Deontic Sensors

19^th May 2016, 12:00 Ashton Lecture Theater
Julian Padget
University of Bath

Abstract

Procedural programs do what we tell them, but not always what we want them
to do, especially when presented with unexpected inputs (that we did not
think about). We assume that such tight control reduces risk, whereas giving
a program more autonomy is scary: ``who knows what it might do?!?!''.

In principle, some autonomy to choose an appropriate action ought to be at
least as good and possibly better, by allowing greater resilience. But a
program's capacity to understand its environment is limited to what the
designer knows or can foresee, which is no better than where we started.

To provide up-to-date interpretation of (aspects of) the environment, we
propose ``deontic sensors'', based on a formal model and realised through
Answer Set programming. These sensors observe program actions and provide
advice on what a program (agent!) can, ought and ought not to do and
illustrate the concept with examples from a variety of socio-technical
system demonstrators.