In the slow (most conservative) scenario, we estimate that $439 million invested in AI in 2021 would yield $49 billion in productivity gains that same year; $634 million invested in 2023 would yield $87 billion in gains in 2023; and $1.6 billion invested in 2028 would deliver $205 billion in productivity gains in 2028.
In the baseline scenario, we estimate that $600 million invested in AI in 2021 would yield $84 billion in productivity gains that same year; $976 million invested in 2023 would yield $143 billion in gains in 2023; and $3.8 billion invested in 2028 would deliver $364 billion in productivity gains in 2028.
In the intense (most optimistic) scenario, we estimate that $697 million invested in AI in 2021 would yield $104 billion in productivity gains that same year; $1.4 billion invested in 2023 would yield $196 billion in gains in 2023; and $7.4 billion invested in 2028 would deliver $532 billion in productivity gains in 2028.
Such a productivity surge would confer enormous benefits. To name only a few: it would generate vast savings, which could be reinvested in innovation and R&D; and it would reduce delivery backlogs, allowing citizens to receive services more promptly.
Consider, again, the benefits for tax collection. In 2017, the Internal Revenue Service fielded just 53% of the calls to its public hotline, with callers waiting 17 minutes for service, on average (That year, an answered call cost the IRS $41). Investing in AI-powered chatbots to answer calls would cut waiting times sharply, as chatbots fielded routine questions and human operators took the most difficult queries.
Whether at the IRS or the U.S. Copyright Office or the Veterans Benefits Administration, more productive workers would deliver better services, improving lives and increasing citizens’ satisfaction with their government.
To put productivity gains of $532 billion (our high-end estimate) in perspective, consider that the amount would equal about 2.4% of America’s GDP of $21.73 trillion in 2019; and it would be 1.7 times greater than the U.S. government’s civilian payroll of $298 billion in fiscal year 2020.
When thinking about our productivity estimates, two caveats are also in order. The first is that productivity gains from AI may reveal themselves in non-conventional ways. “Intangibles such as better responsiveness to customers and increased coordination with suppliers do not always increase the amount or even intrinsic quality of output,” observed MIT’s Erik Brynjolfsson, in an influential 1994 paper, The Productivity Paradox of Information Technology. “But they do help make sure it arrives at the right time, at the right place, with the right attributes for each customer.” What applied to past IT breakthroughs may well apply to future AI breakthroughs.
The second caveat: our productivity projections assume that federal workers and executives are widely empowered to make the most of the new AI tools at their disposal. At present, this is not the case.
1 This model also included other “intelligent technologies” like virtual reality and Internet of Things that were omitted from our U.S. federal government study.