AI boosts research careers but narrow the span of ideas explored: study(spectrum.ieee.org) |
AI boosts research careers but narrow the span of ideas explored: study(spectrum.ieee.org) |
To me this effect doesn’t seem to reflect on AI very much, it seems to reflect on humans. Like maybe this is more evidence of the Babble Hypothesis and the incentives in research than AI, no?
the last decade of US politics demonstrates just how powerful willingness to produce put strips all other critical skills.
AI exacerbates this and exposes fundamental human heuristic frailty.
'It’s not about the architecture per se,' Evans says. 'It’s about the incentives.'"
If this same article had said “adoption of word processors over typewriters increases publishing”, would we still be thinking of the babble hypothesis?
Given the way in which AI is currently used in publishing, it is altogether way too early to label it counterproductive in the research creativity department.
It would have been useful to check whether less original work was already getting more citations before AI adoption. That could reflect broader trends and network effects: heavily cited research areas attract more authors optimizing for citations, so high-productivity researchers end up clustering on the same topics.
But also just thinking about your point for one second: in your mind, how else would they argue for the conclusion if not by checking the trend over time? Like what is the precise implication here?
They do this in various ways, like establishing paper pipelines, collecting rents on labs and committees, focusing on money layer, using their profiles and citation count to help with acceptance of papers of other people , etc. You talk to them and they can’t explain their papers beyond a superficial introduction.
They collect huge citations, travel and give talk on the winner horses, collect credit, which feeds back into this fraudulent scheme. A scientist used to be a scientist not long ago, not a credit collector.
I wonder if Google could invent a new metric to expose them (weak ratio of first authorship, etc).
It's a game of cultural tribalism. The only thing worse for one than not engaging is to upset the status quo unblessed.
Unfortunately, "volume of stuff" is how governments allocate funding, for stuff like education, healthcare, research, etc
Any kind of quality metrics, outside of the free market capitalist "how much profit you made?", will be gamed within a few years to get the same results, for things that aren't designed to turn a profit, since there's no fair and objective metric in that case.
Earlier studies looked at the effect of web search on research. This is all covered near the beginning of the article.
We tend to think that obvious potential is the same as realized potential, for new technology.
For any specific context, there are generally innumerable smaller adaptations and capability thresholds that have to be crossed. And the price for that journey is often temporary loss off overt productivity.
I see it as an overfitting problem. Fundamentally, the topic here seems to be that citation indices and similar metrics are actually flawed indicators, and obsessing over them is just Goodhart's law in action. Ultimately, the argument is that the entire design of those metrics is wrong. To be precise, it was a good metric at first, but now that the scale has changed, it's become bad. This is common in programming too—things that are correct in the beginning but become problematic as they grow larger.
From an individual researcher's perspective, it's rational. You get more citations, your career accelerates. Everyone knows this. Paper counts aren't everything. Citation counts aren't everything. Journal impact factors aren't everything. You shouldn't only play it safe. But everything is tied to those metrics anyway.
Most researchers who give me work are fully aware of these facts. But are they going to change anything? Funding is still distributed based on those metrics.
Max Planck said, 'Science advances one funeral at a time.' Science doesn't progress purely through reasoned argument. The authority of the older generation, research funding networks, journals, and school-specific evaluation criteria all move together.
And honestly, I think discoveries will keep happening—probably quite rapidly. Because AI doesn't have the factional conflicts or interpersonal issues that humans do. It's very good at connecting papers across schools of thought without bias. In other words, the current human system is flawed at consolidating research, but I think AI is actually strong in this area. I expect AI-driven discoveries will continue for some time. The people who ride this wave will clearly be the winners.
Everyone knows things are broken, but no one is trying to fix them. I always think human society is inefficient. I read this post, but I'm more curious about who will actually lead the improvement effort.
I like LLM's but this writing style is like eating the same dish 4 times a day.
Please feel free to disagree with me! I am keen to hear more anecdotes to get more datapoints.
Funny enough, it basically says the exact same thing about software engineering that TFA says about science:
First - developer level productivity has improved
…
Second - overall system flow has slowed down at every stepI think the flattening of progress is the most interesting dimension to the article. For an example a useful biological product discovery with a nonlinear path to get to there, look at the Taq polymerase (https://en.wikipedia.org/wiki/Taq_polymerase). Without some NSF funded exploratory ecological research by Tom Brock in Yellowstone Hot Springs to test the theoretical limit of life at high temperatures (https://en.wikipedia.org/wiki/Thermus_aquaticus) we never get to the Taq polymerase, we never get reliable/robust PCR (https://en.wikipedia.org/wiki/Polymerase_chain_reaction), which is now a gold standard method in both clinical and environmental testing! It is rather improbable to think that large language models would associate those domain connections across the topic (molecular biotechnology + ecology + microbial physiology). I also did some exploratory work with text embedding models people might use for RAG and challenged them with an open source scientific MCA question dataset, generalist embedders performed worse vs. domain specific embedders trained on scientific corpora (doesn't surprise me at all). However, if everything regresses to the median of the universe of possible knowledge, it seems like scientific leaning frontier models would get locked into this asymptotic flattening before turning cashflow positive for model vendors OR they become so locked down that only big pharma, state actors, or big ag can afford the API rates and vetting process.
Crowded trades and crowded research share the same root. The reward is easier to measure than the thing it proxies for. Returns are easier to measure than durable edge, and citations are easier to measure than discovery, so everyone digs the same hole deeper. Which leaves the study’s real question. Are citations measuring discovery at all, or just the crowd?
By definition, creativity cannot be automated, and AI is a fantastic automation machine. It can explore thinking paths at a rate humans cannot match. But creativity is bringing the unthinkable into the thinkable, and that requires sensory experience [1]. Specifically, new definitions and symbols which never existed before. Imagine the concept vector space, and expanding that with new independent dimensions. Is that even possible ? When you look at history the answer is yes !. And each time there was an independent dimension added, it was an act of genius. It is an instructive exercise to name these moments in history where an independent dimension was added to human thought. Some examples in math would be the invention of a number, and in politics could be the idea of democracy. By contrast, LLMs are trapped in the vector space they are trained on, and they lack the feedback loop with sensory experience to be able to create and validate theories.
[1] https://philsci-archive.pitt.edu/28024/1/Scientific_Inventio...
it would be funny if by accelerating the enterprise it actually forced an effort to correct the trajectory.
It's really, _really_ high time we dispensed with the idea that this is "AI". Nobody said they're not useful, but "AI" they are not.
AIs do things no human has done before millions of times a day.
LLMs are aggressively trained to reproduce facts and consequently struggle to reject orthodoxy. There isn't any reason they can't, in principal, make big new discoveries just by getting lucky, which is sort of also how humans do it, but its ok to acknowledge that current AIs aren't so good at certain things.
https://news.ycombinator.com/item?id=48863490
LLMs don't just 'average' their data.
I have a hard time believing that all novel concepts yet to be discovered are contained within that space, though.
The aim of many scientists is discovery, publishing is a side chore to survive and to get funding. Automate paperwork and you get more time for discovering.
Disclosure: Physicist.
I was also in academics myself up to the Master's level (research track), and personally had to deal with the politics of getting support for what I wanted to work on; that experience helped to discourage me from going on to a PhD, as I'd rather have proper leeway to work on what I really prefer and take avenues I find interesting.
many children have an unlimited capacity to ask "why?". many adults are the same
if the abilities of AI are finite, then we will continue to have burning curiosity, questions to ask, and discoveries to make
I said as much - https://news.ycombinator.com/item?id=48870324 - prepare to be downvoted.
> its output can be novel or good, but rarely both at the same time.
> rarely
That is not a viewpoint they can't do something useful and new.
With that criteria, he could be talking about anyone.
I find it rare that people critiquing AI today, actually hold people to the same standards. Or are as enthusiastic about referencing ways machines keep surpassing us, as for ways they have not yet, when speaking about limits for progress.
Well, these AI are never going to die in any real sense, so expect them to make orthodoxy more sticky, not less.
All the factional conflicts are in there, and there are also plenty of reports of people getting weird / toxic / passive aggressive responses from AI.
Because the model is trained with everything, you can in principle get anything out of it. You want to get an answer based on all the right things, while keeping all the wrong things suppressed. But it's easy to get something less than ideal, due to the specifics of training, harnesses, context, prompts etc.
AI-written comment?
Honestly, it feels a bit racist. If you just look at the content, it's clearly not a comment that AI could have written—but just because I used a few em-dashes and have limited vocabulary and formal phrasing, I get called AI and downvoted. It feels like they're saying non-native speakers should just go away. And the downvotes come without any counterargument to the actual content. Does this feel a bit more human now?
When we solve problems we usually follow a heuristically guided energy efficient path. We just prune a lot of possibilities based on our existing knowledge and experience.
Creativity happens when we consciously (or not) go off the beaten path and explore. Most of those explorations are dead ends. But some will yield unexpected connections, patterns etc that we call “creativity” .
An AI system could also go on those kinds of explorations. Today they aren’t it because we are not asking them to.
A lot of the time people state the kind of fundamental limitations of LLMs very confidently when it feels like it is too early for people to really know. Like we are already well past the point where where LLMs are just pre trains on the internet with some RLHF for chatbot… Most of the effort is spent on elaborate reinforcement learning.
Is it unconceivable that future generations of LLMs could be RL’d to use einsteins visual method for theories [1] with the right tooling and geometry representations? Or just something random like that.
[1]. https://www.visualscribing.com/blog/2019-11-11-einstein-on-v...
> The emergence of physically consistent World Models offers a pathway to a synthetic laboratory. By enabling agents to run counterfactual simulations—to experience the physical consequences of a thought experiment—we may finally mechanize the feedback loop between intuition and logic.
LLMs are fundamentally limited by their architecture to only return a token predicted by a statistical inference, essentially lossy decompression.
It's like arguing that taking an image, compressing it with JPEG and low quality, then decompressing it into something blurry with some random color values thrown in is creating new art.
No one is arguing that everything a human has created is good. No one is arguing that LLMs can't be useful.
Sutton is arguing that it can't be novel. Cherry picking a couple of words doesn't change his argument, which is very clear.
I presume you are an expert in some field. Think carefully about the boundary of the field and all the subtlety and complexity of that boundary and all the oversimplification you do to communicate that stuff to lay people. AI is, in some large sense, directed at all lay people, not experts, and even if we wanted to direct it at experts, at the edges of knowledge, there really isn't a lot of training data for that. Mathematics is a sort of exception because it has very clear validation criteria which makes RF particularly easy for it.
So if the speed of propogation of EM waves is the same no matter your frame of reference (along with all the rest of physics) then the speed of light can't be relative (a conclusion that was aided by the Michaelson-Morley experiment) and what are the logical consequences of that.
If I'm incorrect in my understanding I'd appreciate any correction.
Once the Pythagorean theorem was proposed, many different proofs have been identified. In art, once a new style is created it's often straightforward for others to replicate. In physics, the idea of Relativity was what enabled the design of experiments to demonstrate its correctness. Proposing the idea is what's essential.
There are really a lot of people like you who don't respect non-native speakers. You seem to enjoy mocking native English speakers around you, don't you?
You saying it's "You're AI", but I'm genuinely not an AI—I don't understand why you won't accept that.
'To be honest' just maps directly from '솔직하게 말해서','솔직히','까놓고','그리고 사실','명확히하자면','사실' etc in Korean, my native language. There are many phrases that map directly like that. There aren't many connecting words. There just aren't that many direct mappings.
You must ask every foreigner who asks you for directions if they're an AI.Now that I'm reacting this aggressively, are you happy because you think I'm not an AI
I don't really get this sentiment. You can be in a competitive environment, but it means nothing if the incentives are skewed.
Ever heard of "a race to the bottom"? That's a competitive environment too, after all, and no one thinks it's positive when applied to humans.
This seems like some variant of "why don't you short the market and become rich". It doesn't work like that.
Should be interesting to see what happens to the programming profession when there isn't anyone around anymore who actually knows programming.
I work in industry. In that case, nobody who meets me would ever know that I have patents. I would consider them to be a useful add-on for my resume should I ever need one, but it doesn't define me. And what my employer chooses to fund is a matter between consenting adults.
This was also my default thinking, but we really see more and more "nerds" getting into the spotlight. It could be a kind of self-fulfilling situation though: the ones working for that fame make the choices that get them there, such as opting to do research that's more "palatable" to those holding the purse strings, and so have the support to gain and maintain presence. Those who would rather blaze their own path generally get left in obscurity (unless they find something truly game-changing), even if it turns out they're more than the former group.
But a generalization that science work is motivated by fame and fortune just doesn't make sense.
lol how old are these people? You have better chance at fame and glory if you started a stupid YouTube channel.
The first type happens when you are enthusiastically engaged in a topic, which LLMs will likely enhance.
The second type happens as a by-product of solving a, perhaps deeply uncomfortably, difficult problem. This is what people are talking about when they say LLMs will hamper human cognition. Instead of sitting there for an hour and struggling, people will instead reflexively give in and ask an LLM to solve it for them.
I think in most cases, understanding is the point. we don't expect students to derive general relativity before doing astrophysics. re-invention is only a tool for understanding
The flip side is even more interesting. There’s a great number of electrical engineers with significant physics backgrounds who don’t really understand how electricity actually works, but they can still solve useful problems. By understanding I mean they can describe what underlying physical phenomena reactance represents etc.
When the child is able to go to YouTube and find a tutorial rather than having to puzzle it out, yes, it absolute does. We've seen this for decades now.
you didn't re-invent it, you learned it from a teacher and then practiced until you understood it to the required degree
If you can't derive a result in mathematics or physics from first principles you fundamentally do not understand it.