AI policy
Current thinking on AI tool usage as writer and content creator
Why AI?
Protagonist Science is a little digital corner that tries to bring science and society closer together. That mission only works if readers can trust what they find here is not only well sourced, reasoned and argued for; but also not build to steal their time, money or emotional engagement.
Consequently, I have seen the rise of generative AI very critically, as do I see the whole campaign to shove AI chatbots down everybodies’ throats so robber billionaires can become feudal godkings while the rest of us their servs.
Unsurprisingly, I have also been a lukewarmer on AI adoption; I see the innovations in machine learning, transformer architectures etc. as a cool and fundamentally normal, but not magical, technology that is powerful when used appropriately and in bounded use cases, and pretty terrible when implemented as a syncophantic user manipulation and lock-in before enshittification tool.
That all being said, either denying this technology exists now or objecting in principle to its use does not seem to be the right way for me, especially when there are many positive visions of AI use, from kami to community.
So inspired by science journalist Elise Cutts on her newsletter´s AI policy, I thought about what my principles might look like for AI use.
My rules
1. The blog writing is mine. Maybe I am early because I read a lot, but AI-generated prose, even when trained on style, tone and voice of my own material, tends to fall short, be bland, boring, generic and just not really worth engaging with. I will not try to avoid certain constellations or phrases now associated with AI to signal my humaness, I will just keep writing.
2. My name means my accountability. Every factual claim, citation, and argument published here ultimately comes back to me. If I used an AI tool to assist with research on something, but I can’t verify it, it doesn’t run or will be transparently and explicitly caveated as AI generated content. In general, because AI is not accountable, it should not be used in processes where accountability is required.
3. AI for summaries, aggregation, bridging of bounded content tends to be good and useful. One of my understandings of modern LLMs is that they are superhuman token manipulators; and when the control of the original data is mine, reshuffling these tokens on existing, bounded-content in many ways to sundays is just something I find very useful for me and my work. Be it to summarize interview transcripts, check references, or filter and search content for things like ad verbatim quotes is a real time saver.
4. AI-assisted coding and prototyping. I think coding is the most interesting case of bounded utility where the AI can take the druggery of coding while keeping the creative process and architectural decisions on what to code by the human. Now I am not saying anybody should use AI for really critical tasks where accountability is needed (see rule 1), but for prototyping some personal project ideas that live on the web I have to say it has now grown on me.
How I use AI
Research and literature triage: summarizing papers or articles, surfacing related work, helping me get through a stack of potential sources faster. I will still read the primary sources myself before referencing them.
Prototyping: AI may help me brainstorm scenarios or execute certain ideas I have to a point where I feel ready to either discard or take over as human. Just like other humans can help us feel inspired, so can whatever a machine might bring forward. Anything goes. The question is the rigor and process applied afterwards, not how it came about. (this is very much how Feyerabend talks about science under the fancy term epistemological anarchism)
Vibe Coding: for building new html features or data analysis graphs I found it very useful. AI handles the boring, mechanical parts — boilerplate, syntax, debugging. The creative and architectural decisions — what to build and why — stay human.
and I quote directly from Elise Cutts:
The absurdity exception
If a system or task disrespects my humanity, it does not necessarily deserve the respect of a human response. I may use generative AI tools to do “creative work” (writing) for absurd tasks, especially if the entity demanding and/or evaluating the task is itself an AI (e.g. hiring systems). Absurd tasks are those whose point is not to produce something high-quality, creative, meaningful, or valuable but rather to satisfy some arbitrary requirement or check a box. This also includes highly repetitive, non-creative tasks like re-formatting a list of citations.
How I don’t use AI
I don’t use AI to draft my prose.
I don’t let AI hallucinate what it knows.
I don’t use AI to manufacture sources or expert quotes.
I don’t allow AI to make up data or research notes.
I don’t use AI to claim somebody else´s deed.
I don’t craft AI-generated images to mislead.
Why this matters
A blog built on fighting disinformation has a higher bar than most to not perpetuate harmful information.
This policy is a living document. As my tools and workflows change, I’ll update it to match. — but the underlying commitment won’t: my words, my ideas, and my accountability mean I stand behind what’s published here, even if I don´t get it right the first time.
Cheers,
Philipp