Simple question, maybe you guys can help.
So we all know the Turing test - if you interact with an artificial intelligence, and you can't tell if it's artificial, it passes the test. We also know this doesn't mean that AI we ware interacting with would actually be a true knowing intelligence, thanks to the Chinese room thought experiment. We can also observe this with very good chat-bots - they obviously have no knowing, since it only regurgitates based on statistical analysis.
So I was wondering, is there an externally observable metric any of us can come up with that would allow us to gauge if an AI has an actual knowing understanding, in stead of giving logarithmic answers?
The highly advanced AI's like The Akinator are so sophisticated and beyond our comprehension at the moment that it certainly is scary.
Akinator can read minds at this moment and he has been around only a few years and it certainly suggests that he may in fact possess a consciousness/telephatic abilities.
>>896016
Make it play Go.
I'll take a first stab at it:
Doing something unprompted. If something comes out we didn't program as output. Like a chatbot that suddenly starts giving us mathematical proofs, some kind of indication it wants to do this activity.
>>896032
Only if you're a complete simpleton. The Akinator website gets hundreds, if not thousands of unique hits every single day due to pure word of mouth. It doesn't read minds, it has just learned how to manipulate queries and answers in a way that makes it seem all-knowing or psychic, when it's simply following instructions and user inputs.
>>896016
>the Chinese room thought experiment
Doesn't translate into reality because no machine can be programmed for every contingency. It would need to be flexible.
>>896016
What is the difference between "actual knowing understanding" and seeing mathematical patterns in something?
>>896096
ascribing meaning to it, not reacting mechenically
>>896145
Those aren't mutually exclusive.
turing hypothesized that test because there is no test that can absolutely ascertain such things, or rather it is not relevant
the turing test is not just about artificial intelligence
he came up with it as a wager type thing in which he says if there was a machine that was deemed necessarily intelligent then it could fool a person so it is not about human qualities about intelligence or understanding or consciousness and so forth. it is about what a machine is seemingly capable of doing or rather wether or not it is capable of fooling people into thinking that it can "understand"
because these attributes are meaningless in the machine context, it is what we are willing to perceive and not what the machine is capable of
in other words when a machine can produce answer like a human as if it was capable of understanding then that's about as far as you can get about machine "underdstanding"
>>896059
>taking the bait