On the Use of AI in Comments
I’ve updated the Comments Policy at Daily Nous with a passage about using AI to write comments here.
[Manipulation of M.C. Escher’s “Drawing Hands”]
Here’s what I added:
- Regarding the use of artificial intelligence (AI) / large language model (LLM) tools like ChatGPT, Claude, and the like: Generally, please do not submit comments written in whole or in part by these tools.
That said,
-
-
- quoting the writing of these tools in order to discuss their outputs or capabilities, if relevant to the topic of the post, may be acceptable. If you do this, you should clearly identify the text as AI-generated and identify which tool(s) you used.
- translation or minor editing of comments using AI tools or apps/software that makes use of AI may be acceptable, provided that you clearly identify the text that was translated or edited by an AI tool and identify which tool(s)/software you used.
-
As I say [earlier in the comments policy], Daily Nous is a space for philosophers, academics, students, and other interested parties to discuss news and issues related to philosophy and the philosophy profession. Discussion here is a social enterprise entered into on the expectation that our interlocutors are other persons interested in conversing about the topic under consideration, and that the words they are using are their own. I think that’s a reasonable and valuable expectation that helps contribute to the quality of the discussions here, and I want to protect it against behavior that would undermine it.
If you suspect a comment is AI-authored, please let me know about it by emailing me here, and I will take a look and, if I think you’re right, remove it. Please do not make such accusations in the comments thread, as doing so is not conducive to a good discussion atmosphere. Thank you.
Comments welcome.
Sounds good. Do you think you’ll start getting comments wholly generated by autonomous bots at some stage? Or will the “I’m not a robot” captchas be able to stay in front of that?
“Do you think you’ll start getting comments wholly generated by autonomous bots at some stage?”
Yes — in fact there’s some chance that some of the spam I already get is automated llm-written stuff. Fortunately, current spam-filtering technology is pretty good, and with luck it will keep developing in pace with the problems it is meant to address.
Simply by having this minimal policy, the comments section here becomes a better deliberative space than many classrooms (which are permissive or have no policy at all).
“Discussion here is a social enterprise entered into on the expectation that our interlocutors are other persons interested in conversing about the topic under consideration, and that the words they are using are their own.”
Hear, hear.
Here is an argument:
No-AI matters for a blog.
Courses matter more than blogs.
Therefore,
No-AI matters for courses.
No AI was used in the creation of this syllogism
To be fair, he only said “Here’s an argument”; he didn’t say it was valid or sound.
i should have been more charitable
Having a comment section matters for a blog.
Courses matter more than blogs.
Therefore (clearly), having a comment section matters for courses.
I am a regular reader, but haven’t commented before, and am unlikely to do so in future. While I won’t be using AI here, paradoxically I sometimes use AI to revise comments I post elsewhere in order to sound more human. My natural tone quite often gets flagged as AI.
Asking that discussions of philosophy include only contributions composed by humans?
Next thing, you’ll be excluding obvious plagiarism.
I’m being redundant; excluding AI produced comments entails discouraging plagiarism, since generative AI functions exclusively as an automated plagiarism extruding device, lacquered with rampant falsehoods.
Oh, and reliance on generative AI promotes utter laziness and shallowness in thinking. And generative AI is the primary mechanism for what is called ‘enshittification’ of every enterprise and discipline.
What’s not to like about generative AI, especially on a site whose purpose is to promote the study and practice of philosophy?
I appreciate the last part. Accusations of AI use are now far more prevalent than AI use itself, which is ridiculous.
I’ve never used AI to write comments and probably won’t, because I typically type them on my phone while doing other things. If I used AI, my comments would be much better—fewer typos, better organization, and less intemperate (ha!).
That said, I’m not convinced AI use needs moderation. What matters, as Matthew Hammerton explained, is taking responsibility. The real scourge of this blog, to me, is anonymous commenters—denigrating, smearing, and accusing named commenters behind a shield. This creates an asymmetrical situation, where the same commenters eschew accountability, again and again, while many of us continue collecting strays. If an author identifies themself, then they are accountable for their words, whether AI-assisted or not. I’ve read people complain about AI comments because they couldn’t tell who they were talking to—I agree, this feels uncanny. But to me, the exact same problem occurs when responding to—or even more so, being replied to by—anonymous posters, as stable as their pseudonym may be. Many regular commenters here, pseudonymous or not, are bona fide, constructive interlocutors. But others are just extremely disagreeable. I know there’s only so much you can do about this, Justin, but there is something, and I’m hardly seeing it.
So, if I had to suggest a revision to the comments policy, it would be to moderate anonymous posting a little bit more heavily than currently. Those who choose to use AI to write comments should, like our students, be able to explain their thinking and take responsibility for the words they post. AI use becomes a problem when combined with anonymity, but the primary problem, to me, is not AI.
With all that said, thank you for your efforts in tempering the ineluctable tendency of contentious comment sections to quickly become radioactive.
I think putting a restriction on AI is perfectly sensible. I think AI-editing would probably improve a bunch of comments, especially by removing typos, unclarity, and intemperateness. But it’s probably better for responsibility to make people type things out themselves. And the increased friction from requiring typing probably helps when things have a danger of getting overheated. So all things considered, I agree with the balance Justin has drawn.
(Even if I didn’t agree with it, I’d say it’s miles from unreasonable, and it’s his space, so fine. But actually I think it’s probably right.)
I do agree that it would be better to require, or at least encourage, stable and distictive pseudonyms. The best comments sections I’ve been in have not required names, but they have required that everyone stick to a clear label *for that board*. So no one gets to post as ‘Grad Student’, and whatever label one picks, like “Lapsed Modal Realist” or whatever, is constant from thread to thread.
I suspect it’s far too much to ask Justin to enforce such a policy, but I would be happier if everyone voluntarily adopted it.
Hi Brian, I don’t really disagree. The new restriction is fine with me, certainly not “unreasonable.” I’m just not convinced there really was a problem that needed moderation.
Good point—though running your comments through AI might also involve friction, so maybe it’s a wash;
“I’m hardly seeing it”
That is the nature of much of the moderation, which consists in some comments never getting published. Sometimes moderation involves approved comments being removed, but most is of that first sort, and so will be invisible to readers.
I know, I know, I don’t doubt you’re doing the lord’s work. I just said I’m hardly seeing it—because a lot of abuse still slips through the cracks.
As for the point about taking responsibility for the AI-generated text they reproduce, that may be preferable to failing to take responsibility, but I don’t see it as preferable to people not using it, at least in the context of the comments here.
Forget AI for a second. Suppose people commented here simply by reproducing relevant sentences or paragraphs written by other people. They’d quote Aristotle or Russell or Hume or whomever, adding at the end, “my name’s Justin and I endorse this message.” They’re taking responsibility for the words and ideas they borrow, but still, I don’t think that would make for a good comments section. My opinion doesn’t change when we sub in AI.
Wait, where did I suggest it was preferable to use AI than not? I specifically said I don’t and won’t even if it made my comments better, and that reading possibly AI generated comments felt uncanny.
Sorry, I took your reference to Hammerton to suggest that you thought AI-written comments here would be okay so long as people took appropriate responsibility for them, while making a point about pseudonymous commenters not taking responsibility for what they say. Perhaps you were just saying the latter bit.
Some AI comments could be better than some anonymous comments, in part thanks to responsibility, yes, but that’s a very particularist claim, not a general principle about AI vs not AI.
Suppose someone who struggles with verbosity, yet has an interesting point to make, writes out their comment to post on DN and then realizes that it is way to long for a blogpost comment. To correct this, they ask an LLM to rewrite it to half the length. Then, they carefully review the LLM output to make sure that it hasn’t changed their meaning or emphasis anywhere and to ensure that they can take full intellectual responsibility for it. Finally, they post it.
I don’t see what would be wrong with this and would have no problem reading this kind of comment. Indeed, such a commentator has probably done blog readers a favour by saving them from having to read through an unnecessarily long comment to access their interesting point.
In the cases where the use of LLMs is problematic, the problem is not the technology itself but the abdication of intellectual responsibility. For example, imagine someone who wants to oppose a viewpoint expressed on DN but is too lazy to construct an argument against it. So, they instruct an LLM to draft an opposing comment and post it, and then repeat this process for each reply they receive until their interlocutors are exhausted. This kind of thing would greatly degrade the comment section.
What is objectionable here is not the use of AI but the bad faith involved in posting arguments one has not taken the trouble to understand and couldn’t adequately defend. In this respect, this practice is analogous to trolling, which predates LLMs and similarly degrades comment sections. In both cases, people post stuff they haven’t bothered to understand and couldn’t adequately defend because they have an ulterior motive beyond the appropriate one of engaging in a productive conversation.
Having said all of this, I appreciate that it might be very difficult to police the line between commentators who have taken full intellectual responsibility for their comments and those who haven’t. Furthermore, AI tools can greatly increase the damage that bad faith commentators do. Hence, I can understand where Justin is coming from if he feels that a blanket ban on AI is the best way to content moderate this blog.
“Suppose someone who struggles with verbosity, yet has an interesting point to make, writes out their comment to post on DN and then realizes that it is way to long for a blogpost comment.”
This example doesn’t move me one bit, assuming that this person is, like most people, capable of learning how to write better. Either post it and know that it might be less clear or more annoying to read, or don’t post it until you–the person with the actual thoughts–determine a better way to say it. That’s what we expect of interlocutors. We’re starting to act like people have a right for their thoughts to be represented by the clearest or most efficient sentences, and all this talking/writing stuff is just mechanical b.s. that stands between our brilliant ideas and other people.
Suppose this same person has a reasonable amount of evidence that, if prompted, ChatGPT will produce what they think is a suitable post. Would you see it as all the same for them to type out “prompt ChatGPT with: _____”?
I don’t buy the idea that we should obviously expect interlocutors to “[learn] to write better” if they want to be represented by better prose. We shouldn’t underestimate how weighty a requirement this might be in any particular case. I consider it an acceptable practice for someone who is less-articulate to request help from a more articulate friend or ally to express themselves. In my country, there are numerous situations in which access to such assistance is a legal entitlement. Under such circumstances, I don’t see why the choice to use an LLM instead would be illegitimate.
If it’s not a general principle that people cannot use LLMs to express themselves more clearly, then what special features of a philosophy blog comments section mandate it?
To answer your direct question: if Alice believes that, if prompted with prompt X, ChatGPT will produce a post that reflects Alice’s own thinking and that Alice would be willing to defend, then I guess I would not object (on intellectual grounds) to Alice writing a post that says, “<brief summary of argument>; you can see the argument in detail by prompting ChatGPT with X”. But the evidence would have to be extremely strong, and (unless the prompt X is pretty long and doing a lot of work) I do not see such a scenario being remotely plausible at present: the output would need to be guaranteed to reflect Alice’s own thinking.
Moreover, it doesn’t really matter what evidence Alice privately has about the LLM’s behaviour. The communicative onus is on her to persuade the reader that the LLM will behave in an appropriate manner. If she merely asserts that the prompt will work, many readers will quite sensibly not bother to go to the trouble of checking. I would personally consider engaging with a post along the lines of: “<brief summary of argument>; you can see the argument in detail by prompting AliceGPT with X. AliceGPT is an LLM trained extensively on my own thinking, and assessed independently on over 1000 diverse topic prompts as accurately reflecting my thoughts.”
You might have intended a much lower hypothetical standard: Alice merely believing that, if prompted with prompt X, ChatGPT will produce output that would be suitable for posting (but will not necessarily reflect Alice’s own thinking). In such a case, Alice should not propose the LLM as a proxy. She would be free to claim, “When prompted with X, ChatGPT generates outputs that contribute valuably to this conversation,” and provide supporting evidence if she wants to be taken seriously.
In the comments to another blog post, I worked through an example of this, although the use case was tact and respect rather than verbosity. I reproduced a draft comment, and an LLM paraphrase of that comment which was significantly more tactful.
I think this example illustrates why there are some conceptual problems with asking whether an LLM rewrite has changed the meaning or emphasis of prose. I did review the LLM output to assess whether it said what I wanted to say.
I felt uncomfortable, because
This was only a single post. If I was in the habit of using LLMs to edit every post, wouldn’t these slight disparities compound over my posting corpus? What would the implications be?
While I’m genuinely unsure what to think about the intellectual (or interpersonal) merits of using LLMs to improve comment posts, let’s not forget other ethical concerns (e.g. environmental impact and copyright theft).
I am not inclined to post under my own name–I think it would be imprudent, given various facts about my personality–so I wouldn’t post if real names were required. That being said, I respect those who do post under their own name, including you specifically. So I just thought I’d add: cheers to you!
Just to be perfectly clear, I completely understand why many people would rather post anonymously and will defend their right to do so (for instance, not being a US citizen I now think twice before posting under my real name on certain issues…). My concern was that anonymity breeds a lack of accountability which engenders bad behavior. But bad behavior is that which needs to be tamed—I’m only asking that we manage the symptoms not the cause.
I agree with Nicolas Delon that the biggest problem here is the anonymous commenters. They’re often just trolling. They make lots of ad hominem attacks and rarely contribute anything philosophical.
Anonymous commenting can be appropriate on topics which might involve risk for junior faculty. Otherwise, get rid of it.
Sounds good, Justin
On prohibiting public accusations, though: it may be hard to promote a social norm against AI commenting if it’s never clear what comment is AI-written and who’s doing it.
Or are you planning to post a note that something was flagged as AI and removed, every time it happens? If so, that could also help the community track how prevalent the problem is.
But I get the tradeoff for collegiality — it’s not an unreasonable choice and is ultimately your call.
On prohibiting anonymous comments, as some folks are arguing for here: there’s a very real power dynamic between senior and junior scholars, esp. grad students, regardless of topic.
So, banning anonymous comments could have a serious chilling effect on free speech here, and that’s something that needs to be factored into the deliberation…
Good thing no one said we should prohibit anonymous comments! I’m on the record here, years ago when the issue came up, defending the ability to post anonymously precisely for the reasons you mention. I usually get flak here for being too pro free speech and very skeptical of bans in general.
You might not have, Nicolas, but someone (Eric) had:
(Emphasis added.)
Yeah, I missed this bit. You wrote ‘folks,’ and I most definitely didn’t call for a ban. I disagree with Eric on this, if ‘getting rid’ means banning. I’m only calling for more explicit nudges from Justin about conversational norms and stricter reporting and removal mechanisms for offenders who deal in insults and personal attacks.
I think Patrick might be arguing that all topics might involve risk for junior faculty. That may well be right. I don’t have any objection to that!
Instead of a ban on anonymous posters, there could be a ban on personal criticisms. This would not only not chill free speech in any important way, but it would also encourage discussion.
Yes, that and other options seem available to Justin.
I assume, though, he’s not getting paid to run Daily Nous, so a sustainable solution would need to be low effort one. Having to adjudicate whether x is a personal criticism, ad hominem, etc. could blow up his workload here. Same for other suggestions here, incl. mine, as well meaning as they may be.
Even banning and taking down AI comments can be a lot more work, despite good reasons for it…we’ll see…
Content moderation is very, very hard, even more so if you’re a one-person operation. Anyway, on this issue, there’s no pleasing everyone.
Thanks, Justin, for being able to do any of this and creating this community space!
Yes, it must be a huge amount of work maintaining dailynous and I can understand if Justin doesn’t want to take on even more.
I use AI partially to help me compose and edit my own writing, due to my having aphasia(a linguistic disability) after my major stroke. It is what LLMs and AI are really good at helping people do. So, unfortunately, your restrictions are mainly around the mode of presentation in which I do philosophy.
Yes, and presumably, you are still able to take full intellectual responsibility for everything you write, so why would anyone object to you proceeding this way?
I may be misreading the policy, but there isn’t anything stopping you from doing what you normally do, so long as you enclose the relevant text in quotation marks and attribute it to the AI tool, right? Or is your worry that you’d like to do this without the attribution?
More of the latter worry. Let me walk you through my writing process and the AI tools I use. Every time I write, I rely on autocorrect and text suggestions because I have to type with one hand. Both tools use AI trained on my typical word patterns, which makes typing significantly easier. Due to my condition, I can usually get the first few letters of a word correct, but sometimes, depending on the word, my accuracy deteriorates as I continue to spell it. Spelling relies on habit; however, due to my stroke, I must consciously relearn from the beginning, gradually progressing by sounds as I did in second grade, all over again. In both applications, typing just the first few letters will suggest the correct word.
When I totally cannot remember how to spell a word but can recall its pronunciation, I use speech-to-text to find the correct spelling. This tool uses AI trained on my voice, so it typically understands what I am trying to say. The more I use it, the more accurate it becomes.
Due to my aphasia, I sometimes lose access to the word I want to use, though I can remember its first letter and its meaning. Large language models excel at retrieving words based on partial information like this. They are, incidentally, also excellent at crosswords, but it would rob you of the fun of playing them.
I also use a spell checker, which is standard practice now. Interestingly, critics once warned that spell checkers would cause our brains to atrophy. Linguists worried they would weaken spelling proficiency and vocabulary retention. I’m not so sure whether they believe that anymore, or whether people really think it is that bad anymore.
Once I have drafted a paragraph, I run it through several editors to refine it. I typically input the paragraph into Claude and ask it to edit thoughtfully, requesting that it list all edits at the bottom with explanations so I understand the changes and their rationale. Occasionally, I decline the suggested edits, especially when Claude employs em dashes or uses awkward words such as “foundational,” “essential,” or “essentially”, as I do not align with these philosophical labels. However, at other times, it offers valuable suggestions for sentence-level clarity, formality, and precision. Next, I review the text in Grammarly, evaluating any flagged issues and examining how different versions might restructure my sentences and word choices. The whole time, I considered better word choices and sentence structure, and I will change them as needed using a thesaurus. This process is extensive, probably more involved than what most people undertake in their writing.
If I understand correctly, I now need to track every sentence in which I used AI and specify which AI tool I used. This would be a monumental task for me, and frankly, I would probably stop contributing altogether if such requirements were implemented. Additionally, I wonder what counts as “minor edits.” I suspect many people would consider my edits far from minor.
I would propose that the blog owner, Justin, maintain a small list of would-be regular commenters like me who use AI due to disability and allow them the flexibility they need to participate.
More Luddite nonsense. If the AI can construct a better comment than I can, then what is the point of me typing one out? All that matters is the final product, after all. And now we all just have to read inferior, less-informed comments because we have this irrational, bio-centrist prejudice towards words produced by living things? Sad.
Let me push back against this. I want to have conversations with agents who can take responsibility for what they say and be held accountable; who have a stable intellectual character that passes some minimal threshold of decency; and who have their own perspective on things that comes from their experiences of the world and their way of reasoning through it. Chatbots cannot be ‘agents’ in this sense, and hence I am NOT interested in having conversations with them (although they are very useful tools that I regularly use). On the other hand, a human who takes full intellectual responsibility for everything they say, but uses AI as a tool to improve the quality of their comments, is an agent in this sense, and I would be happy to engage with their comments.
Despite this disagreement, we at least appear to agree on one point: an across-the-board opposition to any use of AI, including cases where a human retains full intellectual responsibility for the content, is wrongheaded and is a form of Ludditism.
Thank you, Nick, for this helpful reminder of Poe’s Law.
This is like comedians doing Trump sketches. Reality in professional philosophy has become so parody-proof that the humor only barely comes through.
Pretty sure this is Parody, and a Funny one. We’ll never know who wrote it though! SAD!
I’m curious what people think about the following exchange as it fits into the policy. On a different site (Substack), I was in an exchange today with someone who claimed that Alex Pretti resisted arrest. I suggested the behavior that looked like “resisting arrest” could just be a typical reaction to being pepper sprayed. The other (anonymous) party appealed to their decade of experience in military and law enforcement in the course of claiming that the usual reaction to being pepper sprayed is “not to freak out and fight, but sit down, comply, etc.” I was skeptical. But it’s hard to directly point to evidence against (or for) claims like that. I asked ChatGPT about typical symptoms of being pepper sprayed, and how realistic it would be for those symptoms to be read as “resisting arrest”. While I hope this wasn’t just sycophancy–I tried to word my prompt neutrally–it reinforced my sense that this could absolutely happen. So I replied making explicit that I was quoting a ChatGPT response, and excerpted a bit of its reply to me. (I said: “Fwiw, ChatGPT disagrees…” and then quoted a couple short paragraphs.)
I think that was reasonable online-commenter behavior. If somebody is anonymously appealing to supposed expertise in support of a difficult-to-verify factual claim, checking the claim against an LLM and reporting the results strikes me as appropriate. Is that the sort of thing that would be prohibited by this policy?
If so, I’m not saying it’s a bad policy–even if I’m right that it was reasonable in this case, and maybe I’m not–maybe it’s better to have a blanket policy that catches some good comments in order to stop us from drowning in slop.
Quoting ChatGPT or making reference to its outputs doesn’t violate the policy.
It’s an appeal to a weak authority, however.
Another danger is that the few loud, radically opinionated voices are driving everyone away from this conversation. The increasing consensus and rage in this community on AI use=plagiarism or something like that are forming narrow-minded echo chamber. (Personal disclosure: I am *not* using any LLMs on anything research related, not even proofreading. Nor would I identify myself as remotely pro-AI.)
I completely agree with Alice. The weirdly strong reactionary positions around AI are troublesome. But I think they are also feeding into another related problem that isn’t a new phenomenon at all.
Justin — it’s your blog, and you can moderate it however you want. You obviously put a lot of thought into the problem of how to balance competing desiderata in moderating the discussions, but you are hosting the discussions on a a blog, not in a democratic community in which anyone has a right to post by virtue of citizenship of some sort.
The fundamental reason why discussions at Daily Nous go off the rails and become either tedious or downright toxic is that people get drawn into pissing contests. Most people (I hope) are genuinely interested in using this space to share ideas and listen to one another’s views, but there are some regular posters who seem more interested in using your blog as a soap box or opportunity for…
Well, let’s put it this way — if I’m right that this persistent problem is that discussions can backslide into pissing contents, then the problem is NOT that some people are using AI to to augment their pee pee streams while others are insistent that this isn’t fair. The problem is the contest. Ironically, AI is one of the topics that seems especially prone to this sort of degradation, and I suspect that the pressures and backlash around AI=plagiarism (as Alice put it) get drawn into that. But there are other topics that regularly involve similar dynamics here, many of which predate AI, of course.
This is probably why there are so many reactions to you saying you are instituting a new policy on moderation that relates to AI by pointing out that a bigger issue may be the anonymity issue.
Came across this in Ars Technica this morning, seems somewhat relevant to a discussion of the use of generative AI here (or anywhere else):
https://arstechnica.com/ai/2026/01/how-often-do-ai-chatbots-lead-users-down-a-harmful-path/
“In the newly published paper “Who’s in Charge? Disempowerment Patterns in Real-World LLM Usage,” researchers from Anthropic and the University of Toronto try to quantify the potential for a specific set of “user disempowering” harms by identifying three primary ways that a chatbot can negatively impact a user’s thoughts or actions:
// // //
I dunno, I’m just a clinical psychologist, but reality distortion, belief distortion and action distortion strike me as pretty bad. There is no indication from the now considerable research into the maladaptive psychological effects of generative AI on individuals who routinely use them, that those who study philosophy, or are employed in a professional role in academia, are immune to these effects.
The notion that intelligent, highly educated individuals might use these devices ‘safely’, and ‘innocuously’ is appearing more each day to be an ill-founded assertion, and it’s generally not a good practice to assure ourselves that we’ll be ok, because we’re too smart to be influenced in this way.
Considering the concerns expressed by contributors to this site about the proliferation of generative AI by students as a substitute for actually bothering to read, write, study (you know, rather than actually perform the functions of ‘student’), and the somewhat less discussed practice of faculty using generative AI to produce course materials and grade student work (you know, rather than actually perform the work of ‘faculty’), there is some irony in the defense of the use of generative AI to ‘assist’ in the composition of comments posted here.
As I alluded to in my previous comment, to the extent generative AI functions at all, it does so by way of systematic plagiarism. This is, in fact, its base operation. Because it is incapable of identifying what is true from what is false, and by design creates fictional results in the a form that appears legitimate and authoritative, generative AI’s other primary operating mode is to be a falsehood generator.
A systematically plagiarizing falsehood generator (which cannot operate in any other way) that psychologically harms those who utilize it.
Are there any ethical implications at play here?