Rebooting AI: Building Artificial Intelligence We Can Trust

Word2Vec famous, though, was the discovery that it seemed to work for verbal analogies,

hybrid model, a tiny bit of structure at the micro level, in which regular verbs were generalized by

(Often, a single application, like the traffic routing algorithms used by Waze, will involve multiple techniques,

Language is almost never like that. Virtually every sentence that we encounter requires that we make inferences about how a broad range of background knowledge interrelates with what we read. Deep learning lacks a direct way of representing that knowledge, let alone performing

copy forty different recipes from forty different web pages, automatically converting all imperial measurements into metric, and scaling everything for four portions?

further consequence of the opacity of deep learning is that it doesn’t fit naturally with knowledge about how the world works. There is no easy way to tell a deep network “apples grow on trees” or “apples tend to fall down, not up, when they fall out of trees.”

what you do when you read any text is to build up a cognitive model of the meaning of what the text is saying.

The first we call the gullibility gap, which starts with the fact that we humans did not evolve to distinguish between humans and machines—which leaves us easily fooled. We attribute intelligence to computers because we have evolved and lived among human beings who themselves base their actions on abstractions like ideas, beliefs, and desires.

need a way of understanding space and the geometry of people and objects.

that’s not a point of view that comes at all naturally to a deep learning system. Deep learning is focused around categories, not around individuals.