Mike Tintner writes:
Let's call it the Neo-Maze Test.
I think this type of test is pretty interesting; the objection if any is whether the capabilities of this robot are really getting toward what we would like to consider general intelligence. For example, moving from the simple maze to navigating an office building involves new reasoning abilities such as understanding how to work an elevator. If that ability is programmed explicitly it is suspicious, but if it is somehow learned, that's certainly more interesting. In some ways this idea is similar to another easy-to-define approach to AGI with tangible intermediate goals: recapitulating phylogeny: Start with a fruitfly simulator and build a "brain" capable of passing the Turing-fruitfly test (well, not fooling the other fruitflies so much, but being able to flourish in the fruitfly's world by controlling a fruitfly's body). Then move on to the Turing-mouse test, the Turing-dog test, and the Turing-monkey test. The reason such an approach is distasteful to most AGI researchers is the opinion that it puts a lot of work into doing things that seem completely unrelated to the core task, and even once you have a simulated monkey, are you really very close to AGI? I don't know the answer. ----- This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=231415&user_secret=fabd7936
