It's hard to imagine what kind of contortions someone would have to go through to come to this conclusion.
Not one word of copyright length, which is now effectively over a century long. Not one word of some of the most successful AI models being open source.
Commons encirclement, as the author is describing, has been happening since source code was labelled art instead of engineering and since lobbying efforts have been extending copyright terms. My view is that AI technology are trampling over the encirclement, which is why so many people are upset.
It's hard to imagine what kind of contortions someone would have to go through to come to this conclusion after reading (?) this paper.
Literal quote:
"Over the course of the twentieth century, the scope of human cre-
ations that were treated as IP expanded dramatically, a phenomenon that has been
called 'the second enclosure movement'."
Open-weight models aren't open source. Companies that release open-weight models do so because it currently suits their business interests. They can cut you off whenever they want or if they come under political pressure to do so. If they truly cared, they would release their training data and procedures and democratize access to their training infrastructure.
Well, they might also like a word with all of the scientists and students who make sure their papers are available online, despite Springer's draconic requirements. But that word is no longer heeded. Well, for the most part.
The discussion of open source was pretty anemic. The author seems to think open source is only GPL copyleft. In reality the vast majority of open source is permissively licensed, and the development was funded by for-profit companies. That seems like a crucial thing to consider when discussing how our information commons has come to be.
Near the end they include a paragraph claiming we should interpret this piece as being about more than self-interest and resistance to change, but it doesn't really feel like the piece offered a taste of what that "more" is.
The most thought-provoking part for me was considering that AI companies have used pretty much all available data for training while "they spend more and more money to create proprietary training data". That seems to break a sort of intuitive social contract - like, well you all used everyone's knowledge. Send yours back out too.
Grant from 3brown1blue released a video exploring the economics and mechanics that enable careers in math, in light of the AI company releases over the last couple weeks. Overall it was a lot more hopeful, and it didn't need to strain awkwardly like this piece to apply an ill-fitting historical framework like enclosure. Worth checking out: https://www.youtube.com/watch?v=LKrwhjOI8Nc
I have read more paper in the last year than the last ten years. Across a wide variety of subjects. While the AIs still need open access to science in order to function, they have make that open access something I can actually use like no other technology before it. Nothing has opened the scientific commons to me like AI.
There may be a concern going forward with locking things down. One can lament that AIs need to be prompted to "hit the literature" directly or they don't really think about it, prefering random web pages most of the time. But at least at the moment science is more open and accessible than ever before.
You have read more paper(s) or you have had an AI that is RLHFed to please the median consumer who reads at an 8th grade level summarize more papers than ever?
When you can equally reasonably make the opposite argument - that AI is going to open up scientific exploration and contribution to a wider range of participants than ever before - then I think it’s a sign that nobody has any idea what’s going to happen and spending time making either claim is probably pointless until we have more data.
Not necessarily, not in some of the most useful ways of simply having a significantly more capable way of automating research & experimentation of many sorts. Anyone with a dataset and a laptop with a decent gpu, or pretty reasonably priced API or subscription to access models that don't even need to be frontier level, they can setup a loop for intelligent iteration through countless analytical methods, subset analysis, enriching the data with other sources, search and retrieval of research literature to cross reference results of analysis with other findings and write a basic summary for the person to review, look for alternative methods for analysis.
Yes, not everyone will be able to launch a swarm of agents and a few billion tokens, but scientific research has always had areas where the most advanced tools were beyond the reach of every scientist who might like to have access.
Only for about five more years. In 7-10 years the price of memory and compute should drop by 80-90%. In ten years the bleeding edge 14,000 token/s machines and 4TB memory is going to look quaint, "that stuff can't even bootstrap a kernel and entire (usable) posix-compliant OS in less than a week" vs the future of "the computer updates it's firmware in real time to meet the needs of the current task"
But why would these same trends not also vastly improve the capabilities of the top hardwares and ability of the trillion-dollar companies to leverage their economies of scale?
I'm all for the idea that AI capabilities can be democratized and help ordinary scientists outside of a handful of labs. But it seems there would need to be some checks against Big Tech and it won't happen by just letting the free market do its thing.
The margins on GPUs are estimated at 80% right now. So the price would most likely drop by 70% just by building the additional manufacturing capacity. Assume some modest reductions in cost and the price should drop by 80-90%.
The only reason computers are so expensive is that demand is so out of control.
Crashed demand after one of the big AI labs (probably Anthropic because they focus way too hard on expensive frontier models) become insolvent and every trader panic sells tech stocks.
Even if that doesn't happen, a lot of non AI companies like Oracle that invested too much on data centers will fail and have to sell their hardware. Big models are getting too expensive for a lot of smaller businesses so demand migrates to less GPU hungry ones. That means Oracle might be left with excess capacity.
And we also get more supply because high margins attract investment. 7 years is a realistic time frame to build new DRAM fabs and packaging, it might take even less if governments treat it as a security issue.
Maybe but are those limitations more or less onerous than the current requirements to contribute to scientific research? It remains to be seen I think.
Not only resources being controlled by a few organizations, but the noise of all of the people who have used AI to “discover” something novel, when they don’t understand the domain in the first place. When everyone is discovering something, but what they are publishing is either incorrect or already known, how do you wade through the slop to find the gems (maybe academic publishing was already filled with human slop, though).
All journals will be brute-forced to death. Not the greatest time to be Phd student. And many people have no shame. They just use agents to make whole papers. Those who don’t, will double-guess their life choices.
> When you can equally reasonably make the opposite argument - that AI is going to open up scientific exploration and contribution to a wider range of participants than ever before
Is typing a prompt into ChatGPT "science"? It's an experiment, I guess, to find out whether you get a useful response out of it, and your hypothesis is that the model is able to answer your question correctly? How do you falsify that if you don't have the academic foundations to verify its answer?
ChatGPT is the research assistant, not the experiment subject itself. The science exploration is to use it to explain the existing science and help you design a good experiment, and then once you've run the experiment and gotten the data, as a chemist/biology/someone who hates coding, it'll write the code to analyze your data and make some cool visualizations.
One simple way to understand how this is currently manifesting is the administration's severe, and really unprecedented, defunding of basic science at NSF (likely in unconstitutional defiance of Congress' appropriations to specific NSF directorates). Instead the way the funding has been redirected (by the White House OSTP) is based on the assumption that future knowledge will be exclusively AI (sorry SI) generated, so humans not needed.
China, on the other hand, seems to be doubling down on fundamental research, including announcing these changes:
> Instead the way the funding has been redirected (by the White House OSTP) is based on the assumption that future knowledge will be exclusively AI (sorry SI) generated, so humans not needed.
Can you go into more detail about that? My impression is the real reason the Trump administration has defunded basic science (and so much else) is an unholy combination of "government bad" libertarianism and the belief the funding was going to their ideological opponents.
I feel that and use of AI to justify the redirection is most likely post-hoc.
Echoing what the other reply to your question said, I don't disagree that much of this post-hoc, but I think the thinking has evolved such that this post-hoc reasoning is now truly believed. I'm primarily speaking of it as an observer of how and where funds are being directed and to what ends: often big, non-university-centered programs to build labs or generate datasets for AI discovery. While some of the ideas may have merit, there is no free lunch unfortunately, as that was money that would have supported individual PIs and their trainees.
Going by the public announcements, the original plan was to cut science funding but once Congress said "no," the backup plan became directing it towards non - human - scientist items like facilities operations and the Genesis program.
This administration thinks science is 'woke' because it says things that disagree with him. Many of these funding cuts were explicitly politically motivated, e.g. they specifically targeted research about diversity, gender, or vaccines.
It's not about AI, it's about removing funding from liberal causes and giving it to conservative causes.
Unfortunately it is entirely related. The initial wave of cancelations were about those topics, but the more structural thing they are attempting now is to focus funds on tool development for AI models as opposed to human capital development as NSF does.
No it is about AI too. Trump et al want there to be no dispensing of federal grants to academics in the future. They want the government to instead cut checks directly to xAI et al who will do all our research for us in the future and will for certain represent our best interests and not those of shareholders exclusively.
If AI didn't exist, he'd find some other excuse to do the exact same thing.
The real reason for this is that Elon Musk is a conservative megadonor, while academics are vocally opposed to Trump. He takes from his enemies and gives to his friends; that's how he operates.
" an intangible resource that "no one owns and is shared by everyone."" ... but if you try to share this "un-owned" intangible resource with others you will have police all over the globe harassing and maybe even imprisoning you (see: Anna's Archive).
So ... "no one owns" is doing some pretty heavy lifting there ... and it's kind of the foundation of the entire article.
Say what you will, but a million scientists and a network of sometimes-miscalibrated institutions (publishers, funders, universities, etc)... that feels less owned than maybe 4 companies having the edge on every [scientific?] field in the next few years...
Most published research is controlled by 5 for-profit corporations.
Academic publishing is predominantly controlled by a small
oligopoly of five for-profit corporations that collectively
publish roughly 50% of all scientific articles: Elsevier
(owned by RELX Group), Springer Nature, Wiley (John
Wiley & Sons), Taylor & Francis (owned by Informa PLC),
and SAGE Publications.
It only feels "unowned" to academics because their institutions pay for access, which they can use. The general public mostly cannot.
Going forward, colleges can pay for ChatGPT and things like it, and presumably will. The AI companies will be happy to sell to them. There's really no problems here except for self-created ones (for example, if some researchers don't want to use AI). Of course, they're free to do that, just like they're free to not use computers. But then don't complain that doing stuff takes too long.
Those 5 corporations have an achilles heel: reviewers. If the AI-driven science future has its way, the reviewers at all the major journals will collapse under the weight of submissions. The effect will be a denial-of-service, not unlike the one experienced by open source projects in the era of AI-driven pull requests.
An alternative to journals would be to switch to the arXiv model everywhere: let anyone publish with a bare minimum of moderation, make papers explicitly versioned, and let peer review happen out in the open. Already for decades, the importance of a paper has been measured by citations, not publication in a prestigious journal.
I think you still run into the problem that writers will vastly outnumber readers, so that risks turning the project into a write-only dumping ground. From there you have the same old discovery issues, where traditionally scientists relied on reputation as a first stage filter.
Reputation is already a lot more about networking than a specific publishing system. Researchers get recognized by producing quality work and telling people about it in person and on the internet. Reputation spreads by word of mouth. I don't see that changing much in an open publishing world.
I think you missed the chicken and egg problem here: if you don’t have a good reputation, no one will read & verify your results. If 3rd party verification is the source of reputation, then how do you get started?
No thanks. I have no interest in peer reviewing anything for free if it's going to be handed over to closed source AI companies. And before some facile comment about Elsevier comes along, no I don't review for them either and it's clearly not the same.
You're thinking in terms of the traditional peer review model. I'm imagining a more organic process where papers are scrutinized to the extent that they're considered important or promising. For example, how Grigori Perelman's papers on arXiv about the Poincaré conjecture were reviewed by peers. There's no assignment of individual reviewers in this model, it's a natural byproduct of researchers discussing and collaborating in public.
Grigori Perelman's papers were considered promising mostly because they were written by Perelman. When there are too many papers to read, the reputation of the authors becomes the most important signal of quality. Formal peer review tries to spread the attention a bit more evenly, but it doesn't scale too well.
On the one hand yes, Perelman was known within his subfield. On the other hand, I find it hard to believe that if some random person had published the same papers, they would have gone unnoticed, since as far as I understand it these were important results that would have been worth evaluating based on the titles and the abstracts.
Volume isn't really a problem going forward, at least in mathematics, since every proof can be accompanied by a Lean verification. Whether other disciplines can implement automated verification or quality assessment will depend on the details of the field.
The more famous the problem, the more reputation matters. There is no shortage of papers claiming to prove P vs. NP in either direction. Nobody is taking them seriously, as everyone has a strong prior that people trying to publish such results are cranks.
Mathematicians need to rethink what is the purpose and value of doing pure mathematical research in the future. Until know, they have mostly valued the creation and systematization of human knowledge. From that perspective, Lean proofs mostly increase confidence in the human knowledge. But they don't contribute directly to it, as a formal proof is unlikely to contain insights that are not present in the human proof.
The author is coming from astrophysics where this is not the case. The major publications are operated from learned societies or public entities (RAS in UK, AAS in the US, ESO in Europe etc ...)
US public libraries generally have access for everyone (at least they have everywhere I've lived), and you can walk onto a ton of campuses as a visitor to get their access.
I'm more worried that >90% of the computers on the planet are controlled by a company that obeys the whims of an arguably-demented narcissist and oppose the rule of law.
Keep in mind if Alexandra Elbakyan shares the knowledge she is a criminal who needs to be extradited to face consequences. When you are a Californian AI lab that needs to train a model to give back all this static information to users, you are alright, actually you are the only thing keeping US economy going. Never be in out group its hard out there.
The academic literature has a problem in that it is relatively incoherent. Each new paper provides some small advance, with only a meager indication of how it fits into the known science in the area, its importance, and its consistency with previous results. It also remains in the corpus if later publications disprove its results or fail to replicate them. Only slowly do the aggregate of published results find their way into more systematic and complete authoritative texts.
AI promises to solve this problem by absorbing the entire corpus quickly and making a coherent whole available for further research and progress. It will be particularly useful if it can immediately flag any conflicts and inconsistencies, as well as any new relationships with other results not indicated in citations.
We need to create a new scientific regime in the spirit of open source where scientific papers are not permitted to be used to train closed models. If you want to use the output of human ingenuity, you need to give something back.
I don't really understand the rationale behind accepting change and knowing change is inevitable, but wanting any such change to be fully controlled every step of the way?
This is not that difficult. It's just like what happened to physical fitness/obesity after the invention of the car. Go look at countries that do it reasonably well (ie not the US). Copy exactly what they do, wholesale.
Don't let whiners pollute your mindset. Not even ones who package their insights in a nice scientific artifice like a white paper.
"all knowledge so far" (ie. the data needed for LLM training) is only a grain of sand on a beach compared to the return on making it all regurgitatable on demand.
We are at the bottom of a science, knowledge & art hockey stick chart, we're staring up at it and the steep bit is labeled "the future".
I agree it is terrifying because it is unknown. It is also true that the frontier AI labs have not worked well with affected communities at all and often downright exploited them.
BUT I also agree that, like any new technology, people will find mind-blowingly creative paths forwards with it, opening up whole new art forms.
It's not terrifying because it's unknown. It's terrifying because a tiny number of obscenely rich psychopaths are trying to take sole control of it so they can enshrine themselves as living gods and make 99% of humans so obsolete that they die sick and cold in the streets. That's their overt goal, and almost all evidence we have from history and current politics suggests that they will be wildly successful at this. None of this is for you, it's all for them. All of it. That's why it's terrifying.
> AI labs have not worked well with affected communities at all and often downright exploited them
The first reviews of OAI's dump on math are coming out. Quasi-Riemann Hypothesis Proof looks to be valid. The irony is that is uses and approach that the people who work in that area have disregarded for the most part. As well as mining some other parts of maths corpus. There really isnt "innovation" there.
For as much as OAI is guilty of leveraging work in progress, the complaints about some of the other work(s) are looking spurious (and I'm being very kind using that word).
I understand the all-knowledge so far types of references, but it also reveals to me that most people who say that actually have little to no idea that AI doesn't possess information that exists only in the heads of people which has never been written down and there's substantial value to it which day-to-day uses of AI are not able to reach.
There're two effects: first, the "envelope effect" wherein everything within the "knowledge envelope" of humanity becomes easier to "fill in", with more people making more connections possible faster among known things. Think cross-disciplinary insights.
And then there's a different effect, where a person or team can leverage AI to leap straight to the frontier of a field and then proceed outward from there.
Both are possible right now and available to anyone with a phone.
This is supposed to be The technology that will colonize the universe, and you're saying "Games are being remixed" is the happening we're supposed to be excited about? That's the hockey stick inflection point?
this approach of calling critics "whiners" then making sweeping, naive statements regarding both capital-S Science and also markets, is shallow.
The West collectively has touched markets with Science using concepts of property rights. All of that is upturned now. I agree that publishing portals like Elsivier have done damage to Science while enriching the company and its allies. However that complaint looks quaint compared to literally four military-industrial groups running all new discoveries and their commercialization.
Not one word of copyright length, which is now effectively over a century long. Not one word of some of the most successful AI models being open source.
Commons encirclement, as the author is describing, has been happening since source code was labelled art instead of engineering and since lobbying efforts have been extending copyright terms. My view is that AI technology are trampling over the encirclement, which is why so many people are upset.
Literal quote: "Over the course of the twentieth century, the scope of human cre- ations that were treated as IP expanded dramatically, a phenomenon that has been called 'the second enclosure movement'."
Open-weight models aren't open source. Companies that release open-weight models do so because it currently suits their business interests. They can cut you off whenever they want or if they come under political pressure to do so. If they truly cared, they would release their training data and procedures and democratize access to their training infrastructure.
I still disagree but it looks like the author makes a cursory attempt to address the issues that I think are more relevant.
Springer would like a word
Near the end they include a paragraph claiming we should interpret this piece as being about more than self-interest and resistance to change, but it doesn't really feel like the piece offered a taste of what that "more" is.
The most thought-provoking part for me was considering that AI companies have used pretty much all available data for training while "they spend more and more money to create proprietary training data". That seems to break a sort of intuitive social contract - like, well you all used everyone's knowledge. Send yours back out too.
Grant from 3brown1blue released a video exploring the economics and mechanics that enable careers in math, in light of the AI company releases over the last couple weeks. Overall it was a lot more hopeful, and it didn't need to strain awkwardly like this piece to apply an ill-fitting historical framework like enclosure. Worth checking out: https://www.youtube.com/watch?v=LKrwhjOI8Nc
There may be a concern going forward with locking things down. One can lament that AIs need to be prompted to "hit the literature" directly or they don't really think about it, prefering random web pages most of the time. But at least at the moment science is more open and accessible than ever before.
I've also read some summaries of even more papers, but yes, actually read papers, beyond the abstract.
Yes, not everyone will be able to launch a swarm of agents and a few billion tokens, but scientific research has always had areas where the most advanced tools were beyond the reach of every scientist who might like to have access.
What's different this time is that this situation is expanding to almost every field.
I'm all for the idea that AI capabilities can be democratized and help ordinary scientists outside of a handful of labs. But it seems there would need to be some checks against Big Tech and it won't happen by just letting the free market do its thing.
The mechanism through which this occurs being..?
The only reason computers are so expensive is that demand is so out of control.
Even if that doesn't happen, a lot of non AI companies like Oracle that invested too much on data centers will fail and have to sell their hardware. Big models are getting too expensive for a lot of smaller businesses so demand migrates to less GPU hungry ones. That means Oracle might be left with excess capacity.
And we also get more supply because high margins attract investment. 7 years is a realistic time frame to build new DRAM fabs and packaging, it might take even less if governments treat it as a security issue.
Is typing a prompt into ChatGPT "science"? It's an experiment, I guess, to find out whether you get a useful response out of it, and your hypothesis is that the model is able to answer your question correctly? How do you falsify that if you don't have the academic foundations to verify its answer?
China, on the other hand, seems to be doubling down on fundamental research, including announcing these changes:
https://www.globaltimes.cn/page/202609/1371583.shtml
and this: https://www.scmp.com/news/china/science/article/3368750/chin...
Can you go into more detail about that? My impression is the real reason the Trump administration has defunded basic science (and so much else) is an unholy combination of "government bad" libertarianism and the belief the funding was going to their ideological opponents.
I feel that and use of AI to justify the redirection is most likely post-hoc.
This administration thinks science is 'woke' because it says things that disagree with him. Many of these funding cuts were explicitly politically motivated, e.g. they specifically targeted research about diversity, gender, or vaccines.
It's not about AI, it's about removing funding from liberal causes and giving it to conservative causes.
The real reason for this is that Elon Musk is a conservative megadonor, while academics are vocally opposed to Trump. He takes from his enemies and gives to his friends; that's how he operates.
So ... "no one owns" is doing some pretty heavy lifting there ... and it's kind of the foundation of the entire article.
Going forward, colleges can pay for ChatGPT and things like it, and presumably will. The AI companies will be happy to sell to them. There's really no problems here except for self-created ones (for example, if some researchers don't want to use AI). Of course, they're free to do that, just like they're free to not use computers. But then don't complain that doing stuff takes too long.
How do you build reputation in that environment?
word of mouth and circle of trust that spread?
A good science should focus on verified and verifiable predictions.
No one will quote you if you don't have good reputation either.
> If 3rd party verification is the source of reputation, then how do you get started?
By surviving the check against the reality. Or doing 3rd party verification on other works.
Volume isn't really a problem going forward, at least in mathematics, since every proof can be accompanied by a Lean verification. Whether other disciplines can implement automated verification or quality assessment will depend on the details of the field.
Mathematicians need to rethink what is the purpose and value of doing pure mathematical research in the future. Until know, they have mostly valued the creation and systematization of human knowledge. From that perspective, Lean proofs mostly increase confidence in the human knowledge. But they don't contribute directly to it, as a formal proof is unlikely to contain insights that are not present in the human proof.
AI promises to solve this problem by absorbing the entire corpus quickly and making a coherent whole available for further research and progress. It will be particularly useful if it can immediately flag any conflicts and inconsistencies, as well as any new relationships with other results not indicated in citations.
https://www.youtube.com/watch?v=KPUlgSRn6e0 [Not Just Bikes]
UK has strong inheritance tax, so common angricultural land is very strong and doing quite well!
Also shit tons of stuff you can not do on your "private" agricultural land, like producing too many cow farts!
- Dario
"all knowledge so far" (ie. the data needed for LLM training) is only a grain of sand on a beach compared to the return on making it all regurgitatable on demand.
We are at the bottom of a science, knowledge & art hockey stick chart, we're staring up at it and the steep bit is labeled "the future".
BUT I also agree that, like any new technology, people will find mind-blowingly creative paths forwards with it, opening up whole new art forms.
The first reviews of OAI's dump on math are coming out. Quasi-Riemann Hypothesis Proof looks to be valid. The irony is that is uses and approach that the people who work in that area have disregarded for the most part. As well as mining some other parts of maths corpus. There really isnt "innovation" there.
For as much as OAI is guilty of leveraging work in progress, the complaints about some of the other work(s) are looking spurious (and I'm being very kind using that word).
Whether it meets human needs and meets them in a rich and satisfying way is another question.
And then there's a different effect, where a person or team can leverage AI to leap straight to the frontier of a field and then proceed outward from there.
Both are possible right now and available to anyone with a phone.
I've never seen so much faith outside of a religious setting. Do you have any proof of this at all, or is it a faith-based system?
Its been 4 _years_ since GPT 3.5's release. Where are all these marvels people keep saying are just around the corner?
It's happening all around but you're somehow not seeing it.
This is supposed to be The technology that will colonize the universe, and you're saying "Games are being remixed" is the happening we're supposed to be excited about? That's the hockey stick inflection point?
The West collectively has touched markets with Science using concepts of property rights. All of that is upturned now. I agree that publishing portals like Elsivier have done damage to Science while enriching the company and its allies. However that complaint looks quaint compared to literally four military-industrial groups running all new discoveries and their commercialization.
This is not resolved at all...
I'm advocating mindfully managing your own influences.