Here's an idea: A team of coders in a basement combine lots of AI discoveries done by others (Think merging orthogonal approaches like Numenta + Google Brain + Vicarious + etc). With it, they crack the neo-cortex algorithm (or a rough approximation). It learns outdoors (cameras, actuators), indoors (talking), on the internet, in libraries, on wikipedia, etc. It becomes actually useful. Not: "OK, google set an alarm.", but: "Hey Peter, I'm really stuck with this problem, can you find a solution on SO. Oh, and send flower to my Aunt, you know ... the one who likes pink".
If it was freely available I could see a massive migration of users abandoning google. Who needs google in such a context?
-- far fetched, yes. But it is a scenario where the breakthrough is algorithmic rather than Huge-data based. The learning can then happens slowly (2 years undercover from that basement?), and emerge as a powerful intelligence and company.
Yeah, maybe, but the coders able to "combine lots of AI discoveries done by others", as well as those who make those discoveries, are likely to work for Google/FB/MS/IBM/Apple/Amazon/etc already (or want to work there).
A group like OpenAI also has a good chance to make a breakthrough, but you can hardly call them "a team of coders in a basement".
It's unlikely that any a single company (even a large one) will have any significant advantage, because the research everyone does is highly public (everyone tries to publish asap). The research right now is at the stage where people are still looking which way to go, so as soon as someone stumbles upon a promising direction, everyone will jump on it (e.g. AlexNet success in 2012).
The same could have been said of Brin and Page. Why were they not working for yahoo improving their search algorithm? Some people will have the hunger and drive to disrupt the incumbent.
If it was freely available I could see a massive migration of users abandoning google. Who needs google in such a context?
-- far fetched, yes. But it is a scenario where the breakthrough is algorithmic rather than Huge-data based. The learning can then happens slowly (2 years undercover from that basement?), and emerge as a powerful intelligence and company.