[randcraw]

Personally, I think higher order cognition in AI will be hot. Deep learning has monetized the introduction of AI into numerous mainstream domains (e.g. smart NLP search, vision, speech, game RL), which will motivate and underwrite efforts to push AI beyond the shallow hacks of the past, finally breaking through AI's brittleness problem.

Some problem domains are killer apps, like self driving cars, and on smartphones, personal digital assistants and verbal interfaces. There's no stopping these initiatives. They only question is how far each can go without moon shot levels of investment. But between the economic interests of especially Google and Apple to advance their mobile devices, and the military to make weapons and intel as smart as possible, I'm convinced there's enough critical mass for AI's pile to stay hot for a couple of decades or more.

The trick is to avoid the roadblocks that today's academic agenda inflicts on researchers by demanding they publish frequent shallow incremental novelties. What's needed is 5-10 years to develop the infrastructure that's needed to enable the fielding of robust general reasoning w/ causation and rich knowledgebases.