-minutes + 15-second increment means that you'll only create a game every hour or so. Spend 30-days building databases at 40-minute + 15-second increment games, and you'll only reach ~720 games of analysis per month of preparation.
Self-play at 1-minutes + 0-second increment results in a game win/loss every 2-minutes (maximum), giving you 21600 generated opening book positions per month of analysis. But these 21600 games are of lower-quality.
----------
Spend 1-month building an anti-Stockfish database, an anti-LeelaZero database, an anti-Komodo database... and you're now 3-months into preparation and there's still Shredder, Johnny, and all other programs that may arrive at the contest.
Its not exactly as easy as you think it is. There's no algorithm that automatically builds the best opening book (or "counter" opening book) against your opponents. Its a human choice for what the computer will spend the next-3 months self-analyzing and self-playing.
Lets think about how to build anti-LeelaZero seriously. Which of these networks do you download as the LeelaZero representative? https://lczero.org/networks/
You don't have the time to build an anti-opening database for all of those networks.
From my perspective, cyborg chess is about playing the best game of chess in all time... only stopping once we have discovered "the perfect game" (aka: proving that White-wins, Black-wins, or a Draw is always possible with perfect play)
The computers currently playing Chess, and Go, are incredibly powerful. But they are far from perfect. People have constantly found weaknesses in chess playing programs over the past 20 years, and as a result, have improved chess programs significantly.
Go has only had 2-years in its "cyborg" state, where we can finally use computers as a methodology of exploring the game state.
Most of the resistance you're getting in this thread seems to be due to the fact that you think people who are interested in the game should also be interested in the other thing, and it just seems that a lot of them aren't (including Lee Sedol).
Playing Go challenges your mind with vast complexity and immediate feedback of winning or losing every game. It's a deeply engaging hobby for people who are susceptible to that kind of thing, and it used to be a meaningful career, with competitions, schools, and professional teachers. All of that changes now that software is vastly better at it than any human can ever be. The rush of competing by the strength of the moves you understand and make, by the unaided strength of your own mind, cannot be compared to picking between different engines to make moves for you based on some heuristics about which engine is better at openings. The era of human Go is simply over, for better or worse.
So there was a window between the 90's and today where a human player could have discarded the traditional measuring sticks of "trade-value" and beaten DeepBlue.
Maybe that's not true today or maybe there's a further theory out there?
Google played Alpha Zero vs Stockfish under their own conditions. I think this was a mistake to stay in the labs by themselves.
I think Google would have benefited from participating in WCCC 2019 for example, where Johnny (1200x core cluster) won the day.
----------
Its a completely different field when you're actually competing against someone else. Google did some impressive things in the lab for sure, but they have NOT actually stepped into the competitive ring yet.
I'm sure AlphaZero is good, and probably would make a good showing at one of these contests. But you're putting the cart before the horse here.
EDIT: Maybe MCTS + Neural Nets truly is the superior way of preparing board game knowledge? If so, the next step is building out the opening-database and to start looking for holes where AlphaZero loses. It wasn't a complete blowout: AlphaZero lost some games to Stockfish. Why did AlphaZero lose in those games? Is there a set of opening moves that will lead AlphaZero down that losing path in a true competitive setting?
---------
EDIT: Case in point: Google did NOT incorporate randomness into AlphaZero's algorithm. It always chose the best move it believed in, this leads it prone to opening database attacks. See how the competitive mindset already messed AlphaZero's careful laboratory preparation up already?
Sure, AlphaZero 2.0 might have programmed randomness added to it, if this were a true competitive environment. But that's how cyborg chess evolves: I point out a weakness in the program, exploit it in a game, and then Google goes back to their labs to make something better next year.
Though I do still agree in thinking there is likely a deeper theory yet uncovered.
I have to disagree: even after the discoveries by Alpha Zero, the best chess players are not able to beat Stockfish. Stockfish is just too good at Brute forcing long tactical advantages. Playing really well positionally doesn't matter a lot if your opponent can look 30 moves into the future.
I'm not sure how big the window was, if it existed, but it seems like it might have. Kasparov and Deep Blue themselves were fairly equally matched, it doesn't seem impossible that Kasparov + AI-aided theories would have a window of advantage over the Deep Blue -> Stockfish evolution.
As of now, we have databases for perfect play when 7 pieces remain on the board (including the two kings). The 8-piece tablebase is computationally possible, but I don't believe a comprehensive release has come out yet. Even the current 7-piece tables are incomplete because situations like lone king vs. six opposing pieces haven't been explicitly calculated due to their obviousness.
Computers cost about $0.01 / GFlop (10^12 operaions); $1e47 for 10^50 operations.
The world economy is $1e14/yr.
That's 1e33 years to check every chess position, using all of Earth resources.
You'd need to find symmetries to collapse the search spaceby a factor of 10^33
Probably for a few years. Very soon the input from the human agents will be indistinguishable from the noise in the AlphaGo algorithms.
I agree, and this could also be rephrased as we're not really human any more since the advent of writing or (especially) printing.
Come on, none of us are really "human" anymore since the advent of writing.
As we have seen in the latest championship blitz chess is the future for chess.
Computers compete in their own championship.
This is why AI is more significant to Go than horses and cars are to running.
It's easy for humans to build machines that enable them to travel faster than they could run, but it has zero impact on running as a sport. The existence of cars doesn't change how runners run. But the existence of AI does change how Go players play.
Now that AI is more powerful than human players in chess, it is no longer possible to reach the top of human performance in chess without AI. That's why this is so significant. It's not that a human can't beat the AI. It's that a human can't beat another human without learning from the AI. You simply can't play chess at the highest levels anymore without playing cyborg chess. This is different from the impact of other machines on other sports.
But I disagree with you on the comparison to how we use the internet to enhance our other abilities, such as programming. Computers help us be better programmers, but they can't replace us. AI completely replaces humans as a Go player. The human is no longer needed at all.
I can see this causing a significantly different psychological response. In effect, a human trying to become the best chess player or best go player they can become is trying to more and more closely emulate a computer that they will never actually reach parity with. Another commenter noted that "A Professional Go player is an explorer of truth in a millenarian board", but this is less and less true as AI improves. All knowledge about the best way to play Go was previously gained by humans. In 20 years, humans will probably be meaningless on driving knowledge of perfect Go playing.
Chess players today rely on playing moves which their opponents don't expect and haven't trained on as much as they rely on outplaying their opponents. If there is a perfect chess game or a perfect go game, humans are not going to help find it anymore.
So AI doesn't make it that humans can't meaningfully compete in games such as chess and go, but it does change how they compete. We no longer drive the pursuit of perfect play as "explorers of truth". We no longer learn primarily from each other, we learn by imitating the more powerful, yet unmatchable, AIs. And we no longer compete primarily on sheer expertise in the game. We have to, in some sense, include trick plays that work by exploiting the humanity of our opponents rather than playing for a platonic ideal of the game and seeing who is better from that perspective.
AI has fundamentally and irreversibly altered these games, and I think it's appropriate in some sense to mourn the loss, regardless of our feelings on the new reality we live in.
Mankind should focus on skills that computers can't yet compete at. Hopefully the point in time will be far off where computers will beat humans at everything, but likewise, once they do, there is no reason for the human race to continue to exist.
Cyborg chess is the future of chess. Period. Chess players use computers to train themselves and explore openings in human-only settings, while programmers / cyborgs play correspondence chess.
Go is not finished as a game. A new tool, MCTS + Neural nets, has been developed to explore the "truth" behind the game at ever increasing rates. Its not about how to beat the tool, the question is how to best utilize the tool to improve self-play and self-learning in the game of Go.
Or alternatively: how to best use the tool to play ever more perfect games of Go.
-----------
Come on, none of us are really "human" anymore since the advent of cell phones. We all use our cyborg-capabilities to search the internet and fact-check ourselves every day. Programmers use stack-overflow to teach themselves programming and remember obscure details (using our cyborg capabilities to tag, search, and sift through information ever faster and faster).
Go is the same thing, except we only learned how to become cyborgs two years ago.
----------
Give the world-champions each a copy of AlphaGo on equal computers. Have them play Go against each other WITH computer assistance. I guarantee you that a more beautiful and perfect game will result. Let us welcome the age of Cyborg Go as we step into the future, we shouldn't be scared of it. We've become cyborgs in many other tasks, and Go is no different.