AI and machine learning
Yet another Facebook AI project, this one focused on automated bug fixes. This has been an area of much focus lately, and if this is open source (it's not clear that it is), it could help advance some of the commercial efforts currently getting funded and coming out of larger software companies.
Dropbox is focusing on machine learning to improve its user interface. It sounds like the company has been pretty good so far about tactical applications, but it's unclear to me how much users really care about social graphs, etc, applied to documents. That was a big thing several years ago.
I'm really interested to see if there's a Hadoop (as it's currently defined) resurgence now that AI is picking up. Organizations doing AI still need to store and process lots of data, and build pipelines.
This is clearly the biggest concern to democracy and the most important discussion to have around AI. Oh, wait, it's not. It's kind of sad that lawmakers are so easily (or cynically) attracted to shiny things rather than focusing on more pressing and solvable issues.
This sounds like a stretch to call AI, but taxes do seem like a good place to test this out, because we're talking about money rather than freedom. Just as long as there's human oversight, appeals, etc.
Now that Bitcoin mining has leveled off and every company under the sun is working on embedded AI chips, inferencing seems like Nvidia's biggest challenge.
I haven't seen the full survey, but I'm with the folks in the highlighted responses predicting a slight-to moderate impact of AI in the new few years. Even inside most industries, most applications I see will help with efficiency rather than anything revolutionary.
Potentially interesting work from DeepMind, basically around letting reinforcement learning systems function on new video games that have different scoring mechanisms and scales.
This is one of those common sense capabilities -- being able to predict what will happen next, especially to an object -- that anything resembling AGI will require. The big challenge, of course, is whether a machine can ever understand why something happens.