The secret to effective Turing testing is to ply agile reframing and see whether the AI keeps up. But it won't. This will always be its limitation.I was being glib, and wound up saying something dumb.
It's hard to decide when a thought is mature (much of what I post here on the Slog has been incubating for 40 years or more). When an insight seems fully-baked, even after years of digestion, a greater clarity may appear in the very next microsecond, leaving you aware of how ridiculously incomplete or wrong you'd been.Well, half-dumb, anyway. Challenging an AI to reframe is indeed the smart way to trip up an AI. The problem is that most people have frozen perspectives. Having forgotten that they have the ability to reframe, they'll fail the test, as well. My Turing test strategy rejects all but the most creative people.
It's not like waiting for batter to become cake. It's all eternally batter, making it madness to ever write anything down (the Hindu vedas have a story about this; see the indented portion here).
The dismaying question is this: should I try to think of a better Turing test strategy - one that never rejects humans - or is this one right after all because most humans allow themselves to become as flat and square and inflexible and blinkered and machine-like as machines and therefore shouldn't pass?
How lithe is Margaret Dumont's reframing ability in this clip?
2 comments:
Turing test assumes that humans possess human intelligence. If some (or many) people are effectively unconscious, then distinguishing human from machine becomes a fairly useless enterprise, and the interesting distinction is elsewhere, as you say.
Distinguishing creative people from uncreative people and machines might be an interesting exercise, but it's not a Turing test. A Turing test distinguishes human intelligence from artificial intelligence, period, including unsophisticated human intelligence.
Post a Comment