Should AI memory store thinking patterns, not just facts?
Most current store descriptive facts: , saved notes, conversation summaries, things like 'this user is interested in economics' or 'this user works in engineering.' These help the system recall information but don't capture how a person actually thinks. An alternative idea is to have continuously refined to infer higher-level patterns instead — recurring explanatory frameworks, preferred abstractions, and characteristic reasoning styles. For instance, rather than storing 'interested in economics,' the system would infer something like 'this user tends to explain economic outcomes through incentives and institutional constraints,' or 'this user understands complex systems through interactions and rather than by analyzing individual components in isolation.'
Key points
- Most current s store descriptive facts and preferences directly
- Proposed alternative: infer higher-level patterns like explanatory frameworks and reasoning styles instead
- Example: instead of 'interested in economics,' infer 'explains outcomes via incentives and institutional constraints'
- Still a conceptual discussion — no working or evaluation yet