Recent AI advances, largely in the area of deep learning, are indeed impressive. However, as impressive as they are, they are all still examples of very narrow AI. A deep net is essentially a giant nonlinear equation that takes input and produces output. The deep net knows nothing about how it does this--it lacks the kind of metacognition that human's have. Metacognition is the ability to reason about your reasoning. Since deep nets lack this (at present) they can be very brittle. For instance, an end-to-end deep net for translating Spanish to English takes Spanish sentences as input and produces English sentences as output. The net itself knows nothing about the meaning of the sentences, or whether a particular Spanish word might be hard to translate. It may be able to translate Hola to Hello, but it doesn't know that these are greetings. The ability to understand all that is the goal of general AI: AI that acts more like general human intelligence. Some are concerned about the future implications of AI how AI might be used to have negative effects on society. While this could (and certainly will) happen, at this point with the narrow AI that we have, I believe that the bigger problem will be the negative effects of positively used narrow AI on the economy and society. For instance, the emergence of self-driving vehicles could put a large percentage of drivers out of business around the world. There is little doubt that self-driving vehicles, once they are better than human drivers, are a good thing; however, they are also disruptive. So as with many hard problems, AI will create a conflict between two goods: maintaining employment and decreasing transportation costs and accidents.