The industrialized world has dealt with automation and labor-saving devices for centuries. Industrialization itself is the product of early labor-saving devices — the steam engine and machine tools such as the milling machine. Machines have become more sophisticated since then, but automation has largely supplanted the crudest forms of labor — pure muscle — and posed a limited threat to “creative” workers or intellectuals. That is about to change.
Artificial intelligence (AI) has been recognized as one of the most important developments of the postindustrial era, a source of extraordinary opportunities for business and individuals. In the most rhapsodic versions, AI will enable a brave new world in which individuals will only have to figure out how best to spend their leisure time.
Today, most discussion of AI focuses on potential — the gap between the promise and the reality — and often overlooks the extraordinary progress that has been made and how quickly the future will arrive. The time to start thinking about the implications — good and bad — is now.
Since the dawn of the digital era, computing power has doubled every couple of years, primarily because of improvements in hardware. Only in the last few years have we developed hardware capable of approximating the power of the human brain, which in turn enables software that can mimic its functions. Together the two are driving exponential growth in AI. This combination means that learning — and AI — will accelerate, especially as machines learn how to build and improve themselves.
Machines now best humans in various competitions, from TV game shows like “Jeopardy” to chess and most recently go, a game of strategy that machines were not supposed to be able to master. This is occurring at a time when, most experts agree, AI is still rudimentary.
Yet even in this “prehistoric” age of AI, algorithms (which is really all AI is) already perform 70 percent of all financial transactions, generate most of the content on news platforms and can recognize patterns in speech, handwriting and faces better than most human beings. Former U.S. Treasury Secretary Larry Summers noted that “artificial intelligence is transforming everything from retailing to banking to the provision of medical care.”
The development of driverless cars has dominated many discussions of automation, and there are estimates that self-driving cars alone could cost up to 5 million jobs. But analyst Kevin Drum warns that when AI is smart enough to drive cars it will be smart enough to do many other things as well. By one estimate, within two decades nearly 50 percent of current jobs could be done by AI. As much as 40 percent of the world’s 500 biggest companies are threatened by AI. By 2060, Drum adds, AI will be capable of performing any task currently done by humans — from manual labor to creative tasks like painting, making music or even managing other human beings (or machines). AI will not do all those jobs, but societies need to prepare for the transitions and the problems that such capabilities will create.
Again, much of the discussion of AI has been misplaced, focusing on the technical, or the moral and ethical implications of taking humans out of the decision-making loop — a seeming necessity in many cases given the speed at which these processors work. There are more practical and physical issues that demand attention, however. For example, there are employment consequences. The prospect of massive job losses is already a poisonous force in national politics around the world, triggering populist movements from Amsterdam to Vienna. Policymakers must anticipate and be ready to deal with this problem.
Another concern is the extraordinary power that algorithms provide to their owners. Individuals and companies that hold the rights to such software will potentially control vast swaths of the economy. Notions of social responsibility have not kept pace with the economic power that is being amassed by these algorithm creators and owners. Or to put it another way, Jack Ma (Alibaba), Jeff Bezos (Amazon) and Eric Schmidt (Google) are not just ordinary businessmen with a good product.
Related to this problem is the capacity for control and manipulation that is conferred on governments (or individuals or companies) that use AI. The ever-expanding networks of sensors and monitors deployed around the world, when used in conjunction with AI, afford those entities unparalleled capacity to surveil and control or motivate citizens. In May, police in Wales used cameras and facial recognition software to identify and arrest a man for the first time. China is using AI for facial and voice recognition, preliminary steps in its plan to build a nationwide, intelligent digital surveillance network capable of identifying and locating individuals, and accessing all their records in real time.
These are the real challenges of AI, not the number of kilometers a self-driving car can go without an accident. They demand urgent attention. They are not getting it.