The ability to cope with uncertain environments has been put forward
as one the key strengths of AI methods as opposed to more traditional
computer science approaches. In recent years, a number of application
domains (in particular those involving digital communication and
computational decentralisation) have introduced the autonomy of other
so-called intelligent agents co-inhabiting a common environment as an
additional source of uncertainty. In this talk, I will argue that
beyond the hype around agent technology, autonomy differs substantially
from traditional views of uncertainty, and that it calls for the
development of genuinely new methods for managing the interactions
between autonomous agents. I present an abstract framework that
can be
used as a starting point for designing such methods and
illustrate key
concepts using examples from the area of strategic learning of
communication patterns.