Many examples of AI — either already created, as works in progress, or as envisioned in the future — are commonly cited in books and articles. Examples include:

• Self-driving cars that understand the rules of the road, can perceive objects in their environments (other vehicles, pedestrians, road signs, etc.), and can operate the car to safely and efficiently deliver their passengers to their intended destinations
• Robots that can assist with tasks around the home, which can understand spoken requests, navigate through their surroundings, and perform mechanical actions to achieved desired tasks
• Robots that can successfully stabilize, clear, and/or decontaminate situations or environments that are hazardous to humans
• Computer programs that play games, such as chess or Go, at a level better than the best humans
• Phone apps that tell you what kind of plant you took a picture of
• Phone apps that tell you what kind of bird you are listening to
• Chatbots that can write compelling text given a prompt to start with
• Applications that ingest some text and provide a short and useful summary
• Applications that write computer code based on a textual description of what that code needs to accomplish
• Applications that translate written or spoken text in one language to any other language
• Computer programs that can quickly generate many candidate small molecule structures that might work effectively as drugs or medicines

All of the examples listed above describe things that humans could do, and that seem to require some level of knowledge or expertise for humans to accomplish. In some cases, the interest may lie in the fact that machines can do something better or faster than humans; in some cases, the appeal seems to involve having machines do something that humans find boring, distracting, or dangerous.

Are we just interested in recreating anything that a human can do, but maybe faster, better, or otherwise just freeing up our time for other things? Or are there things that we would like machines to do that humans cannot do, and if so, would we consider them intelligent?

It should be noted that what constitutes "AI" is a moving bar, and will continue to move, perhaps even more quickly now that advances are coming at a rapid pace. One of the things a self-driving car needs to be able to do is plan a route, factoring in criteria such as fastest vs shortest vs least expensive, and taking into account current traffic information. In the past, such as task might have been construed as a hallmark of intelligence, but GPS devices and apps have been around for a long time to provide such functionality, and now seem rather mundane. (Some might argue that planning a route was never about intelligence, but just a problem in graph traversal, through the use of something like Dijkstra's algorithm.) The well-known computer scientist Larry Tesler is often misquoted as having said: “Artificial Intelligence is whatever hasn't been done yet”, but what he actually said was: “Intelligence is whatever machines haven't done yet”. There is a lot of debate as to what actually constitutes intelligence, and as we perhaps move from Narrow AI to General AI (more to follow), such debates will get more firmly grounded in machines that are being built, rather than in speculations about the future.

It should also be noted that all of the examples listed above describe functionality — that is, things that can be accomplished — rather than mechanisms or details by which that functionality is achieved. A recurring theme that we will revisit in some of the following pages involves this separation of outward-facing functionality from inward-facing implementation.