European politics

This is for discussions about news, politics, sports, other games, culture, philosophy etc.
User avatar
Nauru Dolan
Ninja
Posts: 13064
Joined: Sep 17, 2015

Re: European politics

Post by Dolan »

How do you mathematically solve the problem of deciding to write a book and what the book is about.
User avatar
Netherlands Goodspeed
Retired Contributor
Posts: 13002
Joined: Feb 27, 2015

Re: European politics

Post by Goodspeed »

You estimate the return on investment using whatever reward system programmed into you, taking opportunity cost into account. If it's a worthwhile investment you do it.

You decide on the subject based on what will sell best or what you would enjoy writing about or whatever else your reward system tells you to value.

Needless to say, if AI ever get that far, programming the right reward systems is complex and vital. "Get the most possible points in this game" is of course relatively simple and morally uncontroversial.
User avatar
Kiribati princeofcarthage
Retired Contributor
Posts: 8861
Joined: Aug 28, 2015
Location: Milky Way!

Re: European politics

Post by princeofcarthage »

Dolan wrote:How do you mathematically solve the problem of deciding to write a book and what the book is about.
Personality and experience.
Fine line to something great is a strange change.
User avatar
Nauru Dolan
Ninja
Posts: 13064
Joined: Sep 17, 2015

Re: European politics

Post by Dolan »

Goodspeed wrote:You estimate the return on investment using whatever reward system programmed into you [...].
You decide on the subject based on [...] whatever else your reward system tells you to value.
This is the part that's not very clear to me: how is any value programmed into your reward system.
Can't people also change their values in time and if they do, are they actually overriding their value programming?
User avatar
Nauru Dolan
Ninja
Posts: 13064
Joined: Sep 17, 2015

Re: European politics

Post by Dolan »

Also how did people make such decisions before maths was invented/discovered (depending on what's your position on this question).
Let's imagine a situation in ancient Sumer, before numbers were invented for the purpose of keeping track of property and making contracts.
How did they reason about their decisions without having any concept of numbers or maths. How did peasants who were illiterate until just a century ago reason without even knowing how to read or write.
User avatar
Kiribati princeofcarthage
Retired Contributor
Posts: 8861
Joined: Aug 28, 2015
Location: Milky Way!

Re: European politics

Post by princeofcarthage »

My answer answers all your questions dolan
Fine line to something great is a strange change.
User avatar
Netherlands Goodspeed
Retired Contributor
Posts: 13002
Joined: Feb 27, 2015

Re: European politics

Post by Goodspeed »

Dolan wrote:
Goodspeed wrote:You estimate the return on investment using whatever reward system programmed into you [...].
You decide on the subject based on [...] whatever else your reward system tells you to value.
This is the part that's not very clear to me: how is any value programmed into your reward system.
Can't people also change their values in time and if they do, are they actually overriding their value programming?
For humans our reward system is built from a combination of genes and what our peers teach us we should strive for in life. But importantly, the reason our reward system depends on our peers is that we are programmed to want their approval. Program a computer that way, along with the basic needs we have, and it will behave similarly
User avatar
Nauru Dolan
Ninja
Posts: 13064
Joined: Sep 17, 2015

Re: European politics

Post by Dolan »

princeofcarthage wrote:My answer answers all your questions dolan
I was specifically addressing this proposition that the way we make decisions is the result of some kind of mental maths. And that this fact would make human cognition translatable to an automaton that works based on maths and electronic logical gates.
Your answer that personality and experience determine a decision to write a book doesn't seem to address this specific point about how human decisions are determined by mental maths and how they could be automated, translated onto an electronic substrate.
User avatar
Netherlands Goodspeed
Retired Contributor
Posts: 13002
Joined: Feb 27, 2015

Re: European politics

Post by Goodspeed »

Dolan wrote:Also how did people make such decisions before maths was invented/discovered (depending on what's your position on this question).
Let's imagine a situation in ancient Sumer, before numbers were invented for the purpose of keeping track of property and making contracts.
How did they reason about their decisions without having any concept of numbers or maths. How did peasants who were illiterate until just a century ago reason without even knowing how to read or write.
How did AlphaGo come up with completely novel (and superior) ways to play Go?
User avatar
United States of America occamslightsaber
Retired Contributor
Posts: 1326
Joined: May 31, 2019
ESO: L1BERTYPR1ME

Re: European politics

Post by occamslightsaber »

Dolan wrote:Also how did people make such decisions before maths was invented/discovered (depending on what's your position on this question).
Let's imagine a situation in ancient Sumer, before numbers were invented for the purpose of keeping track of property and making contracts.
How did they reason about their decisions without having any concept of numbers or maths. How did peasants who were illiterate until just a century ago reason without even knowing how to read or write.
Image
The scientific term for China creating free units is Mitoe-sis.

I intend all my puns.
User avatar
Nauru Dolan
Ninja
Posts: 13064
Joined: Sep 17, 2015

Re: European politics

Post by Dolan »

Goodspeed wrote:For humans our reward system is built from a combination of genes and what our peers teach us we should strive for in life. But importantly, the reason our reward system depends on our peers is that we are programmed to want their approval. Program a computer that way, along with the basic needs we have, and it will behave similarly
Ok. I'm not sure if that's correct, but it's an answer.
How does this work, let's take a concrete example. Let's say there's a group of 10 people, in which 1 decides to become a musician and another a doctor. If individual decisions are determined by peer pressure (and some genetic givens), how could people from the same group make different decisions?

I didn't take the easiest example with a family because often kids decide to do something completely different from what their parents want and then you could argue they were influenced by their peers, so I jumped directly to the peers example to save time.
User avatar
Nauru Dolan
Ninja
Posts: 13064
Joined: Sep 17, 2015

Re: European politics

Post by Dolan »

Goodspeed wrote:
Dolan wrote:Also how did people make such decisions before maths was invented/discovered (depending on what's your position on this question).
Let's imagine a situation in ancient Sumer, before numbers were invented for the purpose of keeping track of property and making contracts.
How did they reason about their decisions without having any concept of numbers or maths. How did peasants who were illiterate until just a century ago reason without even knowing how to read or write.
How did AlphaGo come up with completely novel (and superior) ways to play Go?
Because the task is so easily reducible to movements on limited paths on a table, this makes it completely solvable through maths alone. There's a limited gamut of possible moves and identifying the optimal one is just a question of simulating each move and finding the one that has better probabilities of leading to victory. You don't need a thinking entity to solve this, just an automaton that is capable of computations.

That doesn't mean that if you play Go as a human you're not using your intelligence to get ahead of an opponent, but the way humans approach the task is different. Humans don't just do an exhaustive evaluation of all possible moves (and moves implicated by each move, etc) before committing to a move, they just make a mental estimation of what the best move is. And that's also because humans can't hold in their memory more than just a few items at a time, while computers only have a hardware limit (which is huge). So, in tackling such a task, computers have two advantages over humans: massive parallel processing which allows them to branch out all possible moves without incurring any significant cost in terms of speed of decision-making (compared to humans) and no practical limit to how many alternative moves they can hold in memory simultaneously, ready to be used, if the opportunity to gain an advantage comes (only hardware can set a limit, but it will still be billions of orders of magnitude larger than the few items that humans can juggle with in their memory at any given point in time).
User avatar
European Union scarm
Howdah
Posts: 1439
Joined: Dec 7, 2018
ESO: Malebranche

Re: European politics

Post by scarm »

Goodspeed wrote:
scarm wrote:Are Go and Chess actually "very complex tasks" though? They are mathematically solvable games in theory, humans just lack the computing power to do so.
Everything is mathematically solvable in theory. Go is plenty complex as a proof of concept for machine learning.
No not everything is mathematically solvable. Human interaction for instance. And considering that AI should ideally be able to solve human problems, that seems like a major issue. Assigning random reward values to strategies or choices available does lead to highly suboptimal results in reality. This kinda is tied to the discussion Jerome and Dolan had the other day, and the major problem social sciences face: how do you operationalize social phenomena, considering their incredible complexity, the amounts of variables that play a role and so on.


If it was true that human behavior could be accurately mapped by assigning payout values, game theory would essentially be all social science ever needs.

To put it bluntly: it is very easy to assign values to chess pieces and say "Taking this bishop is good because i am then +3 Material (ignoring positional disadvantages and such)". It is not so easy to mathematically determine the best option in say a hostage taking scenario. Do i provoke the hostage taker? Do i play for time? Do i accept his demands? Do i order storming the situaion? This even applies to simple shit like going to the supermarket, cheating in exams, flirting and whatever other situation you can come up. Same with writing a book as dolan proposed.

Not to mention that the amount of variables in chess and go is very low in principle, and there is symmetrical odds and perfect information.
User avatar
Nauru Dolan
Ninja
Posts: 13064
Joined: Sep 17, 2015

Re: European politics

Post by Dolan »

occamslightsaber wrote:
Dolan wrote:Also how did people make such decisions before maths was invented/discovered (depending on what's your position on this question).
Let's imagine a situation in ancient Sumer, before numbers were invented for the purpose of keeping track of property and making contracts.
How did they reason about their decisions without having any concept of numbers or maths. How did peasants who were illiterate until just a century ago reason without even knowing how to read or write.
Image
Image
User avatar
Nauru Dolan
Ninja
Posts: 13064
Joined: Sep 17, 2015

Re: European politics

Post by Dolan »

@scarm There's an even more striking example: what about composing some music. How could an AI/algorithm compose something that can convey a feeling, without actually having a feeling?
Every act of cognition that a human does involves the lower areas of the brain that manage emotions. Literally every human decision has an affective substrate in it, otherwise nothing could be relevant to you, in terms of motivation.
You wouldn't feel any impulse to act in any way without those underlying animal drives that got shaped into a much more fine-grained gamut of emotions, motivations, temperamental traits that eventually build into a personality.
User avatar
Kiribati princeofcarthage
Retired Contributor
Posts: 8861
Joined: Aug 28, 2015
Location: Milky Way!

Re: European politics

Post by princeofcarthage »

Goodspeed wrote:
Dolan wrote:Also how did people make such decisions before maths was invented/discovered (depending on what's your position on this question).
Let's imagine a situation in ancient Sumer, before numbers were invented for the purpose of keeping track of property and making contracts.
How did they reason about their decisions without having any concept of numbers or maths. How did peasants who were illiterate until just a century ago reason without even knowing how to read or write.
How did AlphaGo come up with completely novel (and superior) ways to play Go?
Cuz it has simply more computational power but still it is predefined by parameters. It is nearly 100% assurance that with time humans would have came up with same techniques.
Fine line to something great is a strange change.
User avatar
No Flag fightinfrenchman
Ninja
Donator 04
Posts: 23505
Joined: Oct 17, 2015
Location: Pennsylvania

Re: European politics

Post by fightinfrenchman »

Dolan wrote:@scarm There's an even more striking example: what about composing some music. How could an AI/algorithm compose something that can convey a feeling, without actually having a feeling?
What about drawing a picture? Sonny did this in I, Robot
Dromedary Scone Mix is not Alone Mix
Image
User avatar
Netherlands Goodspeed
Retired Contributor
Posts: 13002
Joined: Feb 27, 2015

Re: European politics

Post by Goodspeed »

Also @scarm
Dolan wrote:
Goodspeed wrote:
Dolan wrote:Also how did people make such decisions before maths was invented/discovered (depending on what's your position on this question).
Let's imagine a situation in ancient Sumer, before numbers were invented for the purpose of keeping track of property and making contracts.
How did they reason about their decisions without having any concept of numbers or maths. How did peasants who were illiterate until just a century ago reason without even knowing how to read or write.
How did AlphaGo come up with completely novel (and superior) ways to play Go?
Because the task is so easily reducible to movements on limited paths on a table, this makes it completely solvable through maths alone. There's a limited gamut of possible moves and identifying the optimal one is just a question of simulating each move and finding the one that has better probabilities of leading to victory. You don't need a thinking entity to solve this, just an automaton that is capable of computations.

That doesn't mean that if you play Go as a human you're not using your intelligence to get ahead of an opponent, but the way humans approach the task is different. Humans don't just do an exhaustive evaluation of all possible moves (and moves implicated by each move, etc) before committing to a move, they just make a mental estimation of what the best move is. And that's also because humans can't hold in their memory more than just a few items at a time, while computers only have a hardware limit (which is huge). So, in tackling such a task, computers have two advantages over humans: massive parallel processing which allows them to branch out all possible moves without incurring any significant cost in terms of speed of decision-making (compared to humans) and no practical limit to how many alternative moves they can hold in memory simultaneously, ready to be used, if the opportunity to gain an advantage comes (only hardware can set a limit, but it will still be billions of orders of magnitude larger than the few items that humans can juggle with in their memory at any given point in time).
You don't seem to understand how neural networks like AlphaGo work.
Because the task is so easily reducible to movements on limited paths on a table, this makes it completely solvable through maths alone.
Only in theory. And this is why I said that everything can, in theory, be solved by maths alone. Much like with Go, though, it's just not realistic or even worth trying. Go, like many other tasks humans engage in, is too complex for any finite number of supercomputers to "solve" this way. This is why I told @scarm that it's plenty complex for a proof of concept, and it's why Deepmind chose Go when they started the project. They wanted something that wasn't possible to brute force precisely because it would be a proof of concept.

Neural networks actually don't go through the entire move tree and exhaust every possibility until they've found the correct move. Yes this is how Chess "AI" (used to) work, because in Chess it's possible, as long as your search is efficient enough, to brute force your way to near perfect play. But that's not how AlphaGo works because it's not realistic in Go, no matter how efficient your search. Timestamped, shows the difference between the two games in the amount of possible game states.
Something to note is that before neural networks, Go programs were about 1000 ELO worse than the best professionals. They really weren't anywhere close to competent, due to the above.

When a Go player looks at a position, while there may be 300 possible moves, there are really less than 10 that are worth exploring. And how do you describe which are worth exploring? You can't. Developers have tried and failed miserably. There's too much "it just feels right". Intuition. Anyway you take those 10 candidates, explore them by again finding moves your opponent would likely respond with (using intuition), explore those, etc, until you land on a move that's most promising. Anyone who has played any turn-based strategy game before knows this process.

Deepmind's concept was novel in that its program had a sort of "intuition layer" (in technical terms: policy network), which develops, through machine learning, what can reasonably be described as intuition about which moves are worth exploring and which aren't. This layer of the program, like human intuition, comes up with promising candidates and is then combined with an "old school" search algorithm to evaluate them. The policy network is even developed the same way as human intuition: by playing lots and lots of games and learning from them. It forms connections based on previous results, and uses the memory thereof to find promising moves in future games.

By developing intuition through pure self-play, AG was able to discover ways to play Go that were unknown to humans or considered inferior, caught professional players by surprise, and have since found their way into the meta. Players who refused to adapt to the "AI way" can no longer compete on a pro level. Notably, AG was able to find what can only be described as "creative" new moves, simply by playing against itself. Not by "solving" the game through exhaustive analysis of each possible game state, which is impossible.

I recommend the full doc: https://www.youtube.com/watch?v=WXuK6gekU1Y
User avatar
European Union scarm
Howdah
Posts: 1439
Joined: Dec 7, 2018
ESO: Malebranche

Re: European politics

Post by scarm »

Just to clarify, my argument doesn't rest on AlphaZero actually bruteforcing chess (i probably phrased it poorly). I know it doesn't do this, though i don't know enough about machine learning to really understand how it works. What i am trying to say is that while this on its own might be impressive, i just don't think these games are actual good proof of concept of usability in real life circumstances, because they in the end are relatively simple board games, with a very low amount of variables involved, a closed system and so on. The fact that both are in theory brute-forceable, which social situations aren't, served me to illustrate this. In the end i might be wrong here, because having a clear-cut argument necessitates understanding the process fully.

Another argument in this direction is that you brought up that AI needs a lot of iterations to reach human levels of competency (idk, just refering to what you said). In many potential applications giving it this amount of training does not necessarily seem possible, because you might lack the necessary databases, while you can just let it play digital chess or go forever.
User avatar
Netherlands Goodspeed
Retired Contributor
Posts: 13002
Joined: Feb 27, 2015

Re: European politics

Post by Goodspeed »

I'm only disagreeing with the statement that AI could never get there, and am trying to explain why machine learning is considered a big step forward in the field.

What AG was proof of concept of is that computers can mimic human intuition through machine learning. Of course it's not proof that we may one day have general purpose AI. I never claimed that.

I specifically responded to your point about mathematical solvability because while Go is indeed solvable in theory, in this context it is practically unsolvable and its theoretical solvability in no way helped Deepmind. Making that point, I thought you were saying AG is just another brute force algorithm and were therefore misunderstanding how it works. For all intents and purposes, Go is as solvable as human interaction.

And yeah, machine learning will need to get a whole lot more efficient to be applied generally. We are far from that point.
User avatar
Nauru Dolan
Ninja
Posts: 13064
Joined: Sep 17, 2015

Re: European politics

Post by Dolan »

Goodspeed wrote:Only in theory. And this is why I said that everything can, in theory, be solved by maths alone. Much like with Go, though, it's just not realistic or even worth trying. Go, like many other tasks humans engage in, is too complex for any finite number of supercomputers to "solve" this way. This is why I told @scarm that it's plenty complex for a proof of concept, and it's why Deepmind chose Go when they started the project. They wanted something that wasn't possible to brute force precisely because it would be a proof of concept.

Neural networks actually don't go through the entire move tree and exhaust every possibility until they've found the correct move. Yes this is how Chess "AI" (used to) work, because in Chess it's possible, as long as your search is efficient enough, to brute force your way to near perfect play. But that's not how AlphaGo works because it's not realistic in Go, no matter how efficient your search.

When a Go player looks at a position, while there may be 300 possible moves, there are really less than 10 that are worth exploring. And how do you describe which are worth exploring? You can't. Developers have tried and failed miserably. There's too much "it just feels right". Intuition. Anyway you take those 10 candidates, explore them by again finding moves your opponent would likely respond with (using intuition), explore those, etc, until you land on a move that's most promising. Anyone who has played any turn-based strategy game before knows this process.
But that's how the Monte Carlo Tree Search works. It uses playouts, which are literally game outcomes that have been played out. After each outcome is recorded, moves that were most likely to lead to victory become the new starting point for subsequent optimised moves.
https://en.wikipedia.org/wiki/Monte_Car ... _operation

Apparently this approach has been used to solve chess and Go up to a table size of 9x9. Once the table became larger, it was too computationally expensive, since the branching factor was getting too high. So what they did was use a layered approach, which included multiple heuristics to estimate the best moves (including by using knowledge from played out moves). Different strategies were used at different points in the game too, since at the start of the game there are fewer possible moves than midgame, while the more pieces are placed on the table the number of possible moves may actually decrease in time. Hence, using the same computational strategy all throughout the match wouldn't be optimal.

In the end, there's nothing magical about how such a performance was reached, there's no actual "creativity" behind the winning moves chosen by the different iterations of this Alpha software. It's all based on probabilistic heuristics, combined with a collection of rules developed from domain knowledge gathered either from high-level games previously played by pros or from a database of games played against itself. Which is obviously a body of knowledge that exceeds any human's ability to prepare before such a match, since the human brain doesn't operate with tens of thousands of cores doing parallel processing and storing results without any error.
Deepmind's concept was novel in that its program had a sort of "intuition layer" (in technical terms: policy network), which develops, through machine learning, what can reasonably be described as intuition about which moves are worth exploring and which aren't. This layer of the program, like human intuition, comes up with promising candidates and is then combined with an "old school" search algorithm to evaluate them. The policy network is even developed the same way as human intuition: by playing lots and lots of games and learning from them. It forms connections based on previous results, and uses the memory thereof to find promising moves in future games.
Yes and by doing this it far exceeds any single human's possibilities of preparing for such a match, since humans can't just simulate thousands of simultaneous games in one brain to learn from them and even if there was a technical way for a human to do that, they couldn't use the results since it would far exceed their ability to memorise the results or heuristics learned from them. In the end, it's the computers' quantitative abilities which gives them the edge, there's no actual thinking or creativity behind how they develop their body of knowledge based on which they play. It's all just tried and tested math algorithms that offer the best probabilistic chances of making winning moves.
By developing intuition through pure self-play, AG was able to discover ways to play Go that were unknown to humans or considered inferior, caught professional players by surprise, and have since found their way into the meta. Players who refused to adapt to the "AI way" can no longer compete on a pro level. Notably, AG was able to find what can only be described as "creative" new moves, simply by playing against itself. Not by "solving" the game through exhaustive analysis of each possible game state, which is impossible.

I recommend the full doc: https://www.youtube.com/watch?v=WXuK6gekU1Y
From what I read in the description of Alpha, the software surprised high-level Go players by not focusing on a specific area on the table, like human players typically do, but rather spreading its focus around the table. But that software doesn't actually think, this is just automatically developed algorithms applying optimal moves that resulted from sifting through huge amounts of data from playing against itself. It's a tool that simulated a lot of possible moves and ranked them according to the probability of giving an advantage.
User avatar
Nauru Dolan
Ninja
Posts: 13064
Joined: Sep 17, 2015

Re: European politics

Post by Dolan »

fightinfrenchman wrote:
Dolan wrote:@scarm There's an even more striking example: what about composing some music. How could an AI/algorithm compose something that can convey a feeling, without actually having a feeling?
What about drawing a picture? Sonny did this in I, Robot
He wasn't even looking at the paper while drawing, he was basically building an image, line by line, like a printer
User avatar
No Flag fightinfrenchman
Ninja
Donator 04
Posts: 23505
Joined: Oct 17, 2015
Location: Pennsylvania

Re: European politics

Post by fightinfrenchman »

Dolan wrote:
fightinfrenchman wrote:
Dolan wrote:@scarm There's an even more striking example: what about composing some music. How could an AI/algorithm compose something that can convey a feeling, without actually having a feeling?
What about drawing a picture? Sonny did this in I, Robot
He wasn't even looking at the paper while drawing, he was basically building an image, line by line, like a printer
The fact that I made someone else think about the 2004 film I, Robot is a win in my book tbh
Dromedary Scone Mix is not Alone Mix
Image
User avatar
Nauru Dolan
Ninja
Posts: 13064
Joined: Sep 17, 2015

Re: European politics

Post by Dolan »

I just looked it up on YT, for like 1 min
User avatar
Netherlands Goodspeed
Retired Contributor
Posts: 13002
Joined: Feb 27, 2015

Re: European politics

Post by Goodspeed »

Dolan wrote:
Goodspeed wrote:Only in theory. And this is why I said that everything can, in theory, be solved by maths alone. Much like with Go, though, it's just not realistic or even worth trying. Go, like many other tasks humans engage in, is too complex for any finite number of supercomputers to "solve" this way. This is why I told @scarm that it's plenty complex for a proof of concept, and it's why Deepmind chose Go when they started the project. They wanted something that wasn't possible to brute force precisely because it would be a proof of concept.

Neural networks actually don't go through the entire move tree and exhaust every possibility until they've found the correct move. Yes this is how Chess "AI" (used to) work, because in Chess it's possible, as long as your search is efficient enough, to brute force your way to near perfect play. But that's not how AlphaGo works because it's not realistic in Go, no matter how efficient your search.

When a Go player looks at a position, while there may be 300 possible moves, there are really less than 10 that are worth exploring. And how do you describe which are worth exploring? You can't. Developers have tried and failed miserably. There's too much "it just feels right". Intuition. Anyway you take those 10 candidates, explore them by again finding moves your opponent would likely respond with (using intuition), explore those, etc, until you land on a move that's most promising. Anyone who has played any turn-based strategy game before knows this process.
But that's how the Monte Carlo Tree Search works. It uses playouts, which are literally game outcomes that have been played out. After each outcome is recorded, moves that were most likely to lead to victory become the new starting point for subsequent optimised moves.
https://en.wikipedia.org/wiki/Monte_Car ... _operation
No it's not. Monte carlo search uses random playouts as a way to more efficiently discard entire nodes in the tree (so it still explores every move, just very efficiently) whereas AG's choices about which moves to explore are based on previous experience, and most of the possible moves are not explored at all. It's not just an optimization of an existing algorithm. It's a fundamentally different approach.
Apparently this approach has been used to solve chess and Go up to a table size of 9x9. Once the table became larger, it was too computationally expensive, since the branching factor was getting too high. So what they did was use a layered approach, which included multiple heuristics to estimate the best moves (including by using knowledge from played out moves). Different strategies were used at different points in the game too, since at the start of the game there are fewer possible moves than midgame, while the more pieces are placed on the table the number of possible moves may actually decrease in time. Hence, using the same computational strategy all throughout the match wouldn't be optimal.
Yes like I said, they used the policy network to find promising candidates, which can be compared to human intuition, and then combined it with the old school search algorithm (monte carlo) to explore them. Obviously that last part isn't the impressive part. The thing that developers have tried and failed to do for decades, but Deepmind succeeded at, is mimicing the human ability to quickly determine which of the 300 possible moves are worth exploring. Without that, no amount of efficient searching, monte carlo or otherwise, got anywhere close to the level of the best human Go players.
In the end, there's nothing magical about how such a performance was reached, there's no actual "creativity" behind the winning moves chosen by the different iterations of this Alpha software.
I'm aware there's nothing magical. What I said was that it's a big step forward in the field. I would argue about the creativity though. In my opinion, there's nothing "magical" about our creativity either. What we call creativity is simply applying past experiences in new ways. That's exactly what programs like AG do. It learned from its past experience and applied that knowledge in ways that revolutionized the game. Stamped
It's all based on probabilistic heuristics, combined with a collection of rules developed from domain knowledge gathered either from high-level games previously played by pros or from a database of games played against itself. Which is obviously a body of knowledge that exceeds any human's ability to prepare before such a match, since the human brain doesn't operate with tens of thousands of cores doing parallel processing and storing results without any error.
My knowledge of Go is based on the ~2000 games I've played. Sure, I don't know them all by heart, but neither does the AlphaGo agent that sits down to play a game of Go remember all of the games it played. Like you said it created a collection of rules based on those games, which it then uses in future games. When I think about a move, I apply rules I've learned from playing the game in the past. There's no need for me to go back and actually remember where I learned them, not to mention it would be highly inefficient to have to do so. AG works the same way.
Yes and by doing this it far exceeds any single human's possibilities of preparing for such a match, since humans can't just simulate thousands of simultaneous games in one brain to learn from them and even if there was a technical way for a human to do that, they couldn't use the results since it would far exceed their ability to memorise the results or heuristics learned from them. In the end, it's the computers' quantitative abilities which gives them the edge, there's no actual thinking or creativity behind how they develop their body of knowledge based on which they play. It's all just tried and tested math algorithms that offer the best probabilistic chances of making winning moves.
Of course it's all maths, but I see no reason to believe our brains can't be replicated that way. And yeah, human brains have found ways to learn very efficiently, or rather: It looks efficient from the outside. I think you underestimate the amount of processing power in our brains. In the end, they too are just doing a shit ton of calculations.
I have the feeling your position ultimately stems from your belief that our brains are somehow magical. I see no evidence for this. Creativity and consciousness are not magic.
From what I read in the description of Alpha, the software surprised high-level Go players by not focusing on a specific area on the table, like human players typically do, but rather spreading its focus around the table. But that software doesn't actually think, this is just automatically developed algorithms applying optimal moves that resulted from sifting through huge amounts of data from playing against itself. It's a tool that simulated a lot of possible moves and ranked them according to the probability of giving an advantage.
Human players spread their attention across the board too. That's a very basic concept that you learn as a beginner. The reason the AI is better at that than humans is because it's not held back by the human psychological tendency to focus a little too much on the area of the board that's changing. Hard to explain this to someone who doesn't play the game, but it's just one of many things the program was better at than human players. The biggest surprise was simply that it was better than humans at the game in general, which no one saw coming. At the time it was believed (and for good reason) that computers beating humans at Go was still decades away, if possible at all. The second biggest surprise, which played out in the years afterwards, was how much these neural network-based programs ended up changing the meta.

What does it mean to "actually think" and why does it matter in this context whether or not the program is "actually thinking"? Am I not using my past experiences when I "think" about a move?

Who is online

Users browsing this forum: No registered users and 8 guests

Which top 10 players do you wish to see listed?

All-time

Active last two weeks

Active last month

Supremacy

Treaty

Official

ESOC Patch

Treaty Patch

1v1 Elo

2v2 Elo

3v3 Elo

Power Rating

Which streams do you wish to see listed?

Twitch

Age of Empires III

Age of Empires IV