TITLE: Be Empowered: Guiding Adaptation through Potential Information Flows

SPEAKER: Daniel Polani (Department of Computer Science, University of Hertfordshire)

ABSTRACT:

The last years have seen significant progress in the development of
information-theoretic approaches for modeling behaviour of agents. In
this philosophy, Shannon information is not just used as a simple
measure of correlation, but in fact becomes the main object of
discourse - information is what an agent acquires through its sensors,
processes in its brain and then re-injects into the environment. The
form of these "information flows" can be formally quantified with the
help of Causal Bayesian Networks.

Their structuring gives a precise formalization to the intuition that
"embodiment" is important for the emergence of intelligence. In
addition, there is mounting indication that (Shannon) information is a
dominant resource in biology that organisms adapt and evolve to
optimize.  Thus, the informational view of agents in conjunction with
the Bayesian Network formalism provides a fecund ground for producing
models for intelligent behaviour that are both principled and
biologically plausible.

The talk will present the framework and introduce one quantity of
particular interest, 'empowerment' which is defined as the external
channel capacity of the perception-action loop and which has been
proposed as one possible candidate for a "universal utility" in
organisms.