Don't seek permission, center values

Note to readers: some of these ideas made it into a commentary I wrote with Christine Cuskley:

Dingemanse, Mark & Cuskley, Christine (in press). For robust research, center values, not technology. Behavioral and Brain Sciences. Preprint doi: 10.5281/zenodo.18944023

One topic that often comes up when discussing LLM technology in relation to research integrity is one that I will describe as seeking permission. When looking at the ethical, legal, and societal harms imposed by LLMs (and there are many), sometimes people feel the message ends up altogether too negative. How about this use case I heard of? Aren’t some people getting something useful out of it? Surely we can’t ban the tech outright?

Often this is phrased as a concern about messaging (people won’t accept it if we tell them how bad it is; you need to sugarcoat it by also mentioning something nice). Sometimes it is phrased as a majority argument: it’s already here, everyone is using it, surely it can’t be that bad? (Smoking would like a word.) Sometimes it is a concern about missing the boat: these are skills we need, telling people not to use it is like telling them to go back to quill pens. ¹

What I think is going on in these kinds of cases is that the central thrust of the argument is being missed. When it comes to research integrity, the key is a values-first perspective rather than a tech-first perspective.

A values-first perspective asks: how can we best uphold the values and standards that make our research robust, reproducible and future proof? A tech-first perspective asks: yeah, but how can I use this technology? It puts technology above values. It seeks permission but sidesteps the question of values.

Can I use an LLM for …

One example that came up in a recent session started with literature review: surely it’s not too harmful, a questioner said, to use an LLM for a first stab at a literature review? Multiple participants pushed back on this, saying that actually, LLMs don’t reliably summarise. Also, they provide only the most average consensus view; we don’t know what’s being left out. And LLMs by nature regurgitate without understanding; can we actually identify confidently produced bullshit in a field that we don’t master fully? Further, reading is a hard-won skill: tracing arguments, spending time with papers, separating substance and rhetorics; surely we don’t want to lose this skillset.

A retreat followed: well then, maybe literature review was not the best example, but surely there is some other kind of use that is okay? Someone mentioned programming. Folks brought up deskilling, the security risks of vibe-coding, technological debt. Oh well, that’s not the kind of programming I meant. And so on.

We can play this game all day: seek permission for an edge case, retreat to another case when trouble arises, appeal to amiability or tolerance. Two things to note when a conversation gets to this point. First, it’s a misrepresentation: reluctance to grant permission is portrayed as an intolerant move, a wish to take the other’s toys, when it is nothing like that. What do you need my permission for, anyway? Second, it’s a distraction: it moves the conversation away from values and towards a tech-first perspective.

Values: a matrix for mindful choice

This is where a values-first perspective can help cut through the knots. The approach is not to recommend or forbid particular tech products, or particular uses; it is to provide a matrix for mindful choice. For any attested or conceivable use, you can ask: how would this help or hinder my upholding of high standards of research integrity? If we consider the pushback voiced in response to the ‘literature review’ use case, you can see they appeal to the same core values:

• honesty (reporting processes accurately and being open about margins of uncertainty)
• scrupulousness (being precise and thoughtful)
• transparency (showing one's process and allowing others to build on it)
• independence (being impartial and unswayed by commercial or political interests)
• responsibility (taking into account the environment; accepting accountability for the statements made)

These values do not come out of thin air; they’re straight from a widely adopted code of conduct for research integrity (NCCRI, 2018). They are also not controversial; most scientists will recognise them as principles that characterize robust research. I’ve written about them before, e.g. in my guidance on GenAI and in my post on why synthetic text has no place in science blogging.

If you use these values as a compass to steer by, it’s easier to navigate the landscape of technology use. On the other hand, if you find yourself seeking permission, one useful thing to do is to step back and inspect the underlying value conflict.

As you move from a tech-first to a values-first perspective, the question shifts from “won’t you give me permission?” to “how do I do the best science possible?”. And that, to me, is a question worth asking.

Further reading

• I am an AI Hater, by Anthony Moser
• Against the Uncritical Adoption of ‘AI’ technologies in academia, by Olivia Guest and collaborators
• Netherlands Code of Conduct for Research Integrity, 2018

1. Quick rhetorical intervention here: I don’t hide my distaste of Big Tech LLMs, but I rarely suggest to ban them, so when I get this quip I typically bounce it back: what made you think I suggested that? There’s a good conversation to be had here, but it is not about quill pens. It is about the moral distress that a response like this reveals.[↩]