Reading in between the lines it seems most of the innovation is managing the regulations and privacy concerns of large institutions who want to use shared AI models.
> offers hospitals its experience in managing huge amounts of data in a secure and decentralized way, which is key to ensuring the privacy and security of the private patient information used to feed the algorithm. [...] uses a system to process data locally without sending it to central storage. This helps protect privacy and make better use of resources when different hospitals work together to create reliable AI-based models
Which probably makes sense why CERN developed some expertise in this area.
It feel silly to say that AI is curing cancer. Normally a phrase like that would sinal the apex of the hype cycle but I guess it actually has some meat in this situation. Using AI more like statistical inference to screen for medical conditions or predict treatment could be helpful. I remember Jeremy Howard from fast.ai did that with deep learning to detect things in medical images. Seems like a good thing for CERN to do as long as it works.
There is a venerable pattern in academic circles of branding your research with the language of whatever hype cycle is currently active. The same research can be branded under big data, computer vision, AI, or whatever else the current trends demand.
I can’t grasp what this is about. It starts saying it enables hospital to store and process their data onsite (not really an innovation in any way) but then later says the hospital computers still contact a main server for analysis. So is this just on premise data analysis (“ai”) but also cloud?
I understood it as this: Hospitals keep the data locally and run AI analysis on-site. They then exchange certain parameters (not the raw data) with a central server, which uses these parameters to update and improve the model. The central server doesn't run the models or process the data itself, it's just used to refine the AI models that the hospitals are using locally.
Someone please correct me if I understood it wrong.
Sounds like they use a method called federated learning. I.e. each site can train a local model on their own data and then share model parameters to a central entity that aggregates several sites model parameters. Superior performance can then be achieved (and also raw training data between sites are not shared).
It is and has been for a while, but most of the more flashy and exciting developments in ML and AI don't have very much applicability to LHC event processing. To be able to state any kind of finding about some aspect of physics based on the scattering of particles in the accelerator and their decays in the detector, you need to take the background of all events and make multivariate discriminants on the data in order to enrich your signal as much as possible while throwing as little as possible away. This requires you to have a rigorous and verifiable statistical "paper trail" from start to finish, so you can say with confidence intervals how much signal and background you ought to have, vs how much you measure in your data after processing it. An overly broad black box doesn't really work for this kind of introspection.
Is this really about CERN’s skill or CERN’s access? I bet many startups could revolutionize the medical industry but lack capital or access to patient data or access to customers (gate keeping by hospitals and insurance companies and regulations and so on).
Fwiw, ChatGPT can give reasonable cancer prevention advice which should always be the focus for most of us. Also, tech behind ChatGPT can be used to do a survey of a specific cancer's main research using various RAG based tools or using my free software such as https://github.com/impredicative/newssurvey
Why not both? I think Claude is probably already more useful and helpful than many doctors. It would be nice to one day be able to just chat with an AI bot from home and get prescriptions or test sheets delivered. You could add automatic safeguards for certain medications or for drug interactions.
Obviously we're not there yet but many people in the world have access to no medical care or only expensive and crappy medical care.
> chat with an AI bot from home and get prescriptions or test sheets delivered
That would be a glorious day because most doctors are spectacularly bad at it. It will never happen though because the medical establishment is very protective, and will never allow it if using one's health insurance.
I'm not sure how much of it is hype targeted at potential shareholders vs. reality but Moderna claims that they've incorporated ChatGPT into their business processes where they're using it to, among other things, predict optimal therapeutic dosages for their drugs
Unfortunately, while AI has some helpful uses like this, it will be used for trivial and even worse, nefarious purposes as well. And I do think the nefarious will outweigh the good for sure, but those who develop it and sell it to the highest bidder will undoubtedly point to stuff like this to use the emotions of the general populace to twist their perceptions to ignore the nefarious.
A lot of technology gets entrenched using such mechanisms even though its benefits are questionable. Genetic engineering, what could go wrong? But think of the sick children?
Not knocking what these people are trying to do, but I do think that if you add up all the positives and negatives of AI, the negatives far outweigh the positives.
Projects get funding just like at any other public-funded science institutions: with a lot of scrutiny and advocacy, if you think these 153'000 CHF ($178'000):
26 comments:
Reading in between the lines it seems most of the innovation is managing the regulations and privacy concerns of large institutions who want to use shared AI models.
> offers hospitals its experience in managing huge amounts of data in a secure and decentralized way, which is key to ensuring the privacy and security of the private patient information used to feed the algorithm. [...] uses a system to process data locally without sending it to central storage. This helps protect privacy and make better use of resources when different hospitals work together to create reliable AI-based models
Which probably makes sense why CERN developed some expertise in this area.
It feel silly to say that AI is curing cancer. Normally a phrase like that would sinal the apex of the hype cycle but I guess it actually has some meat in this situation. Using AI more like statistical inference to screen for medical conditions or predict treatment could be helpful. I remember Jeremy Howard from fast.ai did that with deep learning to detect things in medical images. Seems like a good thing for CERN to do as long as it works.
There is a venerable pattern in academic circles of branding your research with the language of whatever hype cycle is currently active. The same research can be branded under big data, computer vision, AI, or whatever else the current trends demand.
I can’t grasp what this is about. It starts saying it enables hospital to store and process their data onsite (not really an innovation in any way) but then later says the hospital computers still contact a main server for analysis. So is this just on premise data analysis (“ai”) but also cloud?
I understood it as this: Hospitals keep the data locally and run AI analysis on-site. They then exchange certain parameters (not the raw data) with a central server, which uses these parameters to update and improve the model. The central server doesn't run the models or process the data itself, it's just used to refine the AI models that the hospitals are using locally.
Someone please correct me if I understood it wrong.
Sounds like they use a method called federated learning. I.e. each site can train a local model on their own data and then share model parameters to a central entity that aggregates several sites model parameters. Superior performance can then be achieved (and also raw training data between sites are not shared).
What is that overhead power and… other things… device in the image?
looks like https://www.hohenloher.com/en/products/fly-one/ this, though it looks as though the CERN images might be on a movable ceiling mounted track.
Seems a bit outside their wheelhouse, no?
model fitting and extrapolation from large data sets seems like it would be exactly their wheelhouse.
I always assumed ML was used in processing data from the LHC. I don't see any other way to deal with so much.
Things like boosted decision trees have been used in several instances, like this[1] or this[2]. Surely other methods as well.
[1]: https://indico.bnl.gov/event/10699/contributions/53933/attac...
[2]: https://inspirehep.net/literature/1665070
It is and has been for a while, but most of the more flashy and exciting developments in ML and AI don't have very much applicability to LHC event processing. To be able to state any kind of finding about some aspect of physics based on the scattering of particles in the accelerator and their decays in the detector, you need to take the background of all events and make multivariate discriminants on the data in order to enrich your signal as much as possible while throwing as little as possible away. This requires you to have a rigorous and verifiable statistical "paper trail" from start to finish, so you can say with confidence intervals how much signal and background you ought to have, vs how much you measure in your data after processing it. An overly broad black box doesn't really work for this kind of introspection.
You mean like the web was also?
Good point!
Considering LLMs are getting larger each generation, CERN is a great place to investigate their applications.
The data captured, stored, processed, and evaluated after just one of their more notable experiments generate petabytes of data.
They know a thing or two that directly apply.
Is this really about CERN’s skill or CERN’s access? I bet many startups could revolutionize the medical industry but lack capital or access to patient data or access to customers (gate keeping by hospitals and insurance companies and regulations and so on).
This is what AI should be doing instead of ChatGPT stuff.
Fwiw, ChatGPT can give reasonable cancer prevention advice which should always be the focus for most of us. Also, tech behind ChatGPT can be used to do a survey of a specific cancer's main research using various RAG based tools or using my free software such as https://github.com/impredicative/newssurvey
Why not both? I think Claude is probably already more useful and helpful than many doctors. It would be nice to one day be able to just chat with an AI bot from home and get prescriptions or test sheets delivered. You could add automatic safeguards for certain medications or for drug interactions.
Obviously we're not there yet but many people in the world have access to no medical care or only expensive and crappy medical care.
> chat with an AI bot from home and get prescriptions or test sheets delivered
That would be a glorious day because most doctors are spectacularly bad at it. It will never happen though because the medical establishment is very protective, and will never allow it if using one's health insurance.
Meanwhile, I have a corresponding Custom GPT that I find useful: https://chatgpt.com/g/g-Myvb8o0yb-medical-lab-tests-advisor
I'm not sure how much of it is hype targeted at potential shareholders vs. reality but Moderna claims that they've incorporated ChatGPT into their business processes where they're using it to, among other things, predict optimal therapeutic dosages for their drugs
https://investors.modernatx.com/news/news-details/2024/Moder...
Unfortunately, while AI has some helpful uses like this, it will be used for trivial and even worse, nefarious purposes as well. And I do think the nefarious will outweigh the good for sure, but those who develop it and sell it to the highest bidder will undoubtedly point to stuff like this to use the emotions of the general populace to twist their perceptions to ignore the nefarious.
A lot of technology gets entrenched using such mechanisms even though its benefits are questionable. Genetic engineering, what could go wrong? But think of the sick children?
Not knocking what these people are trying to do, but I do think that if you add up all the positives and negatives of AI, the negatives far outweigh the positives.
Makes me think CERN has too much money
Nope too little, this 1000x rather than 1% optimization and data collection on ad preferences of tech giants.
CERN's side projects gave us WWW for example, so you could write your comment.
Projects get funding just like at any other public-funded science institutions: with a lot of scrutiny and advocacy, if you think these 153'000 CHF ($178'000):
https://knowledgetransfer.web.cern.ch/kt-fund/projects/novel...
...is too much money, then I'm glad: you probably don't have cancer!