Turing Tests in the Creative Arts

[Hugo Gonçalo Oliveira]

Hello,

Most of you are probably aware of a competition that is going on, focused on Turing Tests for computer generated stories, poetry and DJ sets: http://bregman.dartmouth.edu/turingtests/

I believe some of you have very strong opinions about this kind of competition and this forum is probably the right place for expressing them and discuss this topic.

So I am inviting you to start this discussion.

I look forward for reading your opinions.

[Joe Corneli]

Hi Hugo,

I think possibly the most exciting Turing(-style) test taking place
this week is Lee Sedol vs AlphaGo. It's not positioned as a Turing
test, but some of the discussion around it seems to fall into that
genre.

~~~

From a discussion of earlier games: https://youtu.be/NHRHUHW6HQE?t=4m43s

"This AlphaGo plays just like a human. But everyone knows that. I'll
talk about the ways in which she played unlike a human."

"Do you think maybe her training data was biased towards Japanese
professionals?"

"Her weakness -- she plays too soft, and doesn't know the value of the
sente (the biggest move that's left on the board)."

"She played almost like a human, I just see some mistakes."

~~~

With a game like Go things are constrained, and "playing well" is
likely to seem synonymous with "playing like a human." But of course
most humans are not playing at this elite professional level, and they
do make mistakes.

The goal of AlphaGo is not to fool people into thinking that it is a
human -- the goal of AlphaGo is simply to win.

And I think a lot of "Turing tests" in practice turn out to be the same thing.

-Joe

[Oliver Bown]

Hi all,

indeed I think the community has expressed in many different and eloquent ways the following:

Sure why not have Turing tests, but what we care about is the *rich* description about the systems and how they are perceived by humans, rather than an opaque ‘yes-no’ research output (either it did or it didn’t trick the human). The latter contains not much useful information, whereas the former gives us lots of new research data to chew on (how and why did it achieve these results, etc?). Insofar as we welcome Turing Tests, we should encourage "rich-description Turing Tests”, rather than opaque Turing Tests (rich description is a concept from anthropology that rejects simple theory testing and promotes detailed reporting of any experience or environment, for the purpose of advancing knowledge).

I think this was captured very well by the late Marvin Minsky, reporting on the recent supposed Turing Test “win” last year (or two years ago… what was the name of the agent… Eugene someone?). Minsky totally hammered the handling of the interrogations. The transcripts were all about which was your favourite music etc. Minsky says: “ask whether you can push a cart with a piece of string, and if not why not. Then you know if the system is intelligent”.

Ollie

> --
> You received this message because you are subscribed to the Google Groups "Computational Creativity Forum" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to computational-creativ...@googlegroups.com.
> To post to this group, send email to computational-c...@googlegroups.com.
> Visit this group at https://groups.google.com/group/computational-creativity-forum.
> To view this discussion on the web visit https://groups.google.com/d/msgid/computational-creativity-forum/CAN%2BqofmB-Fa-4APQ6V18X0%2BzqRneAi2LLQSR68xRq%2Bg13TsmFA%40mail.gmail.com.
> For more options, visit https://groups.google.com/d/optout.