Terminology: What is an AI agent?
The field of AI has a complex history, with many cycles of hype and bust. But one thing is constant: scientists never agreed on what AI actually is. There is an adage in the field of AI called the AI Effect. As MIT robotics professor Rodney Brookes put it: “Every time we figure out a piece of it, it stops being magical; we say, 'Oh, that's just a computation.”
When we think of AI, we often think of robots, such as the killer robots of the Terminator movies or HAL from 2001 Space Odyssey, or the cute robots in the kids’ animated movies Wall-E and Big Hero. We call these types of AI systems AI agents, because they combine three essential capacities. They perceive their environment using sensors. They think in order to interpret what they perceive, and to predict how the world will change. And they act using effectors. Another common property of an AI agent is that it learns, either through its own experience, or from human teachers.
This broad definition of AI captures a surprisingly large category of things we interact with today. Consider the ranking algorithm that decides how to best rank the posts on your favorite social media app. This algorithm perceives your actions (which posts you pause to look at, which ones you click on, which ones you like or share). It thinks by calculating your preferences and predicting which future posts you might find interesting. And it acts by displaying new posts in a particular order. It also learns, by observing you and millions of other users, to figure out which features of the content, or your own behavior, help predict which posts you will like next.
An embodied AI is an AI that, well, has a body. A Robocar (also known as a self-driving car, or autonomous vehicle) has a physical body. It can perceive its surroundings using a combination of frontal cameras and a LIDAR (Light Detection and Ranging) typically mounted on its roof. The car thinks by interpreting what it sees (e.g. is that a dog or a plastic bag?), predicting future events (e.g. where will a particular pedestrian be 0.67 seconds from now?), and planning future actions (should I brake or cross the yellow light?). The robocar can also act, by spinning its wheels forwards and backwards, steering them right and left, changing gears, blinking, tooting the horn, and so on. Finally, the car learns from its own experiences as well as the experiences of millions of other cars on the road.
References
Wooldridge, M. A Brief History of Artificial Intelligence: What It Is, Where We Are, and Where We Are Going. (Flatiron Books, 2021).
Kahn, J. It’s alive. Wired 10, 72–77 (2002).
Russell, S. J., Norvig, P., Canny, J. F., Malik, J. M. & Edwards, D. D. Artificial intelligence: a modern approach. vol. 2 (Prentice hall Upper Saddle River, 2003).
Domingos, P. The master algorithm: How the quest for the ultimate learning machine will remake our world. (Basic Books, 2015).