search

openworm / behavioral_syntax

behavioral syntax analysis based on the paper of Andre Brown
Other
5 stars 1 forks source link

Turing Test assumptions?? #40

Closed ghost closed 7 years ago

ghost commented 8 years ago

The Turing Test assumes that it's sufficient to test for some terminal state(human intelligence) in terms of ability and assume that the machine is intelligent. But, if intelligence is the ability to adapt rather than a certain level of intelligence then we should be testing for the ability to evolve rather than obsessing over some measure of similarity.

MichaelCurrie commented 8 years ago

Possibly relevant: here is a paper discussing the Turing test as it relates to worms: https://github.com/openworm/open-worm-analysis-toolbox/blob/master/documentation/Movement%20Validation%20White%20Paper.md

ghost commented 8 years ago

That paper addresses exactly Turing Tests, which I have doubts on. A strong Turing Test wouldn't simply be binary...it would serve as a metric so that we can get an idea of how close the OpenWorm project is getting to its final destination. But, is there a final destination?

A complementary measure which might be useful is some general measure of adaptability or 'intelligence'. I think that Shane Legg of DeepMind has some interesting ideas on how we might do this: http://www.vetta.org/documents/Benelearn-UniversalIntelligence-Talk.pdf

This way, we have a measure that's forward-looking i.e. it anticipates the ability of the worm to adapt to new environments, beyond those that have been studied in a biology lab.

ghost commented 7 years ago

Probably my final comment: there is no such thing as a perfect Turing Test. If we think of it as a distance function that maps agents to values in (0,1) in theory you can come up with an infinite number of fake worms that game the test. In theory there's nothing that would prevent the test from being almost always wrong unless we have reliable prior information about the set of possible digital worms.