January 11, 2018
Over the past few years, artificial intelligence has rapidly matured as a viable field of technology. Machines that learn from experience, adjust to new inputs, and perform tasks once uniquely the domain of humans, have entered our daily lives in ways seen and unseen. Given the current breakneck pace of change and innovation, the question for governments and policymakers is how to harness the benefits of artificial intelligence, and not be trampled by the robot takeover of our nightmares. The answer is simple: make them work for us.
Recently, the IMF’s Managing Director Christine Lagarde convened some of the most distinguished voices in the field of artificial intelligence, including Malcolm Frank of Cognizant; Martin Ford, author of Rise of the Robots: Technology and the Threat of a Jobless Future; Chief Analytics Officer of IBM, Martin Fleming; and Andrew McAfee and Simon Johnson, the latter a former Chief Economist of the IMF, and both professors at MIT.
There are four areas of artificial intelligence and machine learning of importance to the IMF’s work:
- Governance: Both countries and the IMF will need to address the provenance of data, as well as matters of privacy and informed consent before basing analysis or grounding policy advice on Big Data or the algorithms used to generate findings. Big Data is dynamic, heterogeneous, and may originate in sectors that do not map cleanly to the IMF’s existing lines of responsibility or expertise. For example, data generated from e-commerce, the Internet of Things, satellite data, or supply-chain and logistics data are not yet well understood or integrated in how we assess the health of a country’s economy. Both the IMF and countries will need to develop expertise in the use of such micro-level data.
- Labor markets: Labor markets will look different in the next few years. There will be fewer middle-skilled jobs, such as insurance claims processing or jobs performed in a constrained physical space, like fork-lift operator or order expeditor. These sorts of jobs have been more resistant to offshoring or automation so far. But they may disappear soon, as artificial intelligence improves and robots are more able to make decisions based on ambiguous situations. This has implications for education, retirement, and social welfare programs. Large numbers of middle-class jobs may be eliminated, leading to unemployment or underemployment. Some jobs will require extensive retraining to ensure that workers can perform the work. Many countries are already facing rapidly aging populations. Should large numbers of workers leave the labor market prematurely, governments will find it even more difficult to fund social-welfare and retirement benefits.
- Taxes: By implication, should labor markets rapidly shed middle-skilled or low-skilled jobs as many predict, the tax structures of many countries will need to reflect the decreasing share of GDP attributable to wages and salaries. Among the Organization for Economic Cooperation and Development countries, roughly half of government revenue is derived from individual income or social insurance taxes. If labor becomes a smaller part of developed economies, tax structures will need to change to sustain government revenues near current levels, and to avoid creating further disincentives to the creation of jobs. For example, Microsoft founder Bill Gates suggested that a tax might be levied on robots.
- Social equity: Computer-driven decision-making should be open to scrutiny and inspection, and must not simply be automated versions of mental models that embed legacies of social inequality. For instance, some businesses make use of data to offer personalized pricing, based on predictive models about the future revenue stream that a potential customer might provide. Some customers who do not match an optimal profile might be “invited to leave quietly.” Such redlining of particular groups of customers may lead to further marginalization, leading to a self-fulfilling prophesy.
Economists generally build models and then refine them to reduce error and improve robustness. Many artificial intelligence methods are impervious to external analysis, because software based on artificial intelligence learns and adapts as it encounters new data. After millions of iterations, the algorithm itself will have changed substantially. “The algorithm told me to do it,” is unlikely to withstand public inquiry as the basis for policy development.
There is clearly a need for all institutions to keep up as this rapidly changing world impacts their work. So, the IMF will continue to bring in experts to promote information exchange and develop training so that its staff can work with these new technologies as they emerge. In turn, this will help the IMF work with its member countries to make artificial intelligence serve the public good.