hckrnws
EU science advisers back call for a 'CERN for AI' to aid research
by occamschainsaw
I should be for this, because this is how I make my living in the most literal sense.
But to be honest, I am dismayed with how the ecosystem works in the EU. Billions upon billions of funding for research, some of it good and (let's face it) a lot of it not so good. But either way, no commercialization will result out of all this fundamental research. Any commercialisable results will be spun into startups in the US and China decades before any EU entity follows their path. The many scientists trained with this funding; many of them will end up in the US, and maybe of them will end up in European companies, but not in any position that would make them use any of their acquired skills.
At a time when the EU economy is falling behind so evidently, I'm frustrated that we have these people in power proposing the same old paradigms that don't really benefit the continent.
>Billions upon billions of funding for research
That's a hint at the problem. We spend a lot of taxpayer money to train some of the best researchers and scientists in the world, only for them to work for US or Chinese companies boosting their economy instead, because they pay more than the European companies.
Just look at Nvidia, Huawei and Apple, they have an R&D centers next to every top university in Europe or next to every ARM/Nokia/Ericsson office.
The problem isn't the money spent on training/education, it's the lack of top economic opportunities/companies we have in a lot of parts of Europe resulting in our tax money boosting the economies of our economic adversaries, who are less risk averse and more hungry about innovation and monetization.
The lack of top economic opportunities/companies is not strictly a EU problem, it's a everyone except China/US problem. Japanese workers were earning 2x as much as American ones in 1990. Now they're earning less than half.
I think this comes down to USA/China being big countries with a huge and homogenous consumer and stock market. This allows to scale fast to a point where you can simply crush the competition. The other small countries are all protecting whatever they already have to not be completely crushed, which makes it even harder for them to scale.
For example I am working for a successful startup in Belgium. We have a hard time entering our neighbour's market because of existing players, protectionist policies and just how different things are there, and so is it for them. Maybe with a huge capital we could make it happen, but in the end everybody knows we'll all get crushed by an American company, so nobody really wants to invest.
This could be solved with more EU integration, and with less economical protectionism at the national level so that national champions can leave place to bigger European ones. But everyone is blaming EU regulation for this situation and we're moving in the exact opposite direction.
>This could be solved with more EU integration, and with less economical protectionism at the national level
Absolutely no chance, especially on how right wing and nationalistic the EU elections are swinging lately. We'll just see more national protectionism. That's the problem with the EU. It's not one country where all the resources are pooled, like the US or China, but an alliance of several fiefdoms where each of them wants to be king and won't hesitate to backstab the rest (see abusing veto rights) to gain a perceived advantage with populists back home at the expense of everyone else's monetary disadvantage. The constant political squabbling between members is bittern than tooling up and going to war like in the old days, but it's holding us back compared to US and China.
>For example I am working for a successful startup in Belgium.
Would you recommend moving to Belgium for a career in SW development as a non Speaker of Dutch or French? I'm not looking for a fancy big-tech career or to change the world or make insane wealth but just to work in a leisurely environment with no stress and enjoy life.
I'm not sure it is more leisurely than other countries, and the winters are hard due to lack of sunlight. OTOH Brussels is very international, english is perfectly fine for work and life. It is also quite a fun city if you like the vibe and make the effort to dig under the surface. Antwerp / Ghent are less international and speaking dutch would be more important.
Actually I don't like Brussels that much, too crowded, dirty and overpriced for me (sorry if I offended anyone). I liked Ghent and Antwerp far more when I visited, felt more chill and laid back, easier to get places by cycling and walking, while being cleaner and less rough.
Don't know the tech market or how the population feels about non-Dutch/French speaking local. I would of course try to learn the language and integrate but it will take me a while.
No offense taken, I don't think you'll have any issue, and the tech market is relatively good
I agree that the problem is not entirely to be blamed on the respective governments. The lack of private investment is a fundamental part of the puzzle. I have been trying to figure out why private capital loves to take huge risks in the US but is so financially conservative in Europe and I am not yet satisfied with the answers I get.
Well for one, Europe is conservative because it's more into protecting the wealth of the old gentry, the old families who used to run the continent, rather than disrupting it to build new wealth for others, therefore is a lot more risk adverse and instead just rentseeks and monetizes what they have like tourism and valuable real estate .
And secondly, the US can afford to outspend Europe at VC funding since it's not spending actual earned money from taxes, blood sweat and tears, but "monopoly money" they could keep printing for free as they own the credit card for the world reserve currency.
> The many scientists trained with this funding; many of them will end up in the US, and maybe of them will end up in European companies, but not in any position that would make them use any of their acquired skills.
This is (my simplified version of) Zimbabwe's problem[0]. They heavily invested in education, which is fantastic, but there was no economy to absorb their talents into, so the educated youth left for South Africa and onwards.
As James Carville would say on this topic: "It's the economy, stupid."
Poland was dealing with similar brain drain problems, but now that economic opportunities are there educated people are returning.
That's awesome.
While your comment on the bureaucracy of the EU is reasonable and realistic, the comment on the use (or lack of it) of most science has very little to do with Europe.
I've spent literal years implementing the newest natural language processing papers in practice and over 90% of them cannot be replicated or are replicable but the results are fabrications (generally statistical lies or other types of misrepresentation). The text of the paper is almost never in line with the actual numbers and the numbers are not in line with reality.
Perhaps ironically I did feel like I was right to leave the academic world. I wouldn't like to professionally wade through the intellectual equivalent of shit for decades on end. A few years were more than enough.
Did you publish any of it? Your work, as painful as it sounds, is very valuable. While some papers do not replicate, more authors than ever release papers, code and models together.
Nowadays, conferences such as ECIR (European Conference on Information Retrieval) have reproducability tracks that seek such work, and reward your kind of efforts with getting a peer reviewed publication. The results are often not "can be" or "cannot be" replicated, but more nuanced, and often interactions with the original authors are needed as space in publications is limited.
On a related note, giving a paper to a Ph.D. student to replicate is a pretty good warm-up exercise to get started, specially when combined with writing a survey article on a field (so that the week has a balance between "reading and doing"!
I cannot publish it, it has been done under NDA. It's nice of you to say so, I think you may be the first person to tell me this.
My field is different, but as someone from the industry side that keeps up with our counterpart in academia....most of what is being published is worthless. It serves no purpose other than to inflate publishing statistics. It has a lot of relevant keywords, but nothing I can actually use in industry to solve the problem the paper reports to.
For example, there will be a paper on how to fix X and I'll get really excited, but then realize that the idea shows that the person fundamentally doesn't understand the field and then they'll apply some obscure algorithm that is vastly inferior to the state of the art 20 years ago on a tiny model and then claim success.
> At a time when the EU economy is falling behind so evidently
Keep in mind that a lot of the US' current growth comes from its huge budget deficit. Europe has gotten significantly better at the commercialization aspect of research, two out of the three covid vaccines came out of Europe.
> Keep in mind that a lot of the US' current growth comes from its huge budget deficit
I don't see that.
Most innovation in the US is from private companies and private investment.
Government spending is mostly wasteful, and on things that have little to do with growth, such as social security, medicare, military ...
https://www.nationalpriorities.org/budget-basics/federal-bud...
Infrastructure in the US is crumbling, education is a disaster (esp. college costs - crippling), medical costs out of control, housing costs out of control ... It's amazing that there is at least some surface appearance of the economy being healthy, but it's really a sham.
> two out of the three covid vaccines came out of Europe
Moderna and Novavax Covid vaccines are entirely American. Pfizer-BioNTech vaccine was researched in Germany, but productized by Pfizer which is an American company. AstraZeneca vaccine is from UK, so yes in Europe but not in EU. Janssen vaccine was developed in Netherlands and Belgium by Janssen Vaccines, which is a subsidiary of Johnson & Johnson, again an American company.
Also, Trump's Operation Warp Speed gave $1.2 billion to AstraZeneca and $1 billion to Jansen, so America funded even European vaccine development more than Europe did.
> Europe has gotten significantly better at the commercialization aspect of research, two out of the three covid vaccines came out of Europe.
This fallacy made me jump out of my seat.
I provided salient examples to underline a counterclaim to what the OP said. Where is this fallacious?
If you want a broader overview you could also start with the world innovation index which reserves its top spot to the country hosting the CERN.
You can use fancy numbers to pretend that Europe is relevant. Just like France touts its obsolete military and is constantly called on its bluff.
The reality can be seen around you. A few high profile projects doesnt change how things actually are.
Comment was deleted :(
(1) AI:
What you said is all correct, but there are some positive steps as well, for example ELLIS: https://ellis.eu/
ELLIS is a great way to compete with the top US universities. Much more is needed, especially access to computing power, just as the CERN for AI proposal suggests.
(2) EU research:
EU research funding itself is a success:
https://erc.europa.eu/news-events/news/statement-erc-scienti...
Difference to CERN would be that no company is financing high energy physics while AI is frenzily funded.
But getting new models and insights (in the public domain) is a good idea.
Yes, but just as you need commercial stuff you also need research, and those are different things.
I think something like this will be great. It'll be fantastic to go there for a conference, it'll be fantastic to work there for two years even if you eventually go into industry, etc.
I think the problem for commercialisation is the cost of DL accelerators and I think this is what creates the enormous VC reliance, and this in turn leads to many firms being in the US. The EU probably needs some effort to bring DL accelerators closer to the semiconductor production cost-- it shouldn't be 40x what TSMC charges to make the chips, it should be maybe 2x or 3x. I don't know if that's realistic, but considering that chips are so expensive that the electricity costs aren't a big fraction of the cost anymore, I think even badly designed chips, if you get them cheaply, will do.
I think there should be a centralised location though, or well two. One for summer and one for winter, one here in Sweden, the other in Southern France or in Spain.
Didn't big hospitals start to get their own accelerators ? (Or what do you mean by "DL"-oh, Deep Learning, my bad, in context I thought it was something like Direct Linear particle accelerators...)
> no company is financing high energy physics
Medical Technology (eg. Siemens Healthcare, GE Healthcare, Abbott, Medtronics, Phillips) all sponsor high energy physics due to it's applications in medical imaging and some nuclear therapies.
HPC firms like Nvidia, AMD, Intel, Huawei, etc are also major hirers due to the applications of simulations in machine learning (and vice versa). Most ML innovations were sparked out of research that arose from nuclear simulation research after the NTBT was enforced.
And finally, ML R&D companies themselves have funded research in adjacent spaces of High Energy Physics that also had concurrent applications in the ML space.
The issue with CERN (and much of Europe's R&D ecosystem) is organizational.
Great point. I think the next couple years will see private money dry up for all but a few companies, unless there are further massive breakthroughs. A publically funded group would be great to incubate talent in the EU and help keep AI research free and open.
That is because nobody expects HEP to produce commercially viable products. People do expect that of AI. I think it sounds like a great idea, both to stop a bit of AI brain drain to the US, as well as to develop AI for things that aren't immediately profitable to businesses.
Well, nobody expected it, but by funding HEP we got the Web.
That's a very good reason why CERN produced a lot of things that lead to many viable products at the end. The R&D work is usually a rich source of knowledge in itself.
See, the AI funding frenzy is about extracting much more money from it in the future. IMHO it's like the 2000s internet when everything was funded by the VC and the internet was amazing, only to turn into this money extraction racket of a few apps and websites dominating %90 of everything with no fun left. They banked on rent seeking by locking down everything, to the point that nothing new is being made in SV anymore and the only new big thing in that space came from China(TikTok) and Russia(Telegram).
Would suck badly if the people who fund the current AI revolution decide that it's time to recoup their investment. It's not that they are bad people, it's just that they are not doing it as a charity.
Also, OpenAI's rationalisation for switching from non-profit to for-profit revolves around the idea that AI research is very expensive because it requires huge amount of hardware and it's not feasible to raise the required capital for this as a non-profit.
There's the solution for OpenAI becoming open: CERN for AI. Money is provided with expectation of open research.
Yeah, it will be interesting to see how they are going to avoid an "OpenAI" situation.
Unfortunately, EU research money nowadays is free money for private companies. How this works: a 3-year research project with a consortium of academia+industry of 2-3 universities (1M/each) and 4-5 private companies(50k-200k).
With this money, universities do fund PhD students but with a stipend/salary of an average software engineer. In contrast, the "industry partners" do the bare minimum work a couple of days before the quarterly review. And they use 20% of that money to sub-contract their work packages to freelancers (who, btw are paid better than the PhD students) and the other 80% to cover normal OpEx.
And you know... the "work packages" can be something like 5K-10K for a bootstrap/hugo website.
Edit: yes, I over-simplify things a lot but that's the general modus operandi.
I'm not sure how it is for the companies, but as universities involved in EU projects, the amount of bureaucracy is just overwhelming and there are so many nonsensical requirements, that you don't get to do any interesting work. Overall, it barely seems worth it.
There already is the infra for a 'CERN for AI' - it's called CERN.
CERN is already a major leader in the HPC space, but whereas HPC and High Energy Physics labs like LLBL, ANL, NSC Guangzhou already pivoted to ML and precision medicine research in the 2010s, CERN member states weren't nimble enough to support additional pivots within CERN.
Not so fast, not so fast - at first whole, brand new bureaucratic department (or two) needs to be created to properly govern whole process, when that is ready then work on regulations framework can be started - there is one already but for sure that cannot be sufficient enough because every corner case needs to be properly accounted for. Do no forget about required government overlook, registrations, permissions for future usages of this dangerous new tech, all needs to be in place.
CERN is a special place, unless there is support to build a similar institution, it won't happen.
By the way, CERN is already exploring AI for the new accelerator.
Comment was deleted :(
If you want a CERN for AI you need to build a fab for affordable graphics cards with good software support and high vram.
Or you tackle a commercially interesting problem first and fund the research to tape out more efficient chips in cooperation with ASML and NXP.
Haven't they changed their Name to NERDS recently? ;)
https://home.cern/news/news/cern/cern-change-name-70th-anniv...
> The efforts should focus not only on physical and life sciences, but also on humanities and social sciences
:)
> and they should ensure that AI systems are aligned with “European values”, the scientific advisors say.
:(
Something like this would really need to be coupled with creating much more viable markets for investment in AI (and associated technology) in Europe.
Large-scale R&D could be great, but applications of that research should have a good opportunity for creating wealth in Europe.
With the viable markets you wouldn't need nearly so much investment. People can make money from this; they'll invest.
Yes, but unless you fear state funding crowding out public funding, having some state sponsored research doesn't strike me as wrong (could be more basic science, different directions, etc.) - even if there is great investment market.
I don't fear it; I just don't think it's necessary. Just as I don't fear state funded lawnmower manufacturing.
If Europe wants to finally put the Capital Markets Union and some associated bits in place properly it might indeed not be needed.
More defense spending in Europe is probably also creating some new funding opportunities.
That may be useful if paired with a comprehensive industrial policy to boost the domestic semiconductor industry. Otherwise it'd just be a funnel to American companies for EU taxpayers' money.
That industrial policy is already here: https://commission.europa.eu/strategy-and-policy/priorities-...
I should have added "effective" policy, then ;)
How much bigger is CERN than other circular accelerators?
How big would a "AI CERN" server farm need to be relative to the biggest ones in the world?
It cost 6 billion to build the current collider (in an existing tunnel) and CERN has a budget of 1-2 billion per year. In terms of data they are easily producing 100x of previous colliders.
Will it be able to generate an image of the most common misspelling of the large hadron collider?
Almost certainly as we don't have the same fear of our own bodies this side of the Atlantic, e.g. spinning advert cube near my apartment has "Dildo King" on one face.
It'd be amazing to have an institute for publishing completely open models and weights - especially now the Biden administration seems to be going full AI doomer against Open Source models and weights with the recent US AI Safety Institute appointment.
While some regulations might be beneficial, it’s sad that the opinion of an 80-year-old on nascent technology carries so much weight…
> Biden administration seems to be going full AI doomer against Open Source models and weights with the recent US AI Safety Institute appointment.
I haven't been following the US scene, can you elaborate? The last time I looked, tech celebrities like Paul Graham and Alex Karp were dissing Europe for regulating AI saying that Europe will miss out on AI. Did the USA regulate the AI too?
Just a lot of talk.
IIRC big tech companies an OpenAI are lobbying to make making foundational models very hard for non big tech companies.
Comment was deleted :(
[flagged]
Crafted by Rajat
Source Code