Blue Sky Agents
- I think the best way to get to full fledged AI (Turing test
passing artificial general intelligence) is to follow the human
model. Use neurons, build embodied agents, and build cognitive models.
- I also think that's not likely to happen in the next 20 years.
- Blue sky thinking for the next 10 years: we should be able to build
complex and useful neural agents, that learn about their environements, and
interact with people.
- Lots of possible agents come to mind including robots, tutorial agents,
and machine learning assistants.
- However, in the rest of this talk, I'm going to focus on agents for
architecture, because
they work nicely in virtual environments, and could be used both
for the architects and their clients.
- Try to imagine what these agents would look like.
- They'd "understand" about the virtual building or buildings
that are modelled.
- They'd interact with the users (architects and their clients) via
a conversation, and in other ways (e.g. pointers).
- They'd learn about the building, architecture, and conversations.
- They'd be in this restricted domain so would not need to learn everything;
it would be good if they could say, "I don't know" in the conversation
when they didn't know.