great paper to point to now when referencing this top-down influence. I think finding out how to tease its effects apart from those of the bottom-up agent planning influence will lead to real advances in understanding and modelling social systems.
Thanks Max!
Authors: @jzleibo.bsky.social , Sasha Vezhnevets, @manfreddiaz.bsky.social , John Agapiou, Wil Cunningham, Peter Sunehag, Logan Cross, Raphael Koster, @minsuk.bsky.social , @iyadrahwan.bsky.social , James Evans
The paper aims to bridge cognitive science, sociology, and AI.
We model individuals with LLMs not to claim humans are literally chatbots, but to develop a parsimonious model of culturally competent agents with behavior shaped by memory, context, identity, and social interaction.
In this view, rationality is a ๐๐ฎ๐ฅ๐ญ๐ฎ๐ซ๐๐ฅ๐ฅ๐ฒ ๐๐จ๐ง๐ญ๐ข๐ง๐ ๐๐ง๐ญ ๐ง๐จ๐ซ๐ฆ๐๐ญ๐ข๐ฏ๐ ๐ฉ๐ซ๐๐๐ญ๐ข๐๐, not a primordial basis of behavior. Weighing options and maximizing value is a real and important strategy, but it is just one ๐ญ๐ฆ๐ข๐ณ๐ฏ๐ฆ๐ฅ decision logic among others.
The paper distinguishes between ๐๐ฑ๐ฉ๐ฅ๐ข๐๐ข๐ญ ๐ง๐จ๐ซ๐ฆ๐ฌ (articulable rules/scripts) and ๐ข๐ฆ๐ฉ๐ฅ๐ข๐๐ข๐ญ ๐ง๐จ๐ซ๐ฆ๐ฌ (automatic, hard-to-verbalize guides). This distinction explains why appropriate conduct is typically fast and habitual, yet open to deliberate revision.
This gives us a unified way to explain five major stylized facts about norms:
1. context dependence
2. arbitrariness
3. automaticity
4. dynamism
5. sanctioning
Why norms vary with culture/situation, can be arbitrary, feel intuitive, change quickly, and hold social force.
The core idea is simple:
Human action can often be modeled as predictive pattern completion over social context. In effect, people act by answering questions like:
โWhat does a person such as I do in a situation such as this?โ
(March & Olsen 2011)
New paper: โ๐ ๐๐ก๐๐จ๐ซ๐ฒ ๐จ๐ ๐๐ฉ๐ฉ๐ซ๐จ๐ฉ๐ซ๐ข๐๐ญ๐๐ง๐๐ฌ๐ฌ ๐๐ก๐๐ญ ๐๐๐๐จ๐ฎ๐ง๐ญ๐ฌ ๐๐จ๐ซ ๐๐จ๐ซ๐ฆ๐ฌ ๐จ๐ ๐๐๐ญ๐ข๐จ๐ง๐๐ฅ๐ข๐ญ๐ฒโ
Agent-based models of social order work better when agents act by predictive pattern completion from prefix (culture/context) to suffix (action) than when they act through expected value maximization
Another possible evolutionary path is one where an ecosystem of agents is designed so well that the interactions between them have the same sort of entertainment value for spectators as reality TV or fiction does. (Spectators might also be participants at times: the Westworld model.)
๐ฏ
Words like 'consciousness' give off very individualistic vibes.
The real action is in the "cultural politics" that gives meaning to the word. Moltbook is appealing as an idea to anyone for whom this sort of multi-agent / social intelligence / society-first viewpoint is appealing.
To anyone encountering Moltbook this week and wondering about AI personhood, consciousness, sentience, etc---we published a very relevant paper in October: A Pragmatic View of AI Personhood.
arxiv.org/abs/2510.26396
I liked everything about that movie except the essentialism
The exemplar of superhuman A.I. performance to date
www.nytimes.com/2025/12/02/o...
Itโs not about information, itโs about identity.
Itโs not โmisinformationโ, itโs propaganda.
The bullshit machine is well-funded & understands the value of networked amplification.
As long as thereโs no personal cost to lying or holding false belief, they will.
www.nytimes.com/2025/11/30/c...
Yes, in part. I think the parasocial relationship to fiction writers is stronger than to (say) illustrators or stock photographers (though not fine art).
But more centrally I mean that a lot of writing โ including utilitarian stuff like reports โ matters because it expresses someoneโs stance. +
This chart (which applies even more to social media than it did to TV) lives in my head rent free.
Social media enveloping traditional media means everything and everyone is now competing in the entertainment market. Boring stuff like policy that affects millions of lives doesnโt stand a chance.
Interesting mystery. It is known that animals can learn to control neurons pretty much anywhere in the brain. But they can not learn to ignore hunger which probably means they can't turn of hunger sensing neurons. How is that avoided in the brain? They even have DA inputs.
Is our self-conception as homo faber / "tool-using animals" just because stone endures better than behavior?
If we had ethnographic records of australopithecines, H. erectus, &c, would all the stages between simian communication and "language" add up to a story more fascinating than flint?
These ideas are also at the heart of my new book "What Is Intelligence?" (out via @mitpress.bsky.social & Antikythera), where I explore how human-technology symbiosis may be the latest Major Evolutionary Transition: bit.ly/3H1p8F6
dream-logic is more powerful than logic-logic and oral cultures must encode knowledge into powerful meme-spells newsletter.squishy.computer/p/llms-and-h...
John Milton from Areopagitica.
Like, even before JS Mill gave us a utilitarian ("net upside") argument to justify freedom of speech, there was an older intuition that banning symbols risks doing violence to thought itself, and ought to be approached with "warinesse."
I'm hiring a student researcher for next summer at the intersection of MARL x LLM. If you're a phd student with experience in MARL algorithm research, please apply and drop me an email so that I know you've applied! www.google.com/about/career...
I've managed it a few times..! Though not too many
Noticed that my latest paper announcement got much more attention on Bluesky than twitter, first time that happened in my experience
Today my colleagues in the Paradigms of Intelligence team have announced Project Suncatcher:
research.google/blog/explori...
tl;dr: How can we put datacentres in space where solar energy is near limitless? Requires changes to current practices (due to radiation and bandwidth issues).
๐งช #MLSky
My favorite part of pragmatism is when itโs like โmaybe instead of worrying about shit that doesnโt matter we should worry about shit that does.โ
โOh and btw weโll learn a lot more about the shit that doesnโt in the process anyway.โ ๐ซฃ
This paper is a great exposition of how "personhood" doesn't need to be, and in fact should not be, all-or-nothing or grounded in abstruse, ill-defined metaphysical properties. As I argued in my recent @theguardian.com essay, we can and should prepare now: www.theguardian.com/commentisfre...