Communication Patterns in Flexible Problem-solving Systems

Speaker: Chris Holt

Abstract

This is intended to be an informal discussion considering the question of what we can learn from human behaviour as applied to communities of artificial entities, e.g. expert system and neural net hybrids. Since such communities are likely to be insecure, each entity should keep a record of trust and likely accuracy to determine which of its friends should be listened to for any given question. This is not Byzantine generals, however; its closer to zero-information proof-style confidence levels. The balance between broadcast, narrowcast and one-to-one conversations is considered.