AI and the proton

[billhinge]

Interesting use of AI, wonder how many of our other assumptions are 'wrong'?

 

 

[Mr Spock]

Looks to me like a misuse (again) of the term AI. What they have there is a computer model which has been programmed, it's not AI. AI is something which can adapt and improve without human input.

[billhinge]

I think the whole AI debate is semantics, it is a back propagation convolution  which I have no issue calling AI, the architecture is here 

https://docs.nnpdf.science/n3fit/methodology.html#neural-network-architecture

code is opensource

https://nnpdf.mi.infn.it/nnpdf-open-source-code/

Paper here

https://arxiv.org/pdf/2109.02671.pdf

 

[Ouroboros]

2 minutes ago, billhinge said:

I think the whole AI debate is semantics, it is a back propagation convolution  which I have no issue calling AI, the architecture is here 

https://docs.nnpdf.science/n3fit/methodology.html#neural-network-architecture

code is opensource

https://nnpdf.mi.infn.it/nnpdf-open-source-code/

Paper here

https://arxiv.org/pdf/2109.02671.pdf

 

“back propagation convolution” ?   I’m trying to get my head around what that actually means. 🤫  I understand what the individual words mean.  

[Ouroboros]

1 hour ago, billhinge said:

Interesting use of AI, wonder how many of our other assumptions are 'wrong'?

 

 

Unless I missed it I don’t think he says what the tentative AI model of the proton is. Can we assume that the model still contends that the proton consists of three quarks for example? Or are we talking some completely different model?

As for what is AI and what isn’t AI …. Would it be correct the assume that the AI used to reanalyse the collission data is a bit like the so-called AI used to, for example, look at X-ray images to find tumours?  In the medical example the AI has to first be told and learn what tumours look like so it can then examine data it hasn’t seen before. Are we talking about that sort of approach for the proton modelling? 

[FunkyKoval35]

49 minutes ago, billhinge said:

I think the whole AI debate is semantics, it is a back propagation convolution  which I have no issue calling AI, the architecture is here

Agree. Most of the people thinks that inteligence = self-consciousness. Well, mabey in the future, but self-consciousness is not needed for inteligence.

1 hour ago, Mr Spock said:

Looks to me like a misuse (again) of the term AI. What they have there is a computer model which has been programmed, it's not AI. AI is something which can adapt and improve without human input.

With your definition humans are not inteligent because kids are learned at school and by parents - other human input . Inteligence is, first of all, ability to solve problems/tasks. The process of acquiring this ability is irrelevant.

Edited June 22, 2023 by FunkyKoval35

[Mr Spock]

3 minutes ago, FunkyKoval35 said:

With your definition humans are not inteligent because kids are learned at school and by parents - other human input

Kids are taught at school, which is the acquisition of knowledge. Knowledge and intelligence aren't the same thing 

[FunkyKoval35]

10 minutes ago, Mr Spock said:

Kids are taught at school, which is the acquisition of knowledge. Knowledge and intelligence aren't the same thing 

Yes, but kids are also lerned how to solve the problems/tasks - reading, calculating, etc..

Edited June 22, 2023 by FunkyKoval35

[Mr Spock]

6 minutes ago, FunkyKoval35 said:

Yes, but kids are also lerned how to solve the problems/tasks - reading, calculating, etc..

Learning how to solve problems is knowledge. Intelligence is required to solve problems without input of any kind.

[billhinge]

When I did my msc back in 91 my project was on AI machine vision as it was called then. In those days we had to run on transputers and hand code everything in C and assembler to get any speed. AI was just a label

Today as an IT architect I'm still doing interesting things with AI, ontologies, semantics, digital twins etc amongst other things...

There is a concept known as the semantic triangle aka Ogdens triangle that relates labels to concepts and phenomena, provided you agree on concept and phenomena you can give it whatever label you like , eg 'temporal octopus training' since there isn't a standard label that maps label to concept/phenomena   (we actually debated this at the ITU in Geneva several years back)

 

[saac]

14 hours ago, Mr Spock said:

Kids are taught at school, which is the acquisition of knowledge. Knowledge and intelligence aren't the same thing 

Acquisition of knowledge is only part of the education process in a school Michael and it is by no means the focus.  Amongst other things, pupils "learn" skills such as "problem solving"  skills and are then given opportunity to apply those skills to new applications.  All AIs start with an initial programme a mixture of received knowledge and skills; in that respect they are very similar to pupils in a school. 

Jim 

Edited June 22, 2023 by saac

[saac]

14 hours ago, Ouroboros said:

Unless I missed it I don’t think he says what the tentative AI model of the proton is. Can we assume that the model still contends that the proton consists of three quarks for example? Or are we talking some completely different model?

As for what is AI and what isn’t AI …. Would it be correct the assume that the AI used to reanalyse the collission data is a bit like the so-called AI used to, for example, look at X-ray images to find tumours?  In the medical example the AI has to first be told and learn what tumours look like so it can then examine data it hasn’t seen before. Are we talking about that sort of approach for the proton modelling? 

I gave up on the video half way through, found it a bit annoying as he took too many detours - get to the point for crying out loud. Anyway, I suspect that what the AI would have been used for here would be (as you say) analysing the collision data but specifically looking for patterns. The notion itself that the quark composition of the proton is not static  but fluctuates is an established theory,  not new and not something advanced by an AI programme.  I may well be wrong but I would imagine the use of the AI here is to help to process the massive amount of data to find evidence to support the theory. 

Jim 

[billhinge]

7 hours ago, saac said:

I gave up on the video half way through, found it a bit annoying as he took too many detours - get to the point for crying out loud. Anyway, I suspect that what the AI would have been used for here would be (as you say) analysing the collision data but specifically looking for patterns. The notion itself that the quark composition of the proton is not static  but fluctuates is an established theory,  not new and not something advanced by an AI programme.  I may well be wrong but I would imagine the use of the AI here is to help to process the massive amount of data to find evidence to support the theory. 

Jim 

The upshot was that there are many models about quark composition , each research team having their favourite based on personal bias. However the machine has no human bias so it compares the observation data with those predicted  by models over the sum of all models. Thus seemingly ruling out what was previously most favoured and suggesting alternative models to 3 sigma so far.

[michael.h.f.wilkinson]

46 minutes ago, billhinge said:

The upshot was that there are many models about quark composition , each research team having their favourite based on personal bias. However the machine has no human bias so it compares the observation data with those predicted  by models over the sum of all models. Thus seemingly ruling out what was previously most favoured and suggesting alternative models to 3 sigma so far.

AI can have human biases due to data fed during training being biased. I am not saying that that is the case here, but AI being biased is a well-known phenomenon.

[billhinge]

OK less biased, unlike humans it wont say I've spent $50m working on project X and devoted my life to it so I'm going to push model X regardless of whatever evidence is presented that model Y or Z is better.

[saac]

13 hours ago, billhinge said:

The upshot was that there are many models about quark composition , each research team having their favourite based on personal bias. However the machine has no human bias so it compares the observation data with those predicted  by models over the sum of all models. Thus seemingly ruling out what was previously most favoured and suggesting alternative models to 3 sigma so far.

Not really sure personal bias is such a heavy an influence; models are either supported by data or not; the truth will out!. The advantage of using an AI algorithm is of course the ability to incrementally improve on the efficiency of data processing/analysis.  It is just another tool, very powerful one at that.  Anyway, to circle back to the OP, unless I am mistaken, I think the AI here is not advancing a new theory as more  being used to analyse data -  not that this detracts in anyway from its nature (AI).  Quite clever really.  

Jim  

Edited June 23, 2023 by saac

[900SL]

On 22/06/2023 at 10:35, Ouroboros said:

“back propagation convolution” ?   I’m trying to get my head around what that actually means. 🤫  I understand what the individual words mean.  

I think it's an agitator in a sewage plant

[Ouroboros]

13 hours ago, 900SL said:

I think it's an agitator in a sewage plant

Oh yes …. you don’t want an alligator in a sewage plant.