Solve problems like people do
- Use Human-like components (simulated neurons)
- Keep in touch with reality by
- Solve problems like humans do (cognitive models and
architecture)
- Develop interesting systems (AI programs)
- Integrate Learning
- Take advantage of emergent algorithms
- Repeat
- One way to keep in touch with reality is to build
cognitive models.
- Note that a good neurally implmented cognitive model of human
memory would resolve the stability-plasticity dilemma.
- Of course there are a vast range of other aspects of cognition
that could be modelled.
- I've just done NL parsing (imperfectly), there's the Stroop task,
vision, attention ....
- Another constraint would be to have a single model that implemented
a lot of tasks.
- If it could be claimed that it could implement all tasks, then
it would be a cognitive architecture.
- This is in the spirit of ACT and Soar, but redresses some of their
problems.
- It's also in the spirit of Rolls, though his systems aren't unified.
- If we could build a neural cognitive architecture, that really did
work, then we'd have an AI.
- Of course that's difficult, and advancement can simultaneously be
made on other fronts.
