Report of the working group on the interaction between social and cognitive processes at the AAAI 2000 Fall Symposium on Socially Intelligent Agents - the human in the loop, North Falmouth, MA, USA, November 2000.

Participants:

The appropriate level of modelling of social and cognitive processes, that over-simple models can be misleadingly effective in the very short term but over complex models are difficult to generalise from
The need for a rich and complete data source documenting a few real social processes rather than many field studies of different processes, this would enable the detailed modelling of the same social process from different perspectives
The learning vs. the design stance - whether we attempt to explicitly design in social intelligence or whether it needs to be learnt in situ - maybe we need to learn how to train social agents
The inevitable unevenness of the learning curve by academics in this area - sometimes we have to unlearn things for progress
That researchers should be satisfied with less impressive results and focus more on deeper progress - specifically quick hacks to achieve outwardly more impressive results (which are common in the field) should be avoided
The tension between the desired generality for knowledge and tools and the context specific nature of much social knowledge and skills

Can an artificial salesman achieve real sales
A restricted game of Diplomacy with a limited repertoire of allowed messages between players
How long can an artificial agent keep a human interested in a text-only on-line conversation