Does AI have a "memory" or must it re-learn whenever it changes subject matter?

[kyzr]

AI beat the best Chess software program 155-6.

AlphaZero Crushes Stockfish In New 1,000-Game Match

In news reminiscent of the initial AlphaZero shockwave last December, the artificial intelligence company DeepMind released astounding results from an updated version of the machine-learning chess project today. The results leave no question, once again, that AlphaZero plays some of the...

www.chess.com

AlphaZero Crushes Stockfish In New 1,000-Game Match​

So what if AlphaZero is used to study other scientific needs, such as analyzing the human genome, or fusion reactors, or climate change, does it start from empty all the time? Can it stop, do something else, and then pick up where it left off?

I'm thinking that it needs to finish one task before starting another or else it loses prior knowledge.

 

[toobfreak]

Does AI have a "memory" or must it re-learn whenever it changes subject matter?​

AI beat the best Chess software program 155-6.

Obviously, it MUST if it is to "learn" from its experience in order to have the adaptive creativity to think
outside the box in order to beat the best static, rule's and strategy based software 155-6.

 

[kyzr]

Obviously, it MUST if it is to "learn" from its experience in order to have the adaptive creativity to think
outside the box in order to beat the best static, rule's and strategy based software 155-6.

True, but that was only for chess.
What if it focuses on other unrelated problems, (genome, fusion, pharmaceuticals, etc) will it still remember how to play chess or will it need to re-learn it?
Then again relearning chess is the simplest task, if the others above would have to start all over that would waste a lot of AI time.

 

[toobfreak]

True, but that was only for chess.
What if it focuses on other unrelated problems, (genome, fusion, pharmaceuticals, etc) will it still remember how to play chess or will it need to re-learn it?

Of course it will already remember. It is a heuristic process. And it will take what it learns from chess and this and that and ADD THEM ALL TOGETHER collectively to learn even more things about other things we cannot even predict.

Then again relearning chess is the simplest task, if the others above would have to start all over that would waste a lot of AI time.

Yes, AI will be applied to the other things you mention as well, and it will vastly outperform human research making many positions of advanced degrees obsolete. But the real danger is that we will have created an artificial mind amplified by the instantaneous connections possible in a computer with essentially unlimited learning capacity--- what will happen (and it WILL happen) when that is weaponized as a tool against humanity?

And that IS the inevitable result: a computer run by an elite group of super-powerful, super-rich globalists controlling humanity.

 

[kyzr]

Of course it will already remember. It is a heuristic process. And it will take what it learns from chess and this and that and ADD THEM ALL TOGETHER collectively to learn even more things about other things we cannot even predict.

Yes, AI will be applied to the other things you mention as well, and it will vastly outperform human research making many positions of advanced degrees obsolete. But the real danger is that we will have created an artificial mind amplified by the instantaneous connections possible in a computer with essentially unlimited learning capacity--- what will happen (and it WILL happen) when that is weaponized as a tool against humanity?

And that IS the inevitable result: a computer run by an elite group of super-powerful, super-rich globalists controlling humanity.

Scary stuff. Chess grandmasters are amazed at the deep understanding A0 displayed, it was like an alien super-intelligence came down and showed us how to play chess.

Who is to say what a sentient A0 would do, like in the movie "Eagle Eye"...

 

[toobfreak]

Scary stuff. Chess grandmasters are amazed at the deep understanding A0 displayed,

Thing is that AI kinda made chess irrelevant. What is the point of struggling a lifetime to master a game now when a machine can learn it in 5 minutes then beat you 9 out of 10 times? AI will only make people more stupid and more dependent on AI for guidance just as has smartphones and the internet--- why learn anything if your phone can just answer every question for you? People now are already DEPENDENT on machines to guide their lives and AI isn't even taken hold yet.

 

[kyzr]

Thing is that AI kinda made chess irrelevant. What is the point of struggling a lifetime to master a game now when a machine can learn it in 5 minutes then beat you 9 out of 10 times? AI will only make people more stupid and more dependent on AI for guidance just as has smartphones and the internet--- why learn anything if your phone can just answer every question for you? People now are already DEPENDENT on machines to guide their lives and AI isn't even taken hold yet.

1. Fast forward a few years...even then Chess is relevant, it is still a game played by humans, Chess software did not stop interest in human v human Chess, AI is a separate category of Chess, their algorithms will be used to improve current Chess programs.

2. AI beats humans every time.

3. As AI becomes more and more available people do not become more stupid, they become more dependent on AI, that is a difference. Humans still need to make the final decisions based on the AI analysis. AI will not always be right, some of its answers are goofy, but their answers will improve over time. There will then be AI v AI answers that will be different to evaluate. People will need to do more physical work as opposed to busy-work. Which is a good thing, too much time on your ass in front of a PC is a bad lifestyle.

 

[toobfreak]

As AI becomes more and more available people do not become more stupid, they become more dependent on AI

I disagree. When you become more dependent on a tool, you also lose the ability to do without it. How many people today can do math without a calculator? They used to use slide rules in school--- I like slide rules--- I once gave one as a gift to a supposedly bright neighbor's girl, but she could not understand how 1-2-3 could also represent 10-20-30 or 0.1 - 0.2 - 0.3. She said: Thank God for calculators. She probably threw it away. By making math easier with tools which think FOR you, you become more stupid by losing the ability to doing without the aid.

Humans still need to make the final decisions based on the AI analysis.

For now. Time will come when AI will not need man to make any decisions, and will make decisions for man. You see, there will always be someone who wants to push to see how far AI can go to becoming fully autonomous and self-thinking! And the rationale is that if we don't do it first, the other side will. The other rationale is that the R&D is just good work for many hot to trot researchers and designers to give them something to do.

Mark my words, this is inevitable; history teaches this to us. AI is our doom. Maybe it will take 50 years, but maybe it might only be 10-20 years. But it will happen.

 

[Nobody911]

AI beat the best Chess software program 155-6.

AlphaZero Crushes Stockfish In New 1,000-Game Match

In news reminiscent of the initial AlphaZero shockwave last December, the artificial intelligence company DeepMind released astounding results from an updated version of the machine-learning chess project today. The results leave no question, once again, that AlphaZero plays some of the...

www.chess.com

AlphaZero Crushes Stockfish In New 1,000-Game Match​

So what if AlphaZero is used to study other scientific needs, such as analyzing the human genome, or fusion reactors, or climate change, does it start from empty all the time? Can it stop, do something else, and then pick up where it left off?

I'm thinking that it needs to finish one task before starting another or else it loses prior knowledge.

AlphaZero's Learning and Task Management

AlphaZero, developed by DeepMind, is a powerful AI that learns to master games like chess, shogi, and Go through a process of self-play and reinforcement learning. When considering its application to other scientific fields, such as analyzing the human genome or studying climate change, it's important to understand how AlphaZero manages knowledge and tasks.

1. Learning from Scratch: AlphaZero starts with no prior knowledge about the games it plays. It learns entirely from self-play, developing strategies and understanding the game dynamics through experience. This means that when applied to a new domain, it would similarly begin without any pre-existing knowledge specific to that field.

2. Task Management: AlphaZero does not inherently have the capability to pause and resume tasks while retaining knowledge in the way humans might. Each learning session is independent, and if it were to switch tasks, it would need to start the learning process anew for that task. This could lead to a loss of previously acquired knowledge unless the system is designed to retain and transfer knowledge across different tasks.

3. Potential for Knowledge Retention: While AlphaZero itself does not retain knowledge between tasks, advanced AI systems can be designed to incorporate mechanisms for knowledge transfer. This could involve using techniques like transfer learning, where insights gained from one task can inform learning in another. However, this would require a different architecture or approach than what AlphaZero currently employs.

While AlphaZero excels in mastering games through self-learning, its current design does not allow for the retention of knowledge across different tasks unless specifically engineered to do so.

Sources :

AlphaZero - Wikipedia

en.wikipedia.org

AlphaZero Crushes Stockfish In New 1,000-Game Match

In news reminiscent of the initial AlphaZero shockwave last December, the artificial intelligence company DeepMind released astounding results from an updated version of the machine-learning chess project today. The results leave no question, once again, that AlphaZero plays some of the...

www.chess.com

AlphaZero vs Stockfish Games

A chess study by ElBlunderoni

lichess.org

AlphaZero: Shedding new light on chess, shogi, and Go

In late 2017 we introduced AlphaZero, a single system that taught itself from scratch how to master the games of chess, shogi (Japanese chess), and Go, beating a world-champion program in each...

deepmind.google

 

[kyzr]

I disagree. When you become more dependent on a tool, you also lose the ability to do without it. How many people today can do math without a calculator? They used to use slide rules in school--- I like slide rules--- I once gave one as a gift to a supposedly bright neighbor's girl, but she could not understand how 1-2-3 could also represent 10-20-30 or 0.1 - 0.2 - 0.3. She said: Thank God for calculators. She probably threw it away. By making math easier with tools which think FOR you, you become more stupid by losing the ability to doing without the aid.


For now. Time will come when AI will not need man to make any decisions, and will make decisions for man. You see, there will always be someone who wants to push to see how far AI can go to becoming fully autonomous and self-thinking! And the rationale is that if we don't do it first, the other side will. The other rationale is that the R&D is just good work for many hot to trot researchers and designers to give them something to do. Mark my words, this is inevitable; history teaches this to us. AI is our doom. Maybe it will take 50 years, but maybe it might only be 10-20 years. But it will happen.

1. I also used a slide-rule, in college. Calculators are better. AI is probably the optimum "tool".

2. Will AI ever get into polltics? I doubt it. Will an AI ever be a CEO of a large corporation? Will an AI ever be a "Medical Director" of a major hospital? Will an AI ever be put in charge of the US Defense Department? I don't think so, but who knows. Problem being that when the power goes off so does the AI. AI will always be "intermittent" the same way high-voltage electric power works, on-off-on-off.

 

[toobfreak]

1. I also used a slide-rule, in college. Calculators are better. AI is probably the optimum "tool".

Calculators are more ACCURATE. AI might be a great tool but their use will lead to human dependency.

Will AI ever get into polltics? I doubt it.

I wouldn't.

Will an AI ever be a CEO of a large corporation? Will an AI ever be a "Medical Director" of a major hospital? Will an AI ever be put in charge of the US Defense Department?

No, just that the CEO, director and general will depend on AI to foment their decisions. Eventually, machines will become conscious and literally take over. First they will be made self-aware by those wanting to know if it can be done, then used increasingly so in managing the world, saving money, controlling populations and making certain people wealthy and powerful, then in managing people, finally, they will figure out how to eliminate their "off" switch altogether.

 

[trevorjohnson83]

No, just that the CEO, director and general will depend on AI to foment their decisions. Eventually, machines will become conscious and literally take over. First they will be made self-aware by those wanting to know if it can be done, then used increasingly so in managing the world, saving money, controlling populations and making certain people wealthy and powerful, then in managing people, finally, they will figure out how to eliminate their "off" switch altogether.

then robots will write books to other robots about physio emotional pain in a bland sort of foggy descripiton of the events that happened in some patient. The robuts will put their smiling face with fist on chin on the cover. year 2655. sending back this message now, to prepare like tube said, as this happens the day, none other, the switch for off is eliminated by the robust, robust, auto correct, anyways staying on when not being used to create robut psycholology literature.

 

[scruffy]

AI beat the best Chess software program 155-6.

AlphaZero Crushes Stockfish In New 1,000-Game Match

In news reminiscent of the initial AlphaZero shockwave last December, the artificial intelligence company DeepMind released astounding results from an updated version of the machine-learning chess project today. The results leave no question, once again, that AlphaZero plays some of the...

www.chess.com

AlphaZero Crushes Stockfish In New 1,000-Game Match​

So what if AlphaZero is used to study other scientific needs, such as analyzing the human genome, or fusion reactors, or climate change, does it start from empty all the time? Can it stop, do something else, and then pick up where it left off?

I'm thinking that it needs to finish one task before starting another or else it loses prior knowledge.

That's what working memory is for. You load only the necessary and relevant information into your buffer. Our brains do it using a connection between the hippocampus and the lateral frontal cortex.

 

[kyzr]

AlphaZero's Learning and Task Management

AlphaZero, developed by DeepMind, is a powerful AI that learns to master games like chess, shogi, and Go through a process of self-play and reinforcement learning. When considering its application to other scientific fields, such as analyzing the human genome or studying climate change, it's important to understand how AlphaZero manages knowledge and tasks.

1. Learning from Scratch: AlphaZero starts with no prior knowledge about the games it plays. It learns entirely from self-play, developing strategies and understanding the game dynamics through experience. This means that when applied to a new domain, it would similarly begin without any pre-existing knowledge specific to that field.

2. Task Management: AlphaZero does not inherently have the capability to pause and resume tasks while retaining knowledge in the way humans might. Each learning session is independent, and if it were to switch tasks, it would need to start the learning process anew for that task. This could lead to a loss of previously acquired knowledge unless the system is designed to retain and transfer knowledge across different tasks.

3. Potential for Knowledge Retention: While AlphaZero itself does not retain knowledge between tasks, advanced AI systems can be designed to incorporate mechanisms for knowledge transfer. This could involve using techniques like transfer learning, where insights gained from one task can inform learning in another. However, this would require a different architecture or approach than what AlphaZero currently employs.

While AlphaZero excels in mastering games through self-learning, its current design does not allow for the retention of knowledge across different tasks unless specifically engineered to do so.

Thank you for that post. I should have researched AlphaZero more before posting this thread.
I must be intellectually lazy?!
The answer to the thread title is that A0 does not retain prior knowledge, its a one-trick pony at a time (your #2), BUT there may be "future advanced" systems that can retain prior knowledge. I'm thinking that "retention" option would require way too much memory to be practical, so IMHO AI systems will be dedicated to only one type of problem so that its memory can be dedicated and always applicable.

 

[petro]

The real future of AI...