So AI has a number of sub-disciplines. What happened in the 1960s or so is, you think about what I'd like to do is I'd like to build a robot. That's data the robot from Star Trek The Next Generation, that would be the idea. But you can't pose that as a five-year problem so you need to break off little pieces. You need to break a piece. Well, I'm going to try to figure out how maybe we could take an image and figure out what the objects are in the image. If I walk into a room, are there people here? Is there a desk here? Is there a soccer ball here? Those sort of things. I'm going to look at how we can communicate with it. How could I understand language? How could I understand intent and so on? You can break into a series of problems. Each of those has its own sub-problems. For instance, some to understand language. There are all sorts of things. Whenever I use a pronoun, can I figure out which pronoun? Which actual object that pronoun was referring to? Even that's difficult because sometimes I use a pronoun when I discuss but it's not even a noun that I've mentioned before in our conversation, it's a mutually understood noun. There are all sorts of difficulties there. These sub-areas got their own sub-areas and grew into their own areas that had a reason for being or there's no data that was not maybe AI itself. For instance, computer vision is useful even if you're not trying to design AI. If you want to do image search on the web, right? That whole field grew out. You have robotics which even if the robot isn't intelligent as in robots need to do various things; you have robotics, you have computer vision, understanding vision. You have Natural Language Processing which has its own sub-fields. You have Machine Learning which is how do I take historic information and change what I do in the future. You have game-playing things like chess and checkers and go and things like this or more complex maybe realistic games with nations positioning themselves in the Global Theatre. You have theorem proving logic manipulation and you have general planning ideas. All of these were their own subfields and now people who work in those subfields, some of them consider themselves still part of AI and that's why they're in the sub-field. Some of those subfields people working in there don't consider part of AI is its own field and they don't feel connected to artificial intelligence because they have a different reason for wanting to work in that field. Machine learning is one of those. In machine learning, there are reason people come to machine learning just from statistics or from something else and don't come to it because they're interested in questions of artificial intelligence or because they're interested in different questions about data. In fact, AI has had subfields like compilers. How do you write programs and things like that? Or other subfields. Even some of the ones I mentioned before like theorem proving or game-playing, that are no longer considered part of the AI because they are solved. Or at least it's viewed not as an intelligent thing but the thing that naturally a computer would do. One definition of the field of artificial intelligence is, it's the study of all those things that you thought only a human could do when you didn't think a computer could. Now, that's a self-defeating definition because now as soon as I get the computer to do something everyone accepts, "Oh, yeah. Computers could do that." Then it's no longer part of artificial intelligence anymore. Often that's been a practical or a definition of artificial intelligence. What's the state of the art in artificial intelligence? In robotics, self-driving cars and obviously, it's not completely commercially viable yet but I think most people are familiar with how that's progressing; there are examples, they don't work perfectly, they require some data, they require a lot of other things but we have that. In images and image understanding; if you look at Google's image search, that's probably a reasonable commercial product. There things that work better in subdomain certainly much better than certain subdomains like I can search satellite imagery for particular types of weapons or things like that because there's particular financial interest from defense agencies in doing that. In speech recognition, Android's assistant or Siri or things like this are about the general level, if you're talking about phrases in only slightly constrained context. If you talk about phrases in a very constrained contexts like ordering of plane flight or something like that where the conversation is expected to follow a particular script, you can do better. In planning, NASA has planning that is triggering a sequence of actions that goes on. For instance, Remote Agent who's on a space probe that went out to Jupiter. So with the lag, that probe has to be able to think for itself. Because it's not possible to, there's light. Light takes some time; minutes and minutes, sometimes hours to get there and back. Interestingly, the Department of Defense funded DART, which was a planning for logistics. It planned all the logistics of how to get things there for the First Gulf War and they estimate that the amount of savings they had by having that done in a few weeks as opposed to a bunch of logistic planners do it over months as basically paid for all the investment that the Department of Defense has put into artificial intelligence and planning effort, which is quite a large investment. Information Answering; IBM Watson, or even you're serious system or something, that is about the current level. Language; you can go to Google's Machine Translation. Certainly, in a general broad sense that's about the level. Then in game playing, your Alpha Go and Deep Blue are certainly there. Well, there's one slightly better than Alpha Go now. But we have things that basically played Go and chess at the level of human beings if not quite to do that above.