Page 2 of 2
Re: Bruce Wilcox's AI "passes" turing test
Posted: Thu Nov 04, 2010 12:56 pm
by flOvermind
Re: Bruce Wilcox's AI "passes" turing test
Posted: Sat Nov 06, 2010 12:33 pm
by Time
emeraldemon wrote:Time wrote:"What do you think about the world series?"
"I don't really follow the world series"
"Do you feel left out when everyone else is watching sports?"
"Not really"
This sort of response would work for any questions of the form "what do you think about X?" or "do you X?"
I think this would get noticed pretty fast:
"What do you think about being here?"
"I don't really follow being here"
I mean, do you really expect me to come up with a solution to this in 5 lines of explanation?
I'm pretty sure if I was programming and studying AI for a living I could find a solution. As for this particular example, I guess you should preface X by "the." But do we really have to nit it up this hard? The point is basically that if you're studying and programming AI for a living, you can probably fill most holes that are easy to exploit, since you're obviously going to be testing and thinking about those first.
Re: Bruce Wilcox's AI "passes" turing test
Posted: Sat Nov 06, 2010 1:11 pm
by willemien
kirkmc wrote:What a bogus test. You can only ask about certain topics. Of course it's eventually easy to get there. Ask the bot what it thinks of the World Series or the Sopranos and it fails instantly.
not for me i don't know anything about any of the two
(or is that enough proof that i am Human?
Re: Bruce Wilcox's AI "passes" turing test
Posted: Sat Nov 06, 2010 4:27 pm
by hyperpape
@time Why assume that you can fill up the holes that are easy to exploit? Human languages allow an infinite number of meaningful sentences, of which an average human can process a gobsmackingly large number. Why think any solution based on responding to isolated tricky questions can work, without producing something approximating intelligence? That's the primary reason the contest is interesting--simple strategies of the type that are floated here are thought to be unpromising.
Re: Bruce Wilcox's AI "passes" turing test
Posted: Sun Nov 07, 2010 10:20 am
by flOvermind
Time wrote:I'm pretty sure if I was programming and studying AI for a living I could find a solution. As for this particular example, I guess you should preface X by "the." But do we really have to nit it up this hard? The point is basically that if you're studying and programming AI for a living, you can probably fill most holes that are easy to exploit, since you're obviously going to be testing and thinking about those first.
http://xkcd.com/793/
Re: Bruce Wilcox's AI "passes" turing test
Posted: Sun Nov 07, 2010 2:38 pm
by Mike Novack
Possibly should note:
There are various versions of the "Turing Test" (and in the last decade or so a number of proposed alternatives). Leaving alternatives aside, consider that there is a very big difference between ........
a) You are asked to chat to two clients. You are told before hand that one of them is an AI. After the chats you are asked to identify which.
b) You are asked to chat to two client. You are told afterwards that one of them is an AI and you are asked to identify which.
Now "a" is a much more difficult test for an AI to pass. But if it can pass "b" (not identified significantly more than 50%) would you say that it has failed the Turing Test?
How about a "c" version? You are asked to talk to two clients and told before hand that both may be human, one human and one 'bot, or both 'bots. In the scoring, false positives (acutally a human but identified as a 'bot) used to adjust the identification scores for the 'bot.