Lab 4: Robots and Artificial Intelligence

The most important advice we can offer to teachers on this topic is to seek out current readings for students, because the technology changes so quickly.

It's interesting to compare the strengths and weaknesses of computers compared to people. Modern electronic circuits are vastly faster, with a vastly greater storage capacity, than the human brain. But the human brain doesn't just do one thing at a time; every neuron is "running" all the time, giving us a vast parallelism that was beyond our ability to simulate with computers until quite recently. Many of the early AI goals, such as playing chess at human expert level, were finally accomplished by brute speed at considering possible moves in parallel, rather than by understanding how human chess masters think.

There have been several recent achievements of similar goals. The world's best chess player has been a computer since the 1990s, but 2015 was the first victory of a computer Go program over a human champion without handicapping. A computer won the TV game "Jeopardy!" in 2011, generating lots of good press for AI. In that case, in addition to speed, vast amounts of data were indexed to give the computer the knowledge of popular culture required in that game.

In early AI work, researchers devised algorithms and programmed them into computers, just as students do in this course. Modern AI work is quite different: The computer is programmed with a "neural network" modeled on the brain, and then "trained" with millions or billions of examples of whatever the computer is asked to process—pedestrians in road pictures for a self-driving car, the same face in different pictures, tumors in x-ray pictures, spoken words. The researchers don't really know how the computer will use its simulated neurons, and so they are surprised by undesirable results: the simulated cars don't recognize black pedestrians; the Twitter bot learns to tweet racist neo-Nazi propaganda.

In the old days, some people were afraid of AI because of movies about killer humanoid robots. Sometimes enthusiasts proposed to use AI in ways that really were threatening, such as the Strategic Defense Initiative ("Star Wars") of the 1980s, which proposed to put satellites in orbit armed with deadly weapons that onboard computers would deploy autonomously. Luckily, the technology of the time fell so far short of being able to meet the needs of the program that early testing failed spectacularly and the program was cancelled.

Today, some people, including some AI researchers, are afraid of AI because governments and businesses are eagerly deploying neural nets to make decisions about whom to hire, who should get a mortgage, and which prisoners to parole. The threat isn't that the computers try to take over; it's that people want them to! At the same time, AI is saving lives in medical research. Like every other kind of technology, we can neither prohibit AI nor blindly endorse it, but must question who benefits and who is hurt, and try to prevent the harms.

Robots, an AI-enabled technology, are already widely used in manufacturing. Those robots don't look like people; each robot is more like a single arm and hand. Other robots, such as the ones used to explore other planets, are more like cars, but without riders. Self-driving cars on Earth are more controversial. Current technology seems to be better than human drivers on highways, but not yet ready for local streets. But, like AI in general, the autonomous vehicle technology changes quickly.

Pacing

Prepare

Lab Pages

BJC Videos from UC Berkeley

Solutions

Correlation with 2020 AP CS Principles Framework