Addressing some things sequentially:
#1 - “I don’t think we(humans)need to operate at agent speed and don’t need agent-generated reputation. They should only be using to check the reputation that we create.” - I think it’s both. But I very strongly believe we are years away from programmatically computing reputation in a way that could be completely offloaded to machines; there are too many variables at play.
As such, I think human intuition is an essential piece of the puzzle - we need humans in the loop in the domain of reputation. I could make an argument for always, but definitely right now.
This doesn’t negate the need to analyze how we, as humans, derive intrinsic reputation scores for things; it’s actually the opposite. We should analyze our own behavior and try to understand ‘why’ we think the things we think, so that we can make progress on the front of outsourcing reputation generation to machines / so that we can come up with a meaningful variable set.
#2 - “every assertion would attach to an anchor:” - I don’t necessarily think EVERY assertion needs to attach to an anchor (there needs to be ‘root anchors’, for example), but I REALLY like this idea of most claims attaching to an anchor! This is quite interesting and is definitely a pattern that we should all explore more… Every action we take / every decision we make can be tied back to SOMETHING… or some SET of THINGS… This feels like a very powerful pattern to follow… Especially in the context of AI, where we need to trace and learn from decision paths…
#3 - @repboiz I think everyone here is bought into the concept of contextual reputation / understands its importance. And, that is definitely where any sort of architecture needs to start - we cannot say ‘what is the set of criteria that make Actor X reputable’; we need to be saying ‘what is the set of criteria that maker Actor X reputable in Y Context’. BUT, universal reputation CAN be a useful concept, despite its reductionist nature. This universal reputation - or rather, higher-order reputations (can compose together smaller-context reputations into higher-context reputations in a kind of infinite fractal) just need to be compositions of lower-order contextual reputations that logically roll up.