A plausible weak AI to beat the Turing Test?

Is it theoretically possible that a weak AI can beat the Standard Turing Test?

Who cares. This is not philosophy, this is engineering. Let's plough on.

How do you eat an elephant?

Everyone who has read a self help book knows it's "One bite at a time".

But it also helps if you pick a really small elephant.

What is a weak AI?

Well, the strong AI definition refers to the aim to produce artificial intelligences that have mental states, and can be said to have a mind in a similar sense as a human being.

So weak AI is the realm of "clever stuff" that is not clever like a human being. That do not have mental states. The clever tools.

What is the Turing Test?

Hmm. I am not sure why I have this question 4th, if you have got this far and don't know that its Alan Turing's proposal for a game in which a computer pretends to be a human, and a human judge attempts to identify the computer, then I would be surprised.

Harrison Ford gives one to a replicated human in Blade Runner.

So let's find a really small elephant.

This would be some subset of all the possible conversations, restricted on some plausible manner.

  • What about a Turing Tests restricted to greetings? (too simple?)
  • Or some other structured interaction, say a product demonstration or sales interaction.
  • What about hijacking/intercepting some established instant messenger conversation for a short period, to see how long it takes the human to notice?
  • How about playing the Imitation game, in a MMRPG with avatars like World-Of-Warcraft?

Restricted Turing tests
It appears that they have been thinking about small elephants for a while and have already had a competition with restricted Turing Tests. Here is an example of one from google search that provides the following restrictions;
  • Limiting the topic: In order to limit the amount of area that the contestant programs must be able to cope with, the topic of the conversation was to be strictly limited, both for the contestants and the confederates. The judges were required to stay on the subject in their conversations with the agents.
  • Limiting the tenor: Further, only behavior evinced during the course of a natural conversation on the single specified topic would be required to be duplicated faithfully by the contestants. The operative rule precluded the use of ``trickery or guile. Judges should respond naturally, as they would in a conversation with another person.'' (The method of choosing judges served as a further measure against excessive judicial sophistication.)

//@TODO I'm still working on this post, thinking of plausible weak AI Turing tests.


No comments:

Post a Comment

Don't be nasty. Being rude is fine.