If the author's conversation attempt is any guide (and extensive experience working with chatterbots tells me it is) then one simply must question the sobriety of the 12 judges that mistook the bot for human. I mean ... seriously! They'd have had to be either completely drunk, in the throes of an ADD event or limited to holding discussions with high school freshman ... with ADD. There's (obviously) still a loooong way to go before chatterbots have any chance of passing the Turing Test in public.

That's my time... remember to tip your waitstaff.

But you're just not as cool as the Giant Killer Robots Controlled By Monkeys!

http://www.sciam.com/article.cfm?id=monkeys-see-monkey-do-wit&ec=ypi

(Well, hopefully that will be the next stage of their development!)

Well a stonerbot at any rate. 95% of Facebook could be marginally smart cockroaches or a digital thermostat.

I know the so-called interview in the article is a spoof, but its echo response reminds me of that very, very old program "Eliza" which offered an alternative to live human-to-human talking therapy.

Eliza basically took the human's every comment and turned it into a question, e.g.,

Human: "I feel sad"

Eliza: "Why do you feel sad"

etc.etc.et cetera

I would love to have a chance at a conversation with Elbot.

http://www.wordreplacer.com/cache/g081013234430495813

That transcript wouldn't have fooled me. Were the judges humans or computers?

I find the subject of AI experimentation to be fascinating, tho.

The thing that bugs me about this contest is that the Turing Test refers to a very specific form of intelligence, and one which is not likely to be an important or useful form.

Yes, it'd be cool for a computer to be able to talk like a person, but is that really a valuable use for a computer? Some people have no friends and could perhaps buy a simulation, but most everyday human-human conversation is about as stimulating as drying sponges in the first place.

AI probably has the most value in applications like electronic "sense-data" interpretation, high-level modeling of complex systems, and telecommunications switching.

Vast processing power and tracking of millions of different variables without error is well-applied to those kinds of problems.

One thing I enjoyed about Neuromancer and/or Mona Lisa Overdrive was that there were several totally "mundane" AIs that people used constantly without ever thinking about. "Continuity" was a favorite.* It was an AI that tracked everyone's location and activity (while in public) so you could ask it something like, "Where's Marco been the past 2 weeks?" and get an answer like: "Marco Politsias Bahwumy traveled to Rome on September 29. He stayed in room 1254 of the Winston Hotel. He ate breakfast in the hotel buffet with Albert Anders Vozhninkin at 10am October 1st...etc, etc, etc."

I suspect our first "true" AIs will be nothing at all like us in character, and will perform functions that a human never could. It may not even occur to us that they are intelligences for a little while after their birth.

------

* Continuity was a true "intelligence" in that it undertook actions of its own volition and constantly adjusted its own code. One of its hobbies was writing a complete history of human activity in hypertext, and it was near completion, but it couldn't finish because people kept taking actions. A human in one of the books puzzles over it for a moment, but "Continuity was an AI, and AIs did things like that."

And I'm afraid that assuming that the reports are true that 25% of the judges couldn't tell that Elbot wasn't human after 5 minutes then it doesn't reinforce my confidence in the advancement of humankind as a species. It only took me about 3 lines to know for certain that it wasn't a person. Maybe if I was specifically told that the text was coming from a person and I just had to chat with that person for a few minutes for a test, then I might take more than a few lines, but Elbot isn't any better than chat bots were 15 years ago.

Because I think the possibility to create accidental poetry here is tremendous. Given the right questions and responses from the human, who knows what crazy shit the bot will come up with. Could be Vladimir and Estragon all over again.

Seriously, if you have a link to a page where just anyone can interact with the Elbot -- or a bot much like it -- please share.

Also, a question: would the most ethical thing be to attribute the final product to the bot's makers?

If this transcript is typical of Elbot's conversations, then the three judges who were fooled must be complete morons. Or maybe the members of the human control group were instructed to chat in non-sequiturs, in which case this isn't much of a test. Either way, this is a highly unconvincing, poor application of the Turing test.

And as to some of the other commenters who referred to the "smartness" of the chat, comparing it to 15-year-olds on MySpace, it's not about smartness, it's about naturalness. It's relatively straightforward, given some input, to get a computer to spit out very "smart"-sounding output; the trick is making it sound natural. Conversely, however dumb some teenagers might sound while chatting, they certainly always sound natural.

Is there a mirror-opposite of the Turing test for humans? I have several old friends who I would swear are human beings (having known them since high school days, some 35 years ago), but get them talking about politics and they become indistinguishable from any GOP robot. There is no way that they would pass the Turing test.

(BTW, regarding Elbot's chat, I'm sure that, like Pat Buchanan's speeches, it sounds better in the original German.)

It sucks.

This isn't a chatbot at all, and certainly didn't win any contest. The sample conversation was a total joke and not funny. Looks like a boring not funny person invented it. I tried the actual 'chatbot' itself and it failed miserably on any input. Please look at 'Eliza' which is 20-30 years old and actually attempts to mimic minimal understanding of its input.