[un]

True, they are applications of machine learning techniques and can be considered "narrow ai". What I was suggesting is more along the lines of something that can do the tasks of children, see and recongize most things, and understand language. I think one way of achieving that is by "bootstrapping" from understanding of video, If an understanding of what a human does in the video (standing, sitting, limb positions, height, facial expression) in relation to objects and other humans can be automatically inferred, a predictive model could be built with enough data of what a person in any given situation might do. Combine this with robust speech recognition and understanding of the behaviour of common objects I think you would be closer to cracking the goal of language understanding (which I think defines true human level intelligence).

[showerst]

And i'd go so far as to say that this wouldn't just be good business, it would revolutionize human life.

But from what little I understand of the history of AI, people have been throwing resources at this category of problem for 40+ years and made only middling progress.

Not that that means that people (especially saavy entrepeneurs) shouldn't tackle the problem, I just think that by taking the constrained view (and shooting for the moon) you'd be missing a ton of low-hanging fruit in other areas.

With the emergence of commercially viable computing on demand we're tearing down a major barrier to AI, but I think that throwing the idea out and saying 'Hey! Why don't more of us get together and tackle this problem?' betrays a bit of an underestimation of just how tough this problem is.

That said, I would love for you to prove me wrong =) And as another poster mentioned, groups in the areas of surveillance and security (where the money is, at least initially) are making good headway on the problem.

The biggest problem with building predictive models, as another poster pointed out, is feeding them tagged training data. You could really feasibly load up a database with terabytes of data for most of the stimuli people react to these days (or will be able to in the near future), and even most of the basic combinations. It's a matter of tagging and classifying them in the first place that's the problem. Maybe with enough Indians sitting in front of mechanical turk... scrambles off to a calculator

[un]

That's exactly my initial reason for posting this. In the singulariy enthusiast community artificial intelligence is the most important thing we as humans can create (post scarcity, immortality etc).

What I was suggesting is that a "gold rush" type situation form around the building of classifiers, such as that occurred with website creation. There are millions trying to get an income onlne creating content for adsense and affilate programs (and for free with wikipedia and open source). Most will fail at generating reasonable income but in the process a large amount of wealth has been created. The same could be done for classifiers/artifical intelligence. It would require just collecting a thousand or two pieces of data (image/sudio/text) to create a classifier, an effort similar to creating a small website.

The problem is that even in the past there really wasn't a large number of people working on artificial intelligence, maybe just a few hundred or thousand, and computing power wasn't cheap enough to make it work. Computer scientists that do work on it are trying to create more unique algorithms that can be published, rather than creating applications with existing working algorithms (after all they''re paid to create new things, not simply apply what's available). In Larry Page's lecture at AAAS he mentioned that there are less than a hundred people working on artificial (general) intelligence today. (you can see it on youtube).