Bob Higgins wrote:
That is the problem with the work of Futurists - many of the massive changes in our lives comes from seminal inventions whose timing cannot be predicted... I believe AI is in a similar state of waiting for that seminal invention that makes AI practical.
The timing of advances may be unpredictable, in general and overly optimistic. However in the case of AI there could be a bootstrap, since we have the human model to improve upon - and mature hardware already. Perhaps the only thing missing from a functioning AI, using today's technology without a single new breakthrough, is simply an autonomous high-level control unit which will be allowed to learn and grow on its own. The hardware need not be more complicated than a PC so long as all the web is available as the knowledge base. Language parsing is said to be much improved at places like IBM, and that will filter down sooner or later.
First it is necessary to instill - for lack of a better word - greed. This may be the "year of the Gekko" after all ... thanks to the recent election. AI programmers actually know this and are struggling to pull-off the kind of deep mimicry which allows "rewards" to operate in a competitive world of greedy machine learners. How do you program greed into a control unit? Already there are greedy algorithms and problem-solving heuristics which can be coupled with something akin to a rewards credit card or bitcoins. Sounds silly but no more so, to a realist, than is "heaven" as a reward.
Reward's do work at many levels, especially over time in a competitive society - but the solvable problem is the risks they carry when autonomy is allowed. We must realize that "greed" is the imperative behind learning, in the real world, and secondly, that rewards that reinforce greed require both a distinct identity and a strong semblance of "free will" but with flexible limits more realistic than Asimov's 3 laws. For instance, an AI will soon be allowed to post to blogs and even encouraged to embarrass human ignorance, and this could happen in a few years. Who knows but that we may have one or two warming up on Vortex already.
Anyway, philosophy and behavioral sciences will become required courses for the AI programmer :-)
https://en.wikipedia.org/wiki/Reinforcement_learning
