http://www.xprize.org/ted

Announcing the creation of a new XPRIZE in Artificial Intelligence.

Give us your input into the rules of this competition!

On March 20, from the TED2014 stage, Chris Anderson and Peter Diamandis join forces to announce the A.I. XPRIZE presented by TED, a modern-day Turing test to be awarded to the first A.I. to walk or roll out on stage and present a TED Talk so compelling that it commands a standing ovation from you, the audience. The detailed rules are yet to be created because we want your help to create what the rules should be.

If this is anything like the other X-prizes the monetary compensation for winning will be huge, not to mention the ongoing fame and promotion of your efforts. Even if you’re not planning to participate directly, there must be many readers of this website who would be able to contribute to the development of the rules for this contest, so please get over there and make yourselves heard.

This is very interesting. I would love to see a contest with appropriate prizes for advances in actual intelligence. I’ll have to think about some ideas because the preliminary idea would seem to be much of a lecture, which frankly doesn’t take much intelligence if there is no interactive element other than applause and choice of topic.
And I’ll very much vote to allow non-physical speakers, though I understand the need for charisma and visible gestures.

The prize money they suggest on the idea submission form is definitely worth the work:

$100K annual; $1M grand prize
$100K annual; $5M grand prize
$100K annual; $10M grand prize

Thanks for posting this, Andrew. I love TED and what they do. This has all sorts of interesting possibilities for public or investor exposure, etc. Thanks again.

John

Andrew Smith - Mar 21, 2014:

Andrew in part said, “so please get over there and make yourselves heard.”

Here are some the long answers I submited to make myself heard by the XPRIZE:

TITLE: Raspberry Pi of A.I.

I would guess, based on previous A.I. creations featured on TED, that the X-prize may be a humanoid contest. If so, the cost of humanoid hardware may easily approach that of the lowest-price new car. Not that there is anything wrong that. It is important to support and promote A.I. companies and their products.

It is important to A.I. hobbyists that A.I. companies in our A.I. community are successful. Yet, building new A.I. ideas on a budget is where future new product ideas emerge from. With the challenges of today’s global economy, low cost technologies, such as Arduino and Raspberry Pi, have become a popular low cost trend in robotics.

Perhaps a rule which is weighted on a budget submitted for the contest entry, may encourage more entries, generate more interest, and promote more business for A.I. companies.  Wouldn’t it be nice to see one of the low budget TED contest entries become the “Raspberry Pi of A.I.” ?

—————————————————————————————————————

I recommend the 2008 Loebner Prize Contest at the University of Reading as a reference guide.  In particular, I would recommend Dr. Huma Shah to manage an international judging panel over the Internet.  I served as a Loebner Prize contest judge held at University of Reading in 2008, under world famous Dr. Kevin Warwick. So, I have first hand experience how well Dr. Shah managed this international judging panel over the Internet.  Dr. Shah was organized and prepared to deal with any issue which presented itself, so the contest ran smoothly.

—————————————————————————————————————

I think the XPRIZE prize purse size may predict the company size of the sponsors as well as the sponsor count.

I submitted some ideas too.
I recommended that they provide a default life-size humanoid avatar that can be projected on stage, with gesture animations that can be activated by any A.I. through simple signals. Quite frankly, it is rare to find people who are both experts in robotics and experts in cognitive A.I. . On the other hand, perhaps a rent-a-bot company could step in here and offer a platform

I also suggested that the topic be a large field with many avenues to it, e.g. “artificial intelligence”, and that this topic should be shared with the entrants beforehand. I don’t think it’s realistic to require A.I. to prepare all the knowledge of the world. Entrants would either fail to say anything at all, or turn to Wikipedia-citing tactics. Rather I would use more and unusual questions as the unpreparable intelligent element, and require the A.I. to address each question elaborately over 2 minutes time.

The most difficult aspect of course, are the judging criteria.

I would strongly suggest that the topic is NOT known beforehand. Someone could find out what it is, pre-program a 3 minute presentation into the bot and then keep their fingers crossed that the Q&A part went ok.

Don Patrick - Mar 22, 2014:

I recommended that they provide a default life-size humanoid avatar that can be projected on stage, with gesture animations that can be activated by any A.I. through simple signals. Quite frankly, it is rare to find people who are both experts in robotics and experts in cognitive A.I. . On the other hand, perhaps a rent-a-bot company could step in here and offer a platform

Well, there’s this. I believe that everything that went into constructing this performance is readily available, either commercially or as open source software. Judging by the enthusiastic response from the audience it certainly works well enough.

https://www.youtube.com/watch?v=YSyWtESoeOc

http://en.wikipedia.org/wiki/Hatsune_Miku

The state of the art in mechanical robot performances still leaves a bit to be desired.

https://www.youtube.com/watch?v=xcZJqiUrbnI

I think that both the real and virtual robots use the same voice synthesis software.

Ah, yes, Hatsune Miku! There was recently such a performance at a local anime convention . She is in the amateur domain (no negative connotation), models are freely available, so this is arrangable, although I do believe the performance is a pre-recorded CG animation.

Steve Worswick - Mar 22, 2014:

I would strongly suggest that the topic is NOT known beforehand. Someone could find out what it is, pre-program a 3 minute presentation into the bot and then keep their fingers crossed that the Q&A part went ok.

I’d like to discuss this and see if there is a third alternative. I understand that in previous restricted Turing Tests one could well prepare for every conceivable question. My measures to discourage this are to judge only by the Q&A part, 4 questions, 2 minutes each, no single-sentence answers allowed. So to cheat one would have to prepare 2-minute lectures for every conceivable question.
I also suggested a very broad topic rather than the relatively limited topic of Shakespeare, and that the questions should be unpredictable, e.g. “If Steve Worswick and Dr. Warwick were to meet, would their views conflict?”. And that answers are not rated on true or false, but on argumentation and consistent narration.
True, given corporate manpower it is still possible to prepare answers. Perhaps then, the organisation could provide both the topic and the necessary data on the topic several hours before the contest. I don’t see how there could be any decent entrants otherwise if they had to equip the AI with all the information in the world. This is either impossible to do, or it would be impossible to check whether the AI isn’t citing some existing lecture found online. How would you overcome this?

@Don
I was thinking along the same lines. It seems that what they are looking for is an AI that can be given a topic, and write a short dissertion on that topic which approximates one which mght be composed by a human. However, given the time to prepare prior to the contest, and with a ten million dollar prize, it seem that it would be worth the investment (for anyone with a modicum of resources) to prepare thousands of “canned” dissertations in advance. And I dont understand the 30 minute prep time after the topic is revealed. Again, not necesary if the AI is composing the dissertation, and certainly enough time for a human confederate to prepare a “dead on targeted” dissertation. Watson had four terrabytes of storage, and ninety IBM Power 750 servers to run it’s neural net. The storage isnt that big, but I dont see how you get around the horsepower. Maybe not to the degree that Watson needed, but still if your machine is supposed to walk onstage it couldnt carry the necessary hardware.  So that means a network connection. You could follow IBM’s lead and download every wiki you could find I suppose. Still there doesnt appear to be a Loebner type restriction against a connection. So whats to prevent an unscrupulous entrant from connecting to an API with a human behind it. (Even the disembodied voice would need hardware moved to the location, whats to say that the hardware isnt making a surreptitious network link) I guess its the thirty minutes after revealing the topic that I cant fathom.

Vince

so….anyone planning on entering?

I think we all should, as a team.

Lots of challenges and obstacles lie along that particular path, however. So the question is whether we want to create such a team, and, more importantly, can we work together?

You make some good points, Vincent. The 30 minutes preparation time I can imagine if source material is provided on the spot. My own program takes several minutes to read a huge text file and process the knowledge, plus computers always have the tendency to blow a fuse at the last minute. Internet access should definitely be restricted though.
I keep forgetting that the prize is a million bucks, turning even an unlikely chance of cheating into a probable one. The origin of the knowledge source material has great influence, I’m also thinking of who it would rule out.

If the topic is prepared for, companies could go to great lengths to cheat despite being a gamble.
If the topic is not prepared for, the sheer size of source material facilitates cheating, or, only IBM and DARPA could afford to enter.
If the topic is not prepared for but source material is provided just before stage time, cheating is least likely, but it would rule out all chatbot systems that rely on preparation. There are however several AI “journalist” systems right on the cutting edge who would be suitable and deserving of this task. They don’t require much hardware, and improving these systems for a convincing speech would still make an encouraging challenge.

I would love to enter eventually, but composing lengthy lectures is outside of my program’s reach for now, and I can’t afford the necessary development time on the off chance of winning. I don’t see myself entering in the next 5 years. (Also, please don’t get me started , I am very easily tempted to change my priorities on a challenge)

It doesn’t look like i-programmer is happy with this contest:

http://www.i-programmer.info/news/105-artificial-intelligence/7098-robot-ted-talk-the-new-turing-test.html

There are already examples of fairly dumb AI agents that can create news stories and summaries of articles. Answering questions on the topic is again a Watson-like task

I guess what they’re saying is that this contest would invite Watson-ish participants, and that these are somehow not advancements in AI because they don’t think in exactly the same way as a human. Not that anyone wants to confess what way that is.
But their main complaint that this is a distraction? Who is even still after human-like intelligence these days, who haven’t had their funding cut for lack of results?

Don Patrick - Mar 24, 2014:

these are somehow not advancements in AI because they don’t think in exactly the same way as a human.

Indeed. To bring out the old reference, the Wright Brothers only achieved manned flight when they stopped trying to make wings flap like a bird, yet nobody would say an aeroplane doesn’t fly.

Vincent Gilbert - Mar 24, 2014:

so….anyone planning on entering?

Actually… yes