There is a lot of panic about the implications of artificial intelligence technologies for the future of work. Some scholars, such as Nobel laureate Joseph Stiglitz have suggested that AI will exacerbate inequality, and the infamous 47% of jobs being automated statistic still resonates. President Trump has endorsed this techno-panic, especially with his opposition to Amazon’s innovativeness, and his appeals to the rust belt, whose workers have been displaced by technology and globalization. If he is serious about improving access to jobs, however, he will move to embrace new technologies, rather than delay them.
There’s an old paradox in economics: when technology improves the efficiency of a resource, more of that resource will be consumed, vastly increasing output. While insights such as this are central to environmental economics (better energy production increases the amount of energy demanded), it also gives important insights into the labor market. When a new technology makes labor more productive, a firm does not need to hire fewer workers to achieve the same output, but rather it would want to hire a lot more workers to increase its revenue.
To make this example concrete, the impact of the ATM on bank tellers is an illustrative case. Before the ATM was introduced, a teller’s main task was telling: the act of counting cash and updating bank balances. The ATM did this task in its entirety, so one would assume there would be no more need for bank tellers, right? Well, walk into any bank and you’ll notice that they still have tellers. In fact, there are far more tellers today than there were before the introduction of the ATM.
Cheap ATMs automated the time-consuming task of accounting, reducing the space needed in banks, and saving them money. They could also be put up anywhere. As a result, banks needed less space, and could open up more locations. In these locations they needed a lot more tellers. While they originally spent only a fraction of their time dealing with customer issues and improving the quality of customer satisfaction, that was now their main role. The demand for the job increased, though the need for numeracy was deemphasized, while customer service gained importance.
This is how new technologies have always affected the labor market: boosting productivity and increasing jobs. The issue lies not with technology destroying jobs, but the failure of new jobs to signal their demand. Since 1980 the most in-demand occupations were those requiring social skills, since computers are incapable of them—but the emphasis on improving social skills for young people (or retraining programs that make this their focus) are minimal. This is due to a variety of legal and regulatory barriers that make transitioning to service-based careers artificially difficult.
Take care work, one of the most in-demand occupations as a result of an aging population. It is predicted that within 4 years there will be 1.2 million unfilled vacancies for registered nurses, yet there is still a clamor over a lack of jobs. But nursing is a difficult job that not everyone is suited for, right? As with any occupation, it requires training, and job seekers would be willing to invest in that training if it were worth the costs. However, the median salary for nursing, given the lengthy certification process, is likely discouraging people from seeing it as an alternative career path unless they are already passionate about it. Loosening the requirements for care work and allowing a more open labor market would go a long way to increasing the number of people going into the profession. Empathy, emotional skills, and communication are all areas that AI is unable to touch, and will increase the demand for significantly. As the prices of all other items fall due to the productivity gains new technologies bring, there will be more income to be spent on those areas that AI is unable to improve.
Discussions on artificial intelligence often neglect the limits of current AI technology which are likely to remain for the foreseeable future. Current AI is incapable of thinking and performing general tasks, but excels at performing well-defined narrow tasks. This means that jobs as a whole are not at threat, since jobs are complex combinations of multiple tasks, but only that certain skills will not be needed as much. It is not just laymen that fall into this trap, but experts as well. Geoffrey Hinton, “godfather” of deep learning, the basis of all recent AI advancements, said that a radiologist will be obsolete since an AI can detect tumors with greater accuracy. This ignores all the other aspects of the radiologist’s job, from the fiduciary responsibility to the patient, the judgement required in seeing whether a certain risk is worth taking to operate, and effectively communicating findings. Radiologists will not become obsolete, but they will need to focus more on other aspects of their jobs than they currently do, and need to learn to work with AI.
AI then does not automate, so much as it augments. Man plus machine is much more productive than man or machine for as long as we can predict. We should not be worried about AI taking jobs from people, but rather why people are doing jobs that are meant to be for machines. Taking advantage of that productivity gain, and allowing it to free up resources is a must if prosperity is desirable. The focus should not be on how to prevent automation, nor what kind of redistribution is needed when everyone is technologically unemployed, but rather how to best harness technological capability in order to improve the lives of many through more dynamic labor markets.
The Donald Trump administration should rewrite the Obama AI reports, focusing on the federal government role in disincentivizing employment and ensuring that tomorrow’s technologists and mainline industry and service sectors can expand it. There are real risks to jobs that come from technology, but they arise more from regulations preventing labor market adjustment, rather than the technologies themselves.