Initial Themes
We looked at agents along three inter-dependent dimensions: (1) How the agents present themselves to humans; (2) What relations and ecologies they create within the contexts in which humans
use them; and (3) What infrastructures they need.
Questions and topics of discussion included:
(1) Agents: Human-likeness,
self-representation, and personality. What types of responses do
conversational agents give to ethical issues, and how does that influence our
expectations toward them? What types of questions are systematically avoided? How do they
present themselves and how aware are they of biases?
(2) Relations and
ecologies: Contexts of use, human and non-human relations, and ecologies of
interactions.What kinds of relations and ecologies do
conversational agents elicit through their interaction with humans, as well as with
other non-human agents? How do these relations change with shifting contexts of
use? What kinds of relations matter more to humans, and why? In what instances does the authority/power
of a conversational agent become visible, and problematic?
(3) Infrastructures: Training
data, security, privacy, and commercial interests. What material and
immaterial infrastructures, such as human labor, data, and planetary resources,
can be disclosed by using decentered forms of ethnography? How does the
disclosing of infrastructures challenge traditional divisions of design and
use? How could that help us uncover biases and their origin? What would it take
to design an unbiased agent?