Society of Mind outline
The mind as sub‑agents with competing goals; coordination, not unanimity, drives behaviour.
Have you ever wanted to go to the gym but also wanted to stay in bed? Or decided you'd work on something important, only to find yourself doing busywork instead? Or felt torn between what you should do and what you want to do, as though two different people were arguing inside your head?
That's not a metaphor. Or rather, it's a useful one. Your behaviour isn't the output of a single, unified 'you' making deliberate choices. It's the outcome of a coalition: multiple specialized processes---habits, motives, evaluators---each proposing actions based on their own training history and priorities. Context and control arbitrate between them, deciding which proposal gets to run. The rest linger in the background, shaping your mood, your second thoughts, and the sense that you're 'of two minds' about something.
Marvin Minsky called this the Society of Mind: the idea that intelligence emerges not from a single process but from many smaller, semi-independent agents working (and sometimes competing) together. You don't have one monolithic 'self' that decides everything. You have a parliament of routines, each with its own job, its own history, and its own idea of what should happen next.
This explains a lot. It explains why you can 'know better' but do it anyway---one part knows the long-term goal, another part has a well-trained routine that fires first. It explains why changing behaviour is hard---you're not just retraining one thing, you're renegotiating a coalition, and the losing parts don't disappear; they keep proposing until you give them something else to do. It explains why talking to yourself actually works---personifying the parts gives you a control interface, a way to surface the dissent and broker a deal.
So what can you do? First, stop pretending there's a single 'true self' that just needs to try harder. There isn't. Progress comes from better coordination and role design. Clarify what each part is for. Give the parts that are causing problems good jobs---alternative routes that serve their underlying goal without sabotaging the coalition. Surface the losing proposals instead of suppressing them; name them, understand what they want, and either integrate them or retrain them. And use context to seat the right delegates: environments cue which parts show up and which get to vote, so design your contexts to favor the coalition you want.
This isn't a dual-process story---not just 'fast and slow' or 'System 1 and System 2.' There are many agents with different specialties and speeds. It's not about chunks and links either; those describe how representations are structured. This is about decision-making as negotiation, about how the brain arbitrates between competing proposals when cues activate multiple routines at once.
Let's look at the neural architecture that helps us understand this heuristic better.
What neural architecture makes this happen?
These neural systems underpin this heuristic:
The neural hardware
The underlying neural systems that make this heuristic possible
Hierarchical Control
The brain's control systems are layered: higher levels set goals and constraints, lower levels implement sequences. At each level, there's arbitration---deciding which routine gets to run when multiple options are competing. The 'parts' in the society of mind are these competing routines at various levels of the hierarchy, each lobbying to be the one selected.
When you feel torn between going to the gym and staying in bed, you're experiencing this arbitration in real time. The high-level goal ('be healthy') biases the selection, but the low-level routine ('bed is comfortable') has strong infrastructure and makes a compelling proposal. Whichever proposal wins depends on the current task set, the environmental cues, and which parts have been practiced most.
Key takeaway: hierarchical control is where the society negotiates; higher levels bias, lower levels propose.
Task Sets
Task sets configure which parts are 'in the room' for the current decision. When you shift into 'work mode,' certain routines become active and others get suppressed. The parts that show up to the negotiating table depend on what set is loaded.
If the wrong set is active, the wrong coalition assembles. You meant to work, but 'browsing mode' is still loaded, so the busywork parts dominate and the deep-work parts don't even get a vote. Preload the right set to seat the right delegates.
Key takeaway: task sets determine which parts attend the negotiation; design cues to load the intended set.
Circuit Reuse
The same neural circuits support multiple routines, which means the 'parts' aren't cleanly separated. They share infrastructure. When one part updates, others that share those circuits also shift. This is why giving a part a 'good job' often works better than suppressing it---you're redirecting shared resources instead of fighting for exclusive control.
It also explains why parts don't neatly turn off. They're not independent modules; they're overlapping coalitions of shared sub-routines. The part that wants to stay in bed shares circuits with the part that wants to rest after work, and with the part that values comfort. Negotiate with the coalition, not just the surface routine.
Key takeaway: parts share infrastructure; retraining one affects its neighbors.
Plasticity & Stability
The more a part runs, the more stable and entrenched it becomes. Well-practiced routines build infrastructure that makes them faster, louder, and harder to override in the moment. This is why some parts dominate the negotiation even when you 'know better'---they've been practiced so much that they propose actions before other parts even finish loading.
To shift the balance, you need to practice alternative routines until they're stable enough to compete. The old part won't disappear, but the new one will get a seat at the table.
Key takeaway: practice determines which parts dominate; build infrastructure for the coalition you want.
Referenced by
Sources
- analects/overengineering-calming-down.md
- analects/positive-intelligence-context.md
- analects/positive-intelligence-content-pq.md
- analects/positive-intelligence-content-brain-science.md