Rebooting AI: Building Artificial Intelligence We Can Trust

and prototypes, but we can also reason about concepts relative to the theories they are embedded in,

In the technical terminology of linguistics, language tends to be underspecified,

you will have to do something no current machine can do reliably: reason and act on something new and unexpected, based not on some immense database of prior experience, but on a powerful and flexible understanding of the world.

We need to invent a new kind of learning that leverages existing knowledge,

The third contributor to the AI Chasm is what we call the robustness gap. Time and again we have seen that once people in AI find a solution that works some of the time, they assume that with a little more work (and a little more data) it will work all of the time. And that’s just not necessarily so.

Sixth, programs that rely on data that the public can manipulate are often susceptible to being gamed.

And the changing nature of the world makes situational awareness even harder; you don’t want to look at the world as a snapshot, you want to see it as an unfolding movie, to distinguish objects that are toppling over from objects that are stable, and to distinguish cars pulling into a parking spot from cars pulling out of a parking spot.

of formal logic,

If you suddenly want to italicize all foreign words, you could just ask, instead of going through the whole document yourself, word by word.