2. AI is shifting scientists do and what they value far faster than individual scientists—let along scientific institutions—are able to adjust and adapt their value systems. Even when they work well, AI tools disrupt the delicate balance of costs and rewards that structure scientific activity.
3. To get through the AI revolution intact (not to mention other crises such as the GOP's all-out war on science, knowledge, and expertise) we need to be able to view science from outside and reflect on how it is changing and how we might wish to engineer rather than passively accept those changes.
AI Devolution? In any case, this seems like a forthcoming book, or at least an essay.
Importantly, I think we should avoid furthering the misrepresentation of this phenom as AI. As your colleague points out in the “AI Con” (Bender & Hanna), the LLM’s are text-extrusion machines. The only intelligence involved is the decimation of our own, via an eco-disastrous parlor trick.
4. Science of Science, with its data-intensive descriptive approach is useful, but insufficient. Metascience, with its laser focus on a replication crisis in psychology, likewise. We need broader humanistic thinking about the social processes involved in the construction and valuation of knowledge.
STS. You’re talking about STS. Small but mighty. Also, a lot of modern media studies.
📌
5. Today I read a paper by @sabinaleonelli.bsky.social and Alexander Mussgnug that I think illustrates this point perfectly. philsci-archive.pitt.edu/24891/1/Phil...
📌
📌
6. These are the conversations that we need to be having if we want to be part of a science that continues to further our understanding of the physical world, rather than turning into a high-stakes parlor game wherein humans jockeys ride AIs to fame and (localized, limited) glory.
Thank you for this and I hope you are on one of our university's AI committees.
7. The paper looks at how researchers might hand off boring "drudge work" such as data collection and cleaning to AI agents, and what could (and IMO will) go wrong as they do so. Here's a key passage.
I assumed scientists would see the instances of lawyers and judges including fabricated legal citations, case details, etc within legal documents as glaring red flags against implementing AI, especially for the "boring" yet fundamental aspects, in research? Why risk your data integrity and results?
just finishing up a lecture on how choices made during the drudge work are among some of the most impactful ones on downstream analysis. Handing this off to AI would be catastrophic.
8. As I see it, because scientists themselves determine what outputs of science are valued (by what gets published, funded, drives promotion, etc.), when a new technology offers convenience, we are vulnerable to rewarding its use uncritically for many of the reasons brought up in the paper.
9. As a result, science can swing, pendulum-like, far past its equilibrium point, Researchers gravitate to the convenience technology, it gets rewarded out of proportion, and science stalls despite the fact that most practitioners think they are working in a time of unprecedented productivity.
10. I could go on and on. But instead, read the paper. And record this as my vote for having more philosophers, and fewer computer scientists, on the Provost's Expert Committee For AI Futures and Institutional Destiny.
Ok, I've read it. I deleted seven versions of comments. Think all I want to actually say is that I love you, crow man. And I'd rather not argue about the paper.
Aw man, I hate when I have to read the paper. I guess I have to read it sober too? Oppressive.
Looks interesting--I clicked through and was immediately offered an AI summary of the paper, because of course I was.
I am graduating as a biology major and philosophy minor this December in part because I think science and philosophy need each other. Also, they are both really cool subjects, especially when they overlap.
I can't think of a worse idea than this. My experience is that philosophers have little understanding of basic knowledge such as the theory of computation, and hence come to wildly wrong conclusions about AI. 1/2
As Frances Crick wrote, "Philosophers have had such a poor record over the last two thousand years that they would do better to show a certain modesty rather than the lofty superiority that they usually display."
Well Paul Thagard is a huge exception OTOH I’m still waiting to read David Chalmers’ “Could a Large Language Model be Conscious?” because I’m waiting for a time when I need a good laugh
📌 Philosophy of science addressing changes (and destruction) of the scientific enterprise by bad tools (LLMs), horrible policies, and disinformation. Will return after reading the linked paper
And for AI & ethics, I'd strongly recommend @shannonvallor.bsky.social, en.wikipedia.org/wiki/Shannon... years ago, when she was still at Santa Clara U, she gave a great talk for Board at @computerhistory.bsky.social.
Interesting thread. Will read the paper, though I’d take issue with your characterisation of metascience as fixated on replication. While that may be true of some corners, there are broader, historically-rooted perspectives (eg this from RoRI assets.publishing.service.gov.uk/media/685bcd...)
PECFAFID around and find out.
📌
Well, scientists may not be determining what gets funded any more. It's quite clear that Vought/Bhattacharya et al. are determined to destroy peer review.
But AI is not the first technology that comes to mind in this context. Isn’t that what happened in genomics, during the human genome project, with the mechanisation and computerisation of sequencing? I remember sequencing ‘by hand’. Man, that was tedious.
Human beings had done the hard work of transcription, error correcting, gap filling, etc. They then encoded that specific knowledge into algorithms to successfully speed the process up. AI offers the promise of convenience so that human beings don't have to do the hard graft first.
I don't think that anyone would argue that things that increase productivity are necessarily damaging, only that our incentives mean that we risk letting things that are damaging through if they are seen as improving productivity.
If people aren't interrogating why and how it is "drudge work," which is integral, becomes incidental as soon as it is seen as "feminine" (when the labor is feminized), can there be a meaningful, affirmative shift in science's "inquisitional culture" that obviates the need for Convenience AI?
Nice! Just published a closely related essay. yohanjohn.substack.com/p/to-model-b...
There is an old saying "Everybody believes the data, except the engineer who collected the data"*. I wouldn't be comfortable with pure AI-agent data collection. * The second part is "Nobody believes the theory, except the engineer who formulated the theory".
Christ, it's so easy (as a human) to make mistakes in quality control and cleaning of data. And it's certainly not as easy as setting parameters and everything outside is 'false' - that's how you miss black swan events. You only realise this after you've done the work.
A tough one for sure. How can we leverage AI to assist, not to take over.
Ai is not taking over, because ai does not exist. The issue is not ai. It's the aggressive push of "faster, cheaper, go, go, go!!", which would use and abuse any tool. LLMs are a tool and I do think that they are not a tool that needs to be used that much out of a few niche areas...
I receive notices from organizations of science writers. It's normal for these to include universities, national labs, etc that advertise short meetings and educational opportunities usually about molecular biology, neuroscience, and similar. Lately, they are all about using and reporting on AI.
Which is being aggressively pushed by people and corporations for the buck.
REALLY aggressively!
Also important to consider that it's not just ease of use and convenience that's driving people to adopt these tools prematurely. Most of what I've seen among my colleagues is a fear of being left behind. Our department meetings are basically LLM show-and-tell for bragging rights at this point.
📌
📌
📌
Something I wonder about is if we're going to let AI be our self-imposed sophon (ala Three-Body Problem); I worry there are too many people (especially with commercial motive) willing to trade-off verifiability and reproducibility for speed, and that not enough people understand just how bad that is
Glad you liked our work!
The act of data collection is so important. My current research direction was sparked by something I noticed while collecting data for another project. It’s also how new scientists get trained!
If AI is used to help sort data and organize even stress test ideas, it’s possibly useful. If it instead is used to flatter notions or do the rigor of work and true intellectual examination, the laziness of it will out.
I suspect Latour's work is useful here because these issues are also about culture.
100%
How many universities are left that emphasize broad-based education in the humanities? How many public schools even address education beyond “basic skills?” There’s more than one reason we find ourselves here.
We must transcend the limitations of the scientific method: ocrampal.com/chasing-our-...
Thank you for these threads. You're posts like these one of the reasons I use social media.
