On Mon, Nov 29, 2010 at 3:37 PM, Abram Demski <abramdemski@...>
Here's a silly exercise: what's the analogous test to the Turing test for the insect domain? Does it have the same flaws? Worse flaws?
Not a silly test at all I think, since insects display a good number of adaptive traits and problem solving abilities. It could be a full-body Turing test, putting the virtual insect in certain environments and expect it to survive through challenges.
Insects trying to judge between insects and robots sounds possible (ie, see if our fake bee can infiltrate the hive), but does not fit the spirit of the Turing test in that Turing wanted to rule out the need for robots; and furthermore, the insects don't know what's coming, so they aren't proper judges-- they will accept a bee-like thing merely by default.
Or humans distinguishing a virtual insect's behavior, from that of a real insect hooked up to the virtual environment through a cybernetic connection.
Humans judging insects from robot insects fixes the 2nd problem, but not the 1st.
Ok, so did I solve the first? :)
Perhaps though instead of the virtual environment, you could define more specialized problems. It seems that performing these tests in a rigorous way would be more difficult than making the sim-insect run though. I think to make my idea work in a cheap way, you have to abstract the problems that the insects can solve into abstract machine learning problems, and expect a single algorithm to solve them all. That's something I have been saying for ages, luckily I am now working on just such an algorithm!