Can "AI" discover new Physics

[Macavity]

Maybe we're not that into "talking computers"! But: Can "AI" discover new Physics? 🤔
For a change not the Youtube link! (But there is a FUN video embedded!)
https://www.creativemachineslab.com/hidden-variables.html
Interesting stuff, if you're into Robotics? 🥳

[xtreemchaos]

theres not bounds to what Ai can do with a bit of luck it may improve my spelling 😀 no really theres no limit when when we build more quantum computers to check if the Ai was right or not.

[cajen2]

"Open the pod bay doors, Hal".....

[saac]

What if we get an Ai to develop a new Ai which in turn develops a ,,,

Jim 

[vlaiv]

This is very cool

[Ratlet]

It's the interesting thing about the equations in physics, they don't describe how it is, the just describe how it behaves so it's pretty likely ai will find alternative explanations.

An AI could possibly sort through a large number of systems and develop new equations that describe them and then try and amalgamate them into a single equation.  Could be very helpful as we work towards a grand unified theory.

[andrew s]

Yes but... ...your not going to like the answer.

42

Regards Andrew 

[Zeta Reticulan]

I'm a bit sceptical about the 'ghost in the machine'. I just can't see software as consciousness. Although I've been unconscious for much of my life. "I drink therefore I am"  as Des Carts (or someone) purportedly stated. *burp*

[Guest]

6 minutes ago, andrew s said:

Yes but... ...your not going to like the answer.

42

Regards Andrew 

You beat me to it

[DaveL59]

There's a lot of hype around AI but considering that it will have been designed and programmed by us humans. It'll have defined logic and methodologies that we have crafted and coded. Now install it in a fast enough box and it may be able to seem to perform as if it had intelligence and with algorithms perhaps even seem intuitive when it 'predicts' and responds to situations, conversation etc.

But the big question, for it to actually be intelligent as opposed to just another app, can it think outside the box?

Would it ever be able to make a leap of understanding or hypothesis as humans have done to reach the level of development we have, or will it always be tied to follow the paths defined to it ad nauseum?

Where it is perceived to have "learned", for example a new language, is it simply mapping inter-related constructs and deriving a suitable output or has it really learned and can thereby hold complex conversation in a natural style?

Edited July 31, 2022 by DaveL59

[PaulM]

22 minutes ago, DaveL59 said:

There's a lot of hype around AI but considering that it will have been designed and programmed by us humans. It'll have defined logic and methodologies that we have crafted and coded. Now install it in a fast enough box and it may be able to seem to perform as if it had intelligence and with algorithms perhaps even seem intuitive when it 'predicts' and responds to situations, conversation etc.

But the big question, for it to actually be intelligent as opposed to just another app, can it think outside the box?

Would it ever be able to make a leap of understanding or hypothesis as humans have done to reach the level of development we have, or will it always be tied to follow the paths defined to it ad nauseum?

Where it is perceived to have "learned", for example a new language, is it simply mapping inter-related constructs and deriving a suitable output or has it really learned and can thereby hold complex conversation in a natural style?

Couldnt have said it better

AI is a lot of IF statements

If it really worked it would be having a much larger impact on our daily lives

[Macavity]

7 hours ago, Ratlet said:

It's the interesting thing about the equations in physics, they don't describe how it is, the just describe how it behaves so it's pretty likely ai will find alternative explanations.

An AI could possibly sort through a large number of systems and develop new equations that describe them and then try and amalgamate them into a single equation.  Could be very helpful as we work towards a grand unified theory.

Thanks for that - And for actually reading the article? I guess the term "AI" is a trigger? 🤔
Perhaps they should call their (to me) "innocuous, but interesting" ideas, something else?!?
On lighter note, the "constant" in front of the "new parameters" might always be ONE? 🥳

<thinking> This AND Sabine Hossenfelder (watched earlier) all in ONE day? 😸

Edited August 1, 2022 by Macavity

[iantaylor2uk]

There are various types of AI - some of the simplest are no better than simple statistical regression analysis. 

In others, there is "supervised" learning, in which the AI model is "trained" on data, and then it is used on other data, although these methods can only interpolate, not extrapolate.

However, one of the most interesting types of AI is "unsupervised" learning, where the AI program is programmed to "train itself". This was the approach taken by the Deepmind team when it developed a program to play the game Go. The AI trained itself and became good enough to beat the Go world champion, and also produced some very unexpected moves (move 37, game two) during the games (for more details see: https://www.deepmind.com/research/highlighted-research/alphago and https://www.wired.com/2016/03/two-moves-alphago-lee-sedol-redefined-future/)

For unsupervised AI, although the model has been programmed by a human, there is no telling what the final outcome of the AI will be, as we cannot see "under the bonnet" to see how the AI teaches itself, or how it uses that learning to come up with improved algorithms. I would have thought that it would be the unsupervised AI that is the closest thing to human intelligence (as we basically teach ourselves, with the help of others, for the first 20 or so years of our lives). 

[Macavity]

Thanks! I think it important not to "throw the baby out with the bathwater"!?! 🙃
Highlighted by the following paragraph re. CERN (LHC) data selection/analysis:

It has different meanings in different contexts, and CERN is mainly interested in task-oriented, so-called restricted AI, rather than general AI involving aspects such as independent problem-solving, or even artificial consciousness. Particle physicists were among the first groups to use AI techniques in their work, adopting Machine Learning (ML) as far back as 1990.  Beyond ML, physicists at CERN are also interested in the use of Deep Learning to analyse the data deluge from the LHC.

The actual LINKS can be found on the following page:
https://sparks.web.cern.ch/index.php/ai-cern

P.S. Apparently the links copied over... Ah, the JOY of intelligent computers! 😎

Edited August 1, 2022 by Macavity

[Ouroboros]

I am not convinced the AI is doing “discovery”.  Isn't that down to the human looking at the outputted data and interpreting it?

I am wondering whether this is merely a more sophisticated version of something that was done pre-AI which was to programme a computer to randomly rearrange algebraically a huge number of physical formulae in the hope that some interesting relationships might crop up serendipitously. 

[vlaiv]

9 minutes ago, Ouroboros said:

I am not convinced the AI is doing “discovery”.  Isn't that down to the human looking at the outputted data and interpreting it?

I am wondering whether this is merely a more sophisticated version of something that was done pre-AI which was to programme a computer to randomly rearrange algebraically a huge number of physical formulae in the hope that some interesting relationships might crop up serendipitously. 

I think that AI is in fact doing the "discovery" - in the same sense humans are doing the discovery.

Looking at outputted data and interpretation of it - is actually more like "translation" from foreign language of AI then discovery.

AI is actually neural network that is trained to anticipate how the system will behave.

This is something similar to human standing there and concluding - look if apple detaches from the tree - it will always fall straight down with accelerated motion.

That is step one - finding repeating phenomena and understanding sequence of events that describes that phenomena.

If either we or AI can then say - ok, if system is in state A - it will be in state B after given amount of time - then we have variables and equations (of some sorts). Given that AI can predict how system will behave in future time - means it has variables and equations in one form or another (It performed the discovery).

Interesting part is assertion that there is minimum amount of variables that describe certain system (degrees of freedom) and that those don't change whatever framework you put them in - a bit like vectors - they don't change regardless of how you express their coordinates (polar, euclidean, ....)

These variables and equations are "encoded" in neural network of AI in this example and trick is to "translate" or "read the brain" of AI in order to try to understand - how many variables there are, what they represent and what are the equations.

[Ouroboros]

I see you’re a believer. I’m somewhat of an AI sceptic. I think it should be called AS - artificial stupidity. 😜

It’s still just a sophisticated automaton though isn’t it? One has to ask the philosophical question. Can something be described as discovered (in the scientific sense) unless put into its contextual significance. A dog can “discover” a bone, but the archaeologist or palaeontologist is required to understand the scientific significance. In other words to make what we would describe as a scientific discovery. 

[Adam J]

14 hours ago, Macavity said:

Maybe we're not that into "talking computers"! But: Can "AI" discover new Physics? 🤔
For a change not the Youtube link! (But there is a FUN video embedded!)
https://www.creativemachineslab.com/hidden-variables.html
Interesting stuff, if you're into Robotics? 🥳

Don't confuse AI with Artificial General Intelligence, which is essentially a term invented to cover the fact that people are now using the term AI for things that are not infact what traditionally would have counted for AI, hence now AI doesn't mean AI we now have AGI which is what AI used to be. 

Adam 

[vlaiv]

5 minutes ago, Ouroboros said:

It’s still just a sophisticated automaton though isn’t it?

So are we, aren't we?

7 minutes ago, Ouroboros said:

Can something be described as discovered (in the scientific sense) unless put into its contextual significance.

It's all about pattern matching.

Level of sophistication is just how deep pattern matching goes (or IQ for that matter).

[Ouroboros]

15 minutes ago, Adam J said:

Don't confuse AI with Artificial General Intelligence, which is essentially a term invented to cover the fact that people are now using the term AI for things that are not infact what traditionally would have counted for AI, hence now AI doesn't mean AI we now have AGI which is what AI used to be. 

Adam 

Yes, that was discussed  by Stuart Russel in his BBC Reith Lectures last year. Worth a listen.

Artificial (general) intelligence doesn’t of itself worry me as much as the gullibility of humans to believe in it.  

[saac]

14 hours ago, DaveL59 said:

There's a lot of hype around AI but considering that it will have been designed and programmed by us humans. It'll have defined logic and methodologies that we have crafted and coded. Now install it in a fast enough box and it may be able to seem to perform as if it had intelligence and with algorithms perhaps even seem intuitive when it 'predicts' and responds to situations, conversation etc.

But the big question, for it to actually be intelligent as opposed to just another app, can it think outside the box?

Would it ever be able to make a leap of understanding or hypothesis as humans have done to reach the level of development we have, or will it always be tied to follow the paths defined to it ad nauseum?

Where it is perceived to have "learned", for example a new language, is it simply mapping inter-related constructs and deriving a suitable output or has it really learned and can thereby hold complex conversation in a natural style?

The thing is Dave when you say "crafted and coded by us" we also have to recognise the not insignificant imprint of nature (evolution) encoded in our logical schemas.  The way we think has had the hand of time, around 3.7 billion years, moulding , tearing down and making anew, shaping our thought processes. Nature's fail fast philosophy means the design is pretty well refined now over those 3.7 billion years. Afterall no other solution has brought us to the stage of developing an AI itself; maybe the human brain is as good as it gets. It has certainly been that way in those 3.7 billion years! 

Jim 

[Ouroboros]

24 minutes ago, vlaiv said:

So are we, aren't we?

It's all about pattern matching.

Level of sophistication is just how deep pattern matching goes (or IQ for that matter).

Maybe we are. Maybe we’re not. It rather depends who you read.

But yes, in general terms my feeling is you’re probably right.

I still feel this AI is more like a dog than a human being. For the moment. 

Edited August 1, 2022 by Ouroboros

[vlaiv]

11 minutes ago, Ouroboros said:

I still feel this AI is more like a dog than a human being. For the moment. 

I think that dog is much more advanced than most AIs out there.

It is really down to size of neural network - number of neurons and connections.

Most AI neural networks are rather small in size compared to living beings, but they are sort of optimized for particular task.

[iantaylor2uk]

I don't think a dog would come up with new mathematical ideas: https://www.ox.ac.uk/news/2021-12-01-machine-learning-helps-mathematicians-make-new-connections

[Macavity]

1 hour ago, Adam J said:

Don't confuse AI with Artificial General Intelligence, which is essentially a term invented to cover the fact that people are now using the term AI for things that are not infact what traditionally would have counted for AI, hence now AI doesn't mean AI we now have AGI which is what AI used to be.

OKaaaaaaaaay! Above must be a record for appearences of "AI" in once sentence. [teasing]
I think overall, it has been illuminating for everyone? Ever wish you hadn't started something? 😝