The Philosophers' Cocoon

A safe and supportive forum for philosophers.

Owned & Moderated by Marcus Arvan (University of Tampa)

Contact: marvan@ut.edu

Blog mission & moderation policy

IMPORTANT NOTE on how to ensure anonymity in commenting
Job-Market Reporting Thread (2025-26)
Job-Market Discussion Thread (2026-27)

recent posts

about

How are you (not?) using AI?

Now that it’s mid-summer, I’ve been in the grip of doing research but also starting to think about the fall semester. Basically, I’ve been doing things the old-fashioned way, not using AI in research or teaching (as I have a ton of concerns about it)–and my sense, at least anecdotally, is that many in the humanities are pretty resistant to it too.

That said, I am interested to hear whether people are using AI in research and/or teaching, and if so how. (I’m also interested to hear from those who aren’t!).

Any readers willing to share?

Posted in , ,

35 responses to “How are you (not?) using AI?”

  1. Anonymous

    Thanks for raising this question. I am not using AI in research or writing. I enjoy writing and I would not want to relinquish control over the process. To be clear, I struggle writing sometimes, that is, I struggle getting my point clear. But working through this is how I get clear about my own ideas and views. I am not so confident that I would readily identify an idea or view as my own were it obtained with the aid of AI (LLM, in particular).

    Reply
  2. Anonymous

    I use AI basically as Google on steroids. I find it very useful for finding papers on very niche topics, or for answering questions of the sort, “has any one defended X and Y together?” It’s not always perfectly reliable, but it’s useful for this purpose.

    I occasionally tried running philosophical ideas with AI, but I found these conversations useless. I don’t use it for any of the actual writing (I like my own writing style better than the AI’s).

    Reply
    1. Anonymous

      Which AI do you use for finding papers?

      Reply
      1. Anonymous

        Mostly Grok.

  3. Anonymous

    I have sometimes used AI to get a summary of a book to see if it is close enough in topic rather than just title to what I am interested in engaging with to be worth reading, or if I skimmed it a bit to confirm that I captured the main points in my head. Other than that I am not using it. To use it in writing seems false to me– it is important to me that my writing is mine– my voice, my reflections. And I learn to think by writing.

    Reply
  4. Anonymous

    I send the AI a full sentence I wrote sometimes to ask if I used a certain English phrase correctly (or as ChatGPT says, if it’s “idiomatic”). I find this very helpful. I also ask for a list of synonyms for certain words sometimes.

    I tried using AI for research to find sources a few times but it was always very disappointing.

    Reply
  5. Michel

    Not.

    Reply
  6. Anonymous

    I recall using AI on two distinct occasions. One was to suggest different writing styles of sentences from a paper I submitted to an essay competition. The other was to translate SEP articles paragraph by paragraph (that I would proceed and check) to undergrad/grad students from my country. Since most of my writing process takes place on Overleaf and Goodnotes and the resources I value on them do not rely on AI, I simply do not use it nor do I feel any necessity to. If I want to become acquainted with a topic, I go to SEP/IEP or Phil Compass. If I wanna know if a book fits my purposes, I look for book reviews or go to relevant chapters to take a look.

    Reply
  7. Anonymous

    I use ChatGPT to find relevant papers to read. I do it at the end of a writing process — I read enough papers/develop enough interest in something to draft a paper of it, then inevitably along the way I say some things that I feel have probably been discussed before. Then I use ChatGPT to help me find what those things are, because I find it hard to use Google to do that accurately. After some back and forth (‘those papers discuss X, but the view I’m discussing is really Y’), GPT will come up with a decent list of papers, and I’ll skim them all and often find a few that actually are relevant to my paper, and then I’ll read those. This has been a lot faster for me than using google or asking around.

    I also use ChatGPT to find typos. But I don’t use it for style editing or anything else. I’m worried that using it for any kind of editing other than typos will help contribute to something I’m noticing, which is that a lot of people are starting to sound more like AI in their writing (even people who I think don’t use AI.) We see too many AI-written or edited things around and it influences how we write ourselves.

    I’m a little worried about even feeding it my papers to find typos, because it has now read enough of my papers to start imitating my writing style. Actually I find it much more pleasant to read than the usual AI syntax/etc. But it also creeps me out. Overall, worth it, because I have a lot of typos, and I feel embarrassed when I ask humans for feedback and then realize afterwards my paper is typo-ridden.

    Reply
  8. Anonymous

    I have a lot of anxiety about writing and have some difficulty with organization so, I use AI as a kind of writing coach. It stores a complicated outline for my book project, and being able to “talk” to Claude about the points I want to make in each chapter helps me to organize the sections, try out different options in revision, etc. I sometimes ask Claude to summarize my writing as well, again to help me stay organized, and although I never prefer its formulations to my own, sometimes the way that it presents my ideas prompts me to look for a clearer way of articulating them. I also find that making the writing process a little more interactive–checking in with “someone”–helps emotionally. Basically, I’m experimenting with using AI as a tool to assist with what I find particularly difficult about writing, and have been finding it helpful.

    Reply
  9. Anonymous

    It seems that I use AI a bit more than the other commentors. When I start working on a new paper, I will frequently hand-write the argument until I come up with something precise and defensible. Then I will identify areas where I need to do more research: sociological claims, ready-made defenses, etc.

    I will then type my stuff, and plug it into ChatGPT Pro with a full set of instructions to help me identify papers and books that I need to read. I will ask for Chat to give me “must-read” works, “probably read” works, and “probably do not read” things. And I say that each paper or book needs to include a link. I then do the readings myself. Often, it’s pretty bad at identifying what I need, but you just have to use the tool discerningly—obviously.

    The worst / most annoying part about researching is trying to find appropriate sociological claims. Many of my arguments rely on me noticing that people frequently don’t recognize that their premises / concepts have particular unwanted implications, and so I need to find instances where individuals rely on these things but don’t recognize the implication. AI is extraordinarily helpful in this. In particular, there are many different ways that philosophers talk about similar concepts, and sometimes I don’t remember all of these when I ctrl-F a book or paper to find the central concept. Because LLMs have statistical correlations among concepts, Chat may identify passages that I might have otherwise skipped over.

    Additionally, if you’re not using AI to help you get your citations in order, what are you doing!?

    I also agree pretty strongly with the above commentor who says that AI is bad at doing philosophy. It just does not “understand” good philosophy. (William D’Allesandro made an interesting case for why recently: https://philpapers.org/rec/DALLAP-6) Don’t use it to do your writing or thinking. It is particularly bad at presenting arguments in standard form, which (in my opinion) is the pinnacle of good, clear philosophical writing (though not the nicest to read all the time).

    Reply
    1. Daniel

      You ask “Additionally, if you’re not using AI to help you get your citations in order, what are you doing!?” I use Zotero to handle citations.

      Reply
    2. Anonymous

      “Additionally, if you’re not using AI to help you get your citations in order, what are you doing!?”

      I am doing exactly what academics did before AI existed.

      Reply
      1. Anonymous

        Yes, and this was an incredibly tedious process. @Daniel, I never liked Zotero; it required a bit too much manual input. I used to do philpapers’s citation block but that was annoying as well.

    3. Anonymous

      > Additionally, if you’re not using AI to help you get your citations in order, what are you doing!?

      I see comments like this a lot, and hear similar things from colleagues. My answer is always the same: I use BibLaTeX. It does absolutely everything than one could need to do with citations and bibliograpies (and far, far more). It’s customisable when you need to be.

      I have to say *I* am baffled that anyone does anything else (especially use AI). BibLaTeX is an industry standard for a reason. The whole point of modern generative AI is that it is is not deterministic (at least from the user’s point of view) — the same input might result in different output each time. But surely this is the last thing you want from a tool for formatting citations?! When I compile my LaTeX file, I get exactly the same (and exactly correct) formatting, listing and sorting, every time.

      Reply
    4. Anonymous

      Careful with using AI for citations. I know someone who had his paper scanned for typos and it changed citations to hallucinated ones (this person didn’t ask it to touch citations). There were hallucinated citations and it resulted in a paper acceptance being rescinded. Zotero has an amazing plug-in for any Chrome browser (I use Brave) that makes it simple.

      Reply
      1. Malcolm Keating

        That LLMs intervene at the level of semantics in manuscripts even when asked to make merely mechanical changes is a known phenomenon, something that researchers are studying. This happens across models: https://arxiv.org/abs/2603.18161

  10. Anonymous

    I have found the latest Claude models to be extremely useful for a number of things:

    – Checking manuscripts before submission and proofs after typesetting. This has often found errors that I missed on a manual reading.
    – Helping me choose between different possible names, examples, ways of putting things, or other options. I’m indecisive and often spend way too much time on this sort of thing. Claude helps make me more decisive, even when I don’t go along with it.
    – Preliminary feedback on an argument or the novelty of an idea, before investing much time into it.
    – Correcting references in my central bibliography
    – General LaTeX help
    – Helping me understand very dense and difficult articles outside my subfield

    On this basis I would say that Claude has made real improvements to my work and my productivity. Of course it has flaws, but overall I have found it to be a very helpful research assistant.

    Reply
  11. Trevor Hedberg

    I have used AI rudimentarily only to get a better understanding of how it works so that I can talk in a formed way about it when I teach Ethics of AI. I have thus far never used it in my own writing or to aid my research. (That includes amassing and formatting citations — I never found this difficult or time consuming pre-AI and have not changed my behavior.) My main reason for refraining from AI use is that the output of these systems does not feel like my own work: passing it off as my own thoughts, ideas, writing style, etc., would feel inauthentic. Regardless of whether its use is permissible (a different question altogether), I do not like how using it changes my relationship to my work. I want the words on the page and the ideas they represent, however good or flawed, to be the product of my own mind, and I believe this purity of thought is irreparably contaminated if AI-generated content is used.

    Reply
    1. sahpa

      I think teachers of Ethics of AI should strive to have much more than a ‘rudimentary’ grasp of how AI works, especially if that grasp is limited to the chatbot interface and the lower-tier/free subscriptions (which it usually is for people with only rudimentary exposure). The plain fact is that the chatbots do not reflect how people are using AI to do actually consequential things like software engineering, data science, or knowledge work automation. You are basically working with a baby AI that’s inside a sandbox with very sturdy walls. The basic chatbots often have surprising limitations built in, such as — and I am not exaggerating — the inability to browse the web and read live webpages, or to actually read the document you attach in full. To manage token costs for these low-revenue (and often negative-profit) subscription plans (which are effectively subsidized by the higher tiers/more extensive users), the AI companies often silently implement workarounds, such as a search engine connection (so that the AI doesn’t read the page you send it to, but queries a search engine about the page and trusts/’trusts’ what that engine returns). This varies by the model, of course, and not all the AI companies are fully transparent about what the tool can do and when. And it varies by your own local setup — if, for example, you are using a plan provided by your university under a specific enterprise agreement. At my own institution, for instance, our basic Gemini chatbot does not ‘remember’ any previous chats and never ‘learns’ from users at all.

      More performant setups, of course, cost more money. That’s why you’ll almost inevitably end up in per-token pricing plans to use cloud services (storage and compute), etc.; using those tools effectively usually requires more fluency with things like databases and coding; and much of the interaction (the ‘engineering’) doesn’t happen in the chat interface — but rather, for example, by multistep pipelines triggered in a command line interface or by writing code that interacts with the models directly.

      Reply
      1. Anonymous

        I agree, that having a “rudimentary understanding of AI” seems less than sufficient for teaching Ethics or AI. What would you recommend OP (or anyone) read to have a sufficient understanding as a non-engineer/non-scientist?

      2. Anonymous

        Fair – but there is a difference between “rudimentary use” and “rudimentary grasp” – you can learn a lot about how it works by reading about it.

      3. sahpa

        Replying to the Anonymi….

        Both of you are making the same basic assumption (natural to philosophers!) that you can acquire a sufficient understanding of the current landscape of AI integration by studying it — reading explainers and overviews and so on. The problem is that, when it comes to the actual integration of AI into the real world (including, especially, in existing social contexts like large, old institutions), it is still very very much a wild west. There are great explainers out there about the foundational technology and algorithms underlying LLMs, sure, but that’s not what we’re talking about here and it isn’t where most of the ethical questions are. Your best way to get familiar with how these things actually work in context — in your personal and institutional context, with some specific setup of models+subscriptions+infrastructure — is to just start somewhere and start tinkering. You need to take like an ‘engineer’ or ‘inventor’ mindset to the things — try to figure out how they’re working and how they break. Along the way you’ll start identifying places where you need to go study or inquire more, but it won’t be at the high level of “what is the state of AI integration today?” — it’ll be, like, “what are ‘skills’ anyway?” or “what tools did my IT department actually license for me?”

        Having said that, I’ve found two good twitter follows in Timothy B. Lee (also on substack: understandingai.org) and Seth Lazar — who has possibly more thoroughly integrated AI into his workflow than any other philosopher working today.

      4. Daniel

        People have been saying “you need to use the frontier models to understand AI” for long enough that the current bad models have long passed the capabilities of what used to be called the frontier models. Whenever I point this out, they say things like “the new models have fundamentally different capabilities,” but they also said the same sort of thing back when I pointed this out a while ago too. And of course they also always say nobody can be told anything, they have to experience it themselves (or they say I should go check out someone ELSE who is also using AI, but when I check that person out I never find anything useful). It’s almost as if they’ve been deputized into being salespeople for AI companies! It strikes me as very strange.

        It seems one ought to be able to at least gesture towards something that YOU have done with AI that would impress me. When I eventually do pin people down enough to get them to answer that question, the output is almost never impressive.

      5. anon

        @sapha, you mentioned that Seth Lazar has integrated ai into his workflow – I assume you mean he has done this successfully, or to good effects! – can you point me to anything he’s written about how he’s done that, and the outcomes?

      6. Trevor Hedberg

        I never asserted that I had a “rudimentary grasp” of how AI works; I was simply reporting my personal use of the technology. I think most would view my personal use as relatively basic compared to what is possible with the technology. I read several books and dozens of articles on both the underlying computer science and the current ethical issues prior to teaching the course for the first time. I also watched many, many videos to determine which ones would be suitable to illustrate some of these concepts to my students. Your post makes a lot of assumptions about my course and my approach to teaching it that are unwarranted. I do not expect or require that my students will share my personal view on writing with AI, but that’s what I reported because that’s what the OP was asking about.

  12. Anonymous

    I’ve never tried using AI for any purpose. I suppose I’m just not interested enough, and it’s hard enough to master the tools I already use.

    Reply
  13. Anonymous

    I have used AI to help write syllabus statements and think of all the loopholes I need to close.

    Reply
  14. Anon

    I use LLMs a moderate amount. Some example use cases:

    * Literature surveys.
    * Proving mathematical results in journal articles left as exercises to the reader that would take me 10-15 minutes to prove.
    * Explaining or talking through mathematical results.
    * Coding in R, especially when diagram generation is involved.
    * LaTeX help, especially with tikz, stylesheets and bibliographies.
    * Checking my math. At least one substantive error has been caught that would have been embarrassing and was not caught by referees at a very good journal.
    * Spreadsheet help, especially with formulas (for example, in my spreadsheets tracking journal submissions, conference submissions, grant submissions, and expenses).
    * Spell checking.
    * An initial sounding board for arguments.
    * Checking compliance with university policies, such as consulting policies and conflict-of-interest disclosure.
    * Recommending related literatures, especially scientific and technical literatures.
    * Conforming with journal style requirements for accepted manuscripts, for example by constructing custom bibliography styles to meet journal citation conventions.

    I think many philosophers are too hesitant about using LLMs. I think this is sometimes driven by not paying for advanced models. The best models are quite a bit better than freely-available models, which are not adequate for many tasks.

    I don’t use AI agents at all. I don’t think they are yet good enough to do anything that would save me much time. An exception is that many people use agents to stack up compute-intensive requests so that they stay within plan usage limits. I’ve had some tasks that have had to be stretched over a few days due to usage limits, and probably would have been easier if I’d left those to an AI agent.

    Reply
  15. anon.

    I don’t use AI because I enjoy the process of research, writing, and doing philosophy.

    Reply
    1. sahpa

      If you’re just incurious, fine, but have you actually experimented with AI as a tool for philosophical research — beyond, say, having a few short chats with a free model?

      I won’t say that you should gain some decent familiarity with these tools before deciding not to use them in your work (and I definitely won’t argue for that on pain of ‘getting left behind’, which is the most irritating argument ever). Do what you want. But you should gain some decent familiarity with them before deciding that their use is incompatible with research, writing, and doing philosophy, which is what you implied. “I enjoy doing X, so I won’t use AI, because then I wouldn’t be doing X.”

      Reply
      1. Anonymous

        @sahpa, after all these years, I have realized that some people simply do not have the “frontier mindset” to experiment with AI. I am not saying this as a critique, but neutral: they are just stable and good at what they are already doing, and couldn’t care less. They wouldn’t be persuaded by reasoning like that you can do philosophy (differently) with AI. Here’s a (unserious!) conjecture: the less one’s work involves math, the less likely they are tempted to work with AI:)

  16. Anonymous

    I work in a small department, and none of my colleagues work in the area that I am interested in. I have no one to talk to when I have some weird but quasi interesting ideas. I also sometimes have some really naive questions in other areas. In graduate school, I can easily just have a 2-minute conversation with a professor or another fellow graduate student who can give me quick answers or guidances; but in my current school, I have no one to ask, and I do not think those are the kind of questions that warrant an email to an expert in the area.

    I find some AI models are good for my needs. I just ask for resources about some simple questions, and use the AI-generated answers as the start for my further exploration. I do not rely on them, but they give me some good clues. It also works better for me than simply googling key terms.

    Also, I use Claude Code to automatically save the answers as notes and to organize them structurally. It is interesting to review my own random thoughts in a more systematic way.

    Reply
  17. Anonymous

    I use it like I used to use Google: what’s a good recipe for this? What’s a way to do x? Show me info on y etc etc.

    I don’t use it for research. The only exception is that sometimes, when I’m writing and I have a sentence where the syntax or grammar is off, or where something is wrong with it but I can’t work out why, or where I can’t find the right word for what I mean, I’ll ask ChatGPT. That triggers ideas for me, and allows me to find the right words, like when you ask someone what the right word is.

    Reply
  18. Ned

    I don’t use AI. I’ve read enough about social, economic, and environmental concerns that I’m opposed to using AI for ethical reasons (in philosophy teaching/research and most other cases, though recognize there might be exceptions in some contexts). And while I haven’t used the ‘frontier models’ myself, I have read a decent amount about how LLMs work.

    Reply

Leave a Reply

Discover more from The Philosophers' Cocoon

Subscribe now to keep reading and get access to the full archive.

Continue reading