FROM MR. ED TO MR. ROBOT:

In 1915, there were over 22 million horses in the United States; by 1960, about three million. Is it possible that the work available to a great many human beings is ultimately destined to follow the same path? On the surface, it may seem absurd to compare people to horses. Horses, after all, can provide transportation or help out on the farm, but they have very little ability to adapt. When cars, trucks, and tractors came along, horses had nowhere to turn.

People, of course, are intelligent. People can adapt to new roles. From an economic standpoint, perhaps the single most important difference between horses and humans is that people can learn to do new things.

Should we take comfort from that fact? Does it guarantee that we'll always have enough remunerative work to employ the vast majority of our adult population? The reflexive answer might be yes. But there is another point to consider. Unlike the cars, trucks, and tractors that displaced horses, today's machines and algorithms can learn. In other words, modern information technology is not just encroaching on new types of work; it is gradually taking on the single most important capability that has allowed workers to stay ahead of the machines.

This new ability for machines and algorithms to adapt and learn is perhaps best illustrated by recent advances in an area of artificial intelligence known as deep learning. Computer scientists have used artificial neural networks, which operate according to the same essential principles as the biological neurons in the brain, to perform basic pattern recognition tasks for decades. Deep learning takes advantage of recent technical breakthroughs that allow neural networks of unprecedented complexity to be employed in areas like image recognition and language translation. Some deep learning systems can now perform better than humans at recognizing images such as road signs. Chinese researchers recently unveiled an algorithm that can outperform humans at the type of verbal reasoning problems found on IQ tests, while a team at the University of California-Berkeley has used deep-learning techniques to build robots capable of figuring out how to complete tasks that require a high degree of dexterity, like unscrewing the top from a bottle.

Deep learning, as well as a number of other approaches to machine learning, is already being deployed in areas from self-driving cars to algorithms that write news stories. And there is every reason to expect both the capability of these smart algorithms and the number of uses to which they are put to accelerate.

Follow @brothersjudd