Lets Create The worlds best Artificial Intelligence using Crowdsourcing

This is a seed of an idea I got from watching Luis Von Ahn the Brilliant creator of The Captcha you see on webpages and DuoLingo a language learning website.

Have a look at this video to understand exactly where this dude is coming from. In a nut shell, He's using peoples Captcha answers as a tool to help get books digitized when he realizes there is a problem - The computer makes mistakes due to it's not understanding of the language. This led him to the research he's using now that helps people learn languages quickly and easily. The video explains it all better than i can. Just watch then read the rest.



Now that you've seen the video - Here is my idea -

Why cannot we use something similar to create the worlds best Artificial Intelligence using data pooled from millions of internet users all over the world? Once you watch the video.. I think you'll see why I think this is possible.

The research seems to me goes in parallel lines with AI. I used the word pooled, but I'm not talking about data pooled from static sources, but rather user data fed into the computer on purpose to "teach" the AI much like how he uses the Captcha's. The AI would learn by asking the same thing to thousands of people and comparing answers so it can " learn" the correct way to answer, behave etc. This would all be done like the Captcha's in a way where the user doesn't have to jump through hoops or even have to know he's entering data for this purpose.

I'll let this thread sit a bit and if needed I will try to clarify.

Thoughts?

a reply to: JohnPhoenix

Crowdsourcing is probably the best way to entertain the creation of an AI, it tends to keep the end product away from the corporations and military who are always quick to usurp research for their own or hidden benefit.

The real problem however is in the definition of AI. A linear predictive, semi-randomized or structured machine can learn and it can formulate decisions based on it's learned history/programming. However is that AI? I'd say no, it is the emulation of an AI based on a very restrictive set of rules with no emotive based rationalization.

Do we define an AI as having a conscience, some set of programmed or learned rules, so it can make AI decisions somewhat like a human makes decisions? Again, no, it's still an emulation of an AI model, just with a slightly better rule set and with an emotive emulation base.

Maybe we could define an AI has having the capability of all five senses plus a conscience and full learning capability based on some very complex rule set as well as senses and a "conscience." No, it's still not a true AI, it is still an emulation of AI based on a system with sensory capability that uses an extremely complex rule set that utilizes and emotive base to provide additional decision making capabilities. In all of these cases, in order to create the emulation of AI, we must apply some or all of these rules.

Now, what if you took the last example and rather than apply a very complex set of rules, you used a few basic rules like humans, very limited in scope, in a sense so limited the AI had the intelligence, senses and learning capability of a new born infant and simply let it learn in much the same way a child learns. That might, just might, produce a true AI if there is no unseen human-human connection. It would take years for the AI to learn and in the end you could have anything from a conscienceless monster to Mother Teresa or Jesus Christ or some thing that called itself Bob, but if you made many, you might just get exactly one, that worked the way you wanted. Since the AI is a machine, then you could download it's memory and replicate it to mass produce many more application safe AI machines.

I think crowdsourcing is the answer to this endeavor, however it is a very expensive and time consuming program.

ETA in going over your OP, after the AI has developed in the last scenario and is capable of exponential learning, yeah, throw them all onto the internet and see what happens. There is a reasonable probability that it would create a better AI, which in return would create an even better AI, ad infinitum.

Cheers - Dave