Terminology: Narrow vs general AI
In the media and public discourse, it is common to use the term AI to refer both to general AI agents (which can do the whole combination of perceiving, thinking, acting, and maybe learning) and narrow software tools that have only a subset of these faculties.
An example of a narrow AI system is the ranking algorithm that decides in which order to show you the social media posts of your friends or people you follow. Another example is the algorithm that calculates a route for you to take in order to reach your destination, whether by foot, by car, or public transport. Yet another example is the computer vision algorithm that detects and tags your friends in photos, and can help you find other photos of the same people.
Consider an image captioning AI which can tell you which animal, bird or plant is in a given picture. This AI algorithm performs perception and reasoning, but not action. It perceives the image, and analyzes its content to make a guess about its contents. But it does not make any substantive action, beyond just spitting out text labels. It is therefore considered a narrow AI.
Sometimes, we consider an AI narrow because it can perform only one specific task, even though it is technically an AI agent capable of perception, reasoning and action. An example is an AI that plays Chess: it can perceive the game board, it can think by strategizing about its next move, and it can act by moving pieces around the board. Nevertheless, this AI cannot do anything else, and is therefore still considered narrow.
In the remainder of the book, I’m going to use embodied robot cartoons whenever depicting an AI system, regardless of whether they are narrow or general. Just remember, in reality these robots are not always agents. The cartoons are just abstractions to simplify our discussion and visual presentation.
A special kind of general AI agent is the so-called Artificial General Intelligence (AGI), or strong AI. This is an AI system, not yet in existence, that can perform any intellectual task as good as a human.
References
Goertzel, B. Artificial General Intelligence: Concept, State of the Art, and Future Prospects. Journal of Artificial General Intelligence; Vienna 5, 1–48 (2014).