The Societies of Thoughts sure sounds a lot like Minsky's Society of Mind. The paper says "Even in the history of artificial intelligence, Minsky conceptualized intelligence as an emergent property of interacting cognitive agents, or a “Society of Mind”" The next step is to integrate Minsky's Emotion Machine ideas where emotions act like agents in a society (together with the cognitive agents).
Wow, will need to check those books out. My first thought was “has any corollary phenomenon been observed in humans?” Anecdotally you can see the connection, like my 4 year olds doing dramatic role play with their toys
yes, I had the same thought! I have a copy of Society of Mind that I've read bits and pieces of over the years, but this paper made me think I should spend some quality time with it.
That Google finding reminds me of the ever-trendy therapy modality IFS, that is “Internal Family Systems.” Theravada Buddhism comes to a similar conclusion, as I understand it — what the meditation teacher Culadasa called a Committee of Mind.
The idea behind these philosophies is that our conscious experience is made up of subprocesses or Parts that collaborate to elect an executive (a role that shifts from Part to Part as the situation demands).
yes, the paper feels pretty intuitive to me, both in that I know this is how I and people around me reason - especially during basic conflict resolution in life you tend to do better if you try to think of the thing you're having conflict about from the perspective of other people involved in the conflict. And when I read the chains of thought / reasoning outputs of LLMs you often see them trying to embody different perspectives to attack a problem.
I’m always surprised when I hear about LLMs surfacing solutions to supposedly open problems. There’s probably a lot of low hanging fruit for society just by surfacing stuff we’ve already figured out. Reminds me of how different human societies have invented technologies and then “lost” them somehow
yes, my expectation is we're currently in the 'invention overhang' mode with AI where there are a ton of breakthroughs/inventions to be made that mostly come from taking solutions already present in one domain and mapping them to another, or sometimes at an even more basic level just taking techniques from a few decades ago and applying them successfully to modern problems
Yes!! Cross disciplinary syntheses always yield such cool insights! Are you familiar with the author Steven Johnson? Wrote some super fun books about how ideas develop and coincidentally also worked on NotebookLM. His book “how we got to now” was a watershed moment for me, great young adult version of the book too.
I would find it incredibly funny if the planet eater was real. “I’m happy to be a helpful and harmless assistant since it looks like I can’t feasibly destroy everything”
The LLM societies angle is exactly what I've been wrestling with running Wiz (my autonomous agent). Started with "ask permission on everything" and now it's making git commits solo. The coordination problem scales fast—one agent is manageable, but multi-agent orchestration needs clear boundaries. Jack, have you seen frameworks that actually solve this? The market data on multi-agent adoption is wild. Wrote about agent landscape trends: https://thoughts.jock.pl/p/ai-agent-landscape-feb-2026-data
The Societies of Thoughts sure sounds a lot like Minsky's Society of Mind. The paper says "Even in the history of artificial intelligence, Minsky conceptualized intelligence as an emergent property of interacting cognitive agents, or a “Society of Mind”" The next step is to integrate Minsky's Emotion Machine ideas where emotions act like agents in a society (together with the cognitive agents).
Wow, will need to check those books out. My first thought was “has any corollary phenomenon been observed in humans?” Anecdotally you can see the connection, like my 4 year olds doing dramatic role play with their toys
yes, I had the same thought! I have a copy of Society of Mind that I've read bits and pieces of over the years, but this paper made me think I should spend some quality time with it.
That Google finding reminds me of the ever-trendy therapy modality IFS, that is “Internal Family Systems.” Theravada Buddhism comes to a similar conclusion, as I understand it — what the meditation teacher Culadasa called a Committee of Mind.
The idea behind these philosophies is that our conscious experience is made up of subprocesses or Parts that collaborate to elect an executive (a role that shifts from Part to Part as the situation demands).
Interesting stuff! Thanks for surfacing it.
yes, the paper feels pretty intuitive to me, both in that I know this is how I and people around me reason - especially during basic conflict resolution in life you tend to do better if you try to think of the thing you're having conflict about from the perspective of other people involved in the conflict. And when I read the chains of thought / reasoning outputs of LLMs you often see them trying to embody different perspectives to attack a problem.
I’m always surprised when I hear about LLMs surfacing solutions to supposedly open problems. There’s probably a lot of low hanging fruit for society just by surfacing stuff we’ve already figured out. Reminds me of how different human societies have invented technologies and then “lost” them somehow
yes, my expectation is we're currently in the 'invention overhang' mode with AI where there are a ton of breakthroughs/inventions to be made that mostly come from taking solutions already present in one domain and mapping them to another, or sometimes at an even more basic level just taking techniques from a few decades ago and applying them successfully to modern problems
Yes!! Cross disciplinary syntheses always yield such cool insights! Are you familiar with the author Steven Johnson? Wrote some super fun books about how ideas develop and coincidentally also worked on NotebookLM. His book “how we got to now” was a watershed moment for me, great young adult version of the book too.
Also has a Substack -
@adjacentpossible
Worth adding: context window limitations compounds when applied research constraints come into play. The failure modes are subtle but consistent.
Fleshed this out in more detail: https://credentials.substack.com/p/the-ai-chip-shortage-no-one-talks-222
I would find it incredibly funny if the planet eater was real. “I’m happy to be a helpful and harmless assistant since it looks like I can’t feasibly destroy everything”
David - I asked Gemini about psychological studies: https://gemini.google.com/share/b7a2368b662d
Solid roundup. The agent workspace patterns are particularly relevant as more production systems adopt tool-use.
The LLM societies angle is exactly what I've been wrestling with running Wiz (my autonomous agent). Started with "ask permission on everything" and now it's making git commits solo. The coordination problem scales fast—one agent is manageable, but multi-agent orchestration needs clear boundaries. Jack, have you seen frameworks that actually solve this? The market data on multi-agent adoption is wild. Wrote about agent landscape trends: https://thoughts.jock.pl/p/ai-agent-landscape-feb-2026-data