Alexander Kustov

Thanks for sharing your experience as a historian!

OK, folks, real talk. I get many people here don't like AI, and they may have good reasons. Me announcing I used AI for writing wasn't the best way to start a conversation.

I was fine taking personal dunks for it. But tagging my employer to fire me or threatening my family is where I draw the line

I agree. I'm much more pessimistic about the AI impact on teaching than research :(

Thanks for this point, this makes sense!

Granted, we may disagree on the exact percentile. But, as I mention in the piece, I'm not just talking about US scholars at R1s but social scientists globally.

Thanks, Jeff. Given that many folks here seem to deny AI can do any useful stuff, I'll take your critique as a big concession.

No, AI can't better identify important questions yet, but it can generate hypotheses and other novel ideas better than many scholars.

Thanks, I appreciate it. I'd be curious to know which points you're skeptical of the most.

Btw, I'm quite uncertain about some of them too.

Right. Though, to be fair, I didn't (mean to) lump them. I explicitly said "social science research" which, from my perspective, by definition excludes humanities (some edge cases aside).

Count me (postdoc, Brown, HIV prevention / digital mental health) among those who generally agrees with these points. The sophistication, nuance, and domain awareness one must bring to these tools’ capabilities will look different for each of us. But candor and collaboration are the way forward.

That's correct. My choice set is getting an AI subscription or--similarly priced--few hours of undergrad RA work out of pocket. Hope this clarifies things.

4/ Academics hold AI to absurd double standards. We criticize AI hallucinations while tolerating p-hacking, non-replicable findings, and data errors in peer-reviewed work.

But we also know that very few published papers are genuinely useful. AI is held to a standard we never applied to ourselves.

Sorry, just to clarify, I'm not provided with any funded RAs as a part of my position. I'm also not required to mentor folks who are not my students. I can only apply for grants or pay for RAs out of pocket. So, what are you saying exactly I should do??

academia very badly needs people who are neither dogmatically anti-AI nor credulous boosters. the potential is real and the problems are very real. but many critics refuse to even engage with the question of the how and when the things work in the first place.

Thanks for a very direct response to my actual question. I appreciate it a lot!

I myself am uncertain about some of these, so it'd be indeed nice if people started talking about it seriously.

hmmm. I think I take issue with and disagree with 1,4,7, and 16. but other than that I agree with most of what this thread says.

OK, sir, are you going to pay for my RAs? I hope you do realize this a paid job we're talking about, right?

This is a good post. Highly recommend it as a companion to my earlier writing regardless of where you personally stand on the issue.

Can AI Replace Social Science Researchers? No. No it can't. Come on, now.

New post: Can AI Replace Social Science Researchers? (No. No it can't. Come on, now.)

davekarpf.beehiiv.com/p/can-ai-rep...

Thanks Ben, I agree!

Academics Need to Wake Up on AI Ten theses for folks who haven't noticed the ground shifting under their feet

Full arguments, evidence, and links in the original pieces:

Part I: alexanderkustov.substack.com/p/academics-...

Part II: alexanderkustov.substack.com/p/academics-...

Agree or disagree, I'd rather have the constructive conversation and pushback than the knee-jerk fight.

20/ Research can lack "soul" and still serve the public. Most academic research is publicly funded. Taxpayers fund universities to produce knowledge, not for professors to self-actualize.

So, yes, you should access costs and benefits, but be open to using those tools if they produce better results.

19/ I want to be very careful care. Academic Bsky has not been a productive venue for this debate with professional threats and "ai/dr" pile-ons over a substantive disagreement. But it can be better.

The real cost: grad students and junior scholars watching this learn to keep their mouths shut.

18/ AI detectors do not work. My Claude-generated post passed every major detector as "100% human."

Disclosure norms also sound reasonable but likely select for dishonesty in practice. I disclosed my AI use for my post and got hundreds of personal threats. So, nobody would do it in equilibrium.

17/ Skill atrophy is a real risk—especially for the next generation of scholars. Outsourcing source evaluation, literature reviews, and data coding can undermine deep understanding. For established researchers, the risk is low.

For students, we urgently need to figure things out.

16/ AI exposes what was already broken. "If AI can do your research, your research was never good." I agree—but that is an indictment of social science, not a defense against AI or smart attack against me.

The replication crisis alongside papers that nobody reads were all pre-existing conditions.

15/ Most published papers are never cited or read. So, it's probably already mostly read by AI, not humans. That is, your primary audience is increasingly LLMs.

With agentic tools, it applies to most academics now. Ensuring your work is machine-readable is a good first step.

14/ Publication lag makes AI capability critiques structurally obsolete. Citing a 2025 paper on GPT-4 limitations to argue against AI in March 2026 is like citing a 2005 study on flip phones to argue against smartphones (which is not to say that smartphones are good for you).

13/ User expertise still vastly determines output quality. Agentic AI is decidedly different from copy-pasting from a chatbot. It's about executing productive tasks well from scratch.

Saying "anyone could do this" if something is AI-produced is like saying anyone with a stove can cook a great meal.

12/ AI's "jagged frontier" explains the polarization. Superhuman at some tasks, embarrassingly bad at others. Critics point to the troughs, enthusiasts to the peaks. Both are right about their corner. Very few people hold both truths at once.

11/ Qualitative research will increase in relative value. If AI can synthesize literature and run regressions, the premium shifts to what it cannot do: fieldwork, interviews, archival work—generating new data from hard-to-reach contexts that did not previously exist.