Best Amazon Prime Day deals: Mumsnet favourites

Best Amazon Prime Day deals:
Mumsnet favourites

Shop now

Please or to access all these features

University staff common room

This board is for university-based professionals. Find discussions about A Levels and universities on our Further education forum.

AI use in humanities PhDs

49 replies

NutsAndMay · 10/12/2025 08:58

How are you handling this? We are seeing AI all the time in proposals, and consider it a reason to turn them down. But I’ve recently found a funded PhD student using large swatches of Ai (re-)written material in a thesis. The institution doesn’t really seem to have a process or policy on this. As a humanities scholar I think it’s ludicrous and totally unacceptable! What do you think? And what do you do?

OP posts:
IntrinsicWorth · 10/12/2025 23:41

NutsAndMay · 10/12/2025 08:58

How are you handling this? We are seeing AI all the time in proposals, and consider it a reason to turn them down. But I’ve recently found a funded PhD student using large swatches of Ai (re-)written material in a thesis. The institution doesn’t really seem to have a process or policy on this. As a humanities scholar I think it’s ludicrous and totally unacceptable! What do you think? And what do you do?

Can you not just judge it on the quality of the arguments and content?

Made up references, poor arguments/ poor verbal reasoning, would be a fail. If you know your field very well, just trust your own instincts.

Despite all the hype, general AI like Copilot, ChatGPT, is REALLY bad at specificity, currency and nuance, and also at producing truly novel theories and thoughts. It’s like a giant scraping device and needs human input to not be a zero sum self-affirming technology.

Crispynoodle · 10/12/2025 23:45

It’s difficult to prove/detect AI but they need to be able to answer questions on their work

Bungle2168 · 10/12/2025 23:49

@catontheironingboard “If you know your field, and you want to produce really new, original work not rehashed mediocre stuff, AI is actually a timewaster rather than a useful tool. My job is to produce original thought and research that no-one has produced before. I’m not going to do that by using AI, which by design just recombines all the mediocrity that’s out there already, in order to produce yet more mediocrity.”

I hear what you are saying, but producing entirely original thought sounds fanciful in practice. No one researches in a vacuum and almost all academic output is iterative: Einstein built upon Newton et al. Read any APA paper for the incessant name dropping of prior researchers!

Original research is a myth.

catontheironingboard · 11/12/2025 00:11

Bungle2168 · 10/12/2025 23:49

@catontheironingboard “If you know your field, and you want to produce really new, original work not rehashed mediocre stuff, AI is actually a timewaster rather than a useful tool. My job is to produce original thought and research that no-one has produced before. I’m not going to do that by using AI, which by design just recombines all the mediocrity that’s out there already, in order to produce yet more mediocrity.”

I hear what you are saying, but producing entirely original thought sounds fanciful in practice. No one researches in a vacuum and almost all academic output is iterative: Einstein built upon Newton et al. Read any APA paper for the incessant name dropping of prior researchers!

Original research is a myth.

Are you in the humanities? (Einstein is irrelevant here.) Mediocre research can be iterative, but it’s still meant to produce new thought and new material. That’s the entire point. That’s what distinguishes really good research from mere “output”.

If you’re happy to produce mediocre research “outputs” then fine. But really good research in the humanities? Not likely. How is an AI going to find a new manuscript annotation in a rare book it can’t access or read, let alone decipher whose handwriting it is in? Is it going to be in the library looking at handwritten papers and letters, trawl through archival papers and objects, produce new theories over broad time periods by synthesising hundreds or thousands of different kinds of ideas and material, much of which is undigitised and will probably remain so?

It can’t even summarise my students’ 1500 word essays correctly in a bullet-pointed plan; of course it can’t do original research.

IntrinsicWorth · 11/12/2025 00:12

catontheironingboard · 10/12/2025 23:27

😆 hah! It’s the quantitative sciences and software/tech industries that are most at risk from AI - not the humanities. Are you under the impression that AI can actually think?

Parial agree - in that I think AI is loads better at pattern recognition and quantitative extrapolation than humans. The immediate gains in AI will be in quantitative disciplines but I’m not sure what will happen if it exhausts human left field thought, and if people stop correcting it: even though it is adequate at coding, for example, it cannot do novel creative computational thinking. It’s just computing at the end of the day. It can learn linguistic patterns but a lot of the seeming knowledge it regurgitates is just a version of stuff humans have already written or thought.

It is insanely bad at feeling, empathy, intuition and non-verbal communication. Basically, the things that make us human.

catontheironingboard · 11/12/2025 00:30

It can learn linguistic patterns but a lot of the seeming knowledge it regurgitates is just a version of stuff humans have already written or thought.

Well, in essence it’s just a very fast recombinative search and predictive text engine. If you want it to produce a mediocre version of a structure or a rehashed version of material that’s already out there, it can do that. (Even though a lot of the time AI summaries also misrepresent the sources they apparently make use of.)

As you say, what it can’t do is the stuff that we’re meant to be doing in humanities research - all the human stuff. If one’s idea of research is just reading a load of digitised papers, then writing a combination of those back into something “iterative”, then that isn’t usually good humanities research (though it might well qualify as an “output” - that is, if someone actually wants to publish it.)

But AI sitting for years in libraries or archives or an author’s papers or parish record offices, patiently building up a picture of an author’s works, or of a historical trend, or archival material on a topic, or working on a book about a big cultural shift over time? Nah. AI isn’t going to help anyone publish a big humanities book with OUP or Yale or MIT, the kind that really advances a discipline; or to produce new work on someone or something not yet studied or digitised; or to look at a lot of critical or historical material and see trends nobody’s seen yet; or produce a new interpretation of a text; or think up a topic nobody’s looked at before, and so on. And you only become skilled at research and at a discipline by doing it yourself. AI risks deskilling researchers in all fields so that they haven’t any longer got the experience or knowledge to know when it’s wrong, or to think in a nuanced way about what it isn’t able to see or do.

IntrinsicWorth · 11/12/2025 00:37

catontheironingboard · 11/12/2025 00:30

It can learn linguistic patterns but a lot of the seeming knowledge it regurgitates is just a version of stuff humans have already written or thought.

Well, in essence it’s just a very fast recombinative search and predictive text engine. If you want it to produce a mediocre version of a structure or a rehashed version of material that’s already out there, it can do that. (Even though a lot of the time AI summaries also misrepresent the sources they apparently make use of.)

As you say, what it can’t do is the stuff that we’re meant to be doing in humanities research - all the human stuff. If one’s idea of research is just reading a load of digitised papers, then writing a combination of those back into something “iterative”, then that isn’t usually good humanities research (though it might well qualify as an “output” - that is, if someone actually wants to publish it.)

But AI sitting for years in libraries or archives or an author’s papers or parish record offices, patiently building up a picture of an author’s works, or of a historical trend, or archival material on a topic, or working on a book about a big cultural shift over time? Nah. AI isn’t going to help anyone publish a big humanities book with OUP or Yale or MIT, the kind that really advances a discipline; or to produce new work on someone or something not yet studied or digitised; or to look at a lot of critical or historical material and see trends nobody’s seen yet; or produce a new interpretation of a text; or think up a topic nobody’s looked at before, and so on. And you only become skilled at research and at a discipline by doing it yourself. AI risks deskilling researchers in all fields so that they haven’t any longer got the experience or knowledge to know when it’s wrong, or to think in a nuanced way about what it isn’t able to see or do.

Edited

Yeah, I agree in part.

But academics are stubborn, proud, tenacious individuals and genAI Is one of many emporors’ new clothes situations. No-one is yet going to wholly AI universities to get AI degrees based on AI “research”, and your AI-dependent PhD candidate will likely fail for ethical/ integrity/ lack of graft reasons.

Bungle2168 · 11/12/2025 01:11

@catontheironingboard My broader point is that you - or any researcher worth his or her salt - will have some command over existing literature. This will, naturally, influence different aspects of your own research: thematic, methodological, epistemic, and so on. It is the reason bibliographies exist.

Could you be a historian and claim complete ignorance of Herodotus? Of course not. Barthes encountered Propp; Derrida, Levi-Straus. De Beauvoir spawned generations of Feminists. Much of left wing academia is doctrinally aligned with Gramsci if not Marx himself.

You are correct that AI cannot (yet) access non-digitized sources or do primary research. But are we really saying that the value of research lies only in obscurity?

Researchers pursue activities that will attract monies from funding bodies. As a consequence no money = no research, however interesting or valuable the topic may be. That is the dirty secret of academia, and idealism plays no role whatsoever.

catontheironingboard · 11/12/2025 01:40

Bungle2168 · 11/12/2025 01:11

@catontheironingboard My broader point is that you - or any researcher worth his or her salt - will have some command over existing literature. This will, naturally, influence different aspects of your own research: thematic, methodological, epistemic, and so on. It is the reason bibliographies exist.

Could you be a historian and claim complete ignorance of Herodotus? Of course not. Barthes encountered Propp; Derrida, Levi-Straus. De Beauvoir spawned generations of Feminists. Much of left wing academia is doctrinally aligned with Gramsci if not Marx himself.

You are correct that AI cannot (yet) access non-digitized sources or do primary research. But are we really saying that the value of research lies only in obscurity?

Researchers pursue activities that will attract monies from funding bodies. As a consequence no money = no research, however interesting or valuable the topic may be. That is the dirty secret of academia, and idealism plays no role whatsoever.

But AI is not actually reading or knowing anything. It recombines patterns. It doesn’t actually know or understand anything. If you want it to summarise Kant, it simply searches and rehashes other people’s summaries of Kant. It has no actual understanding of the original text. It’s not thought, and it’s not a real mind. It’s just a simulation of it - plausible linguistic patterns.

I think it’s actually quite possible that real skills and real knowledge disciplines, especially niche ones that take a long time to learn, get crowded out by AI slop, like an academic version of dead internet theory. It will most certainly be used as a way of defunding large parts of the university and academia, getting rid of entire disciplines, and deskilling large parts of the general workforce.

Only in the last decade all the talk was of upskilling and the knowledge economy. Today, the likely future looks more like the past: a very few people in an elite who siphon off the profits from AI; and a population who are systematically deskilled into “meat suit” manual and low-skill jobs, which can be paid very little but provide a convenient labour source for the elite. Goodbye to the Enlightenment, the era of mass higher education, culture and any ideals beyond a crude commercialism. Hello to the corporate drone world where AI produces endless tasteless fake slop for other bots to consume, while our kids are all told that plumbing and being healthcare assistants following AI commands are the only jobs out there, because anything else is simply programmed up by an LLM instead of a person.

We’d do well to resist the creeping tyranny of AI slop while we can, or it will render large parts of the internet unusable, and every bit of general life ugly, nonsensical and commercial.

QBTheRoundestOfBees · 11/12/2025 07:03

Bungle2168 · 10/12/2025 23:49

@catontheironingboard “If you know your field, and you want to produce really new, original work not rehashed mediocre stuff, AI is actually a timewaster rather than a useful tool. My job is to produce original thought and research that no-one has produced before. I’m not going to do that by using AI, which by design just recombines all the mediocrity that’s out there already, in order to produce yet more mediocrity.”

I hear what you are saying, but producing entirely original thought sounds fanciful in practice. No one researches in a vacuum and almost all academic output is iterative: Einstein built upon Newton et al. Read any APA paper for the incessant name dropping of prior researchers!

Original research is a myth.

Come on, researchers need to demonstrate that they know the state of the field and what their original contribution is.

LilyCanna · 11/12/2025 07:30

Because AI doesn’t actually think or ‘know’ anything, it also doesn’t ‘know’ when it is making stuff up. It is notorious for, for example, generating references made up of non-existent research papers. This isn’t a question of “Can it trawl through handwritten documents in libraries?” but being asked to search online databases accurately, and it fails. As a non-expert I’d think that ‘if no data exists, don’t fabricate it’ would be an easy instruction to insert into these programs’ fundamental workings, but hey, what do I know?
And then as the proportion of AI-generated material on the internet increases, these AI-generated errors will be fed back into the input of AI generation…

Bungle2168 · 11/12/2025 07:34

QBTheRoundestOfBees · 11/12/2025 07:03

Come on, researchers need to demonstrate that they know the state of the field and what their original contribution is.

All research is derivative of what came before; it is merely a question of degree.

Alltheprettyseahorses · 11/12/2025 08:49

Students using AI aren't doing the work and imo don't deserve their degree. How can they deserve it if their 3 years are spent telling a chatbot to give them 3000 words on a topic? They won't have learned a single thing and will be a massive disappointment at best to their first employer who will be expecting a fairly intelligent graduate and will get, well...

You should never outsource your brain. It destroys what it is to be human.

QBTheRoundestOfBees · 11/12/2025 09:14

Bungle2168 · 11/12/2025 07:34

All research is derivative of what came before; it is merely a question of degree.

Yes, as I said, you need to show contribution to the field, which requires knowing your field and where you are making an original point of whatever kind, even if it is taking a fresh approach to existing knowledge. I don’t think I am arguing that research is not derivative, but researchers do have to offer something new and original to be published. I am not sure how you are understanding original here but never mind.

tinydynamine · 11/12/2025 09:27

When I'm feeling down about my life at university, I ask ChatGPT to come up with a rant about my gripe du jour...either in the style of Slavoj Żiżek, Donald Trump or Dame Edna. Never fails to make me laugh.

Ramblethroughthebrambles · 11/12/2025 14:06

I'm surprised the OP's institution doesn't yet have a policy on this. But whilst most institutions still have PhD learning outcomes along the lines of ability to summarize complex material & original contribution to knowledge then I can't see how it can be acceptable to have AI writing even part of the thesis. However, learning outcomes might look different in the future e.g. 'critically review and adapt AI output on previous research to develop novel research questions'. I'm assuming current LLM approaches to AI will hit a dead end because of the limitations other posters mention. We should probably assume though that there will be future AI that is better at simulating human understanding. We'll then need to completely rethink the purpose of education - will we be teaching people to develop knowledge with or without the assistance of AI tools? Perhaps one day we'll have two kinds of degrees - with and without use of AI and the latter will be seen as an elite kind of education.

NutsAndMay · 11/12/2025 23:22

@Ramblethroughthebrambles There is policy for undergraduate assessment, at an institution level, and at a discipline level we prohibit the use of GenAI on every assignment brief. But we don’t (yet) have any policy at PhD level. We need it. I also wouldn’t want to examine a PhD at another institution without one, now I realise we can’t rely on good conscience or even shared views across the sector.

OP posts:
goingoingone · 11/12/2025 23:57

This reply has been deleted

This has been deleted by MNHQ for breaking our Talk Guidelines.

Acinonyx2 · 12/12/2025 09:07

Having a policy and being able to meaningfully implement consequences are two very different things unfortunately. Many places have students sign a statement about what AI they have used (if any) and how exactly. It's not hard to 'eye ball' and see when it's been over-used but it's next to impossible to prove it in any quantifiably appropriate way. This is what we are finding. At least PhD students have a viva but many of our masters don't. We did have a case that we tried to fight - and after many months back and forth - we had to let it pass. And if we start insisting on vivas all round, students can claim anxiety etc. It's also unprecedented to have a 'good' thesis then fail on the viva. We're still trying to figure out a solution and in the meantime have gone back to handwritten in person exams.

As for what AI 'knows' well, what do any of us 'know' really.....?

ParmaVioletTea · 12/12/2025 10:43

surreygirly · 10/12/2025 09:22

With the proliferation of AI the more important question is what is the point of doing a humanities subject
Whilst interesting they will not offer people job opportunities within 5 years
AI is sucking up jobs every day now
Students need to do science or medical based or learn a trade

Tell me you know nothing about knowledge without telling me ...

PotolKimchi · 12/12/2025 12:28

@catontheironingboard Sorry I think I agree with you on this- apologies if that wasn't clear. I don't think you can 'critique' AI or AI generated work without deep knowledge yourself. This is what I bring to the table- to my research and to my students. And although yes, research is based on work that is produced in the past, original research does exist, and it is not possible to do this without having deep critical thinking skills.

NutsAndMay · 12/12/2025 21:21

This reply has been deleted

This has been deleted by MNHQ for breaking our Talk Guidelines.

Well there are the obvious things - the dashes without a space, the typical sentence structures - and then there’s just the ‘voice’ of it, which I find stands out a mile once you get the sense of what you’re looking for. AI writes stuff that sounds fancy but doesn’t really say much, and definitely doesn’t manage to analyse a source. I suppose the applicants (and students) I’m thinking of who are resorting to AI are not the strong candidates, and so they don’t do anything/enough to cover their tracks. Conceivably there might be middling applicants (and students) who use it and we can’t tell…

OP posts:
NutsAndMay · 12/12/2025 21:24

And no, I wouldn’t rely on AI detection - I find that very variable. And it’s powered by AI 🤨 Can we be 100% certain, probably not. But with a PhD applicant I’m not sure it matters. If the proposal isn’t good enough it’s not going to progress, whether that’s because it’s AI or because it’s just not good.

OP posts:
aridapricot · 13/12/2025 11:48

I have a PhD student who recently submitted a draft with plainly made up references and a Methodology section which didn't match what had been approved by the Ethics committee and which was in fact wildly out of line with common practice in the field.
Luckily, I was able to argue it was assessed work (the student had submitted it as part of one of periodic reviews they have).
I say "luckily" because my university will only investigate AI cases if they happen in assessed work. Initially they even tried to argue back that the piece in question was not "assessed" and hence no official investigation would be conducted. Which brings me to the question - if it's only "assessed work" that is a problem, and if only work submitted for periodic reviews at PGR level (which in my university happen annually) count as "assessed", then a student could be submitted AI-generated work for the best part of one year without anything being done?
The advice I got initially was to have a "supportive" conversation with the student to make sure they understood what had done wrong... about half of your references were made up and you don't know why that's wrong, really?!

New posts on this thread. Refresh page