Generative AI for more than a year has raised almost as many worries about the risks and uncertainties that come with the technology as it has about the benefits.
However, while most of those concerns have been focused on security, privacy, disinformation, and ethics, economics should be added to the list, according to an official with the International Monetary Fund (IMF).
Generative AI promises significant boosts to the global economy through improved productivity and similar gains, Gita Gopinath, first deputy managing director with IMF, said during a talk at the recent AI for Good Global Summit in Geneva, Switzerland. That said, it could also exacerbate the next worldwide economic downturn that is bound to hit sometime, Gopinath said.
“The widespread use of AI could turn an ordinary downturn into a deep and prolonged economic crisis by causing large-scale disruptions in labor markets, in financial markets, and in supply chains,” she said.
Raising the Alarm
The IMF for months has been beating the drum about the economic changes that generative AI will bring to society and the need to begin preparing now given the breakneck pace of innovation around the still-emerging technology. In January, the IMF released a report that said AI will complement some jobs but eliminate others, and the challenge for countries is to ensure that the use of AI benefits humanity.
In advanced economies, about 60 percent of jobs will be affected by AI, with half benefiting and thus improving productivity, and the other half possibly replaced by AI, lowering labor demand and leading to reduced wages and hiring, IMF Managing Director Kristalina Georgieva wrote in an accompany blog post. Emerging markets and low-income countries will experience fewer immediate disruptions, but also are unlikely to derive the same benefits as their wealthier counterparts, Georgieva wrote.
In May, she said generative AI is slamming into the global labor market “like a tsunami,” adding that “we have very little time to get people ready for it.”
Economic Benefits and Risks
In her talk in Switzerland, Gopinath said that the productivity and other economic gains from generative AI could help give the global economy a lift at a time when growth outlooks are weak. It also could accelerate problems in times of sharp downturns.
When looking at the labor market, the rising use of automation tools can offer a lesson when a global economic crisis hits. When times are good and companies are pulling in profits, they tend to invest in both automation and workers, even if their value to the bottom line declines. But when a downturn hits, they cut costs by cutting workers, and often those jobs don’t come back.
Gopinath noted research that found that since the mid-1980s, almost 90% of automation-related losses in the United States happened in the first years of recessions and companies will stick with the automation even after the crisis lifts.
AI likely will make this worse. About 30% of jobs in advanced economies may be replaced by AI, with figures reach 20% in emerging markets and 18% for low-income countries.
“In other words, the pool of potentially replaceable workers in future downturns will be bigger than anything we’ve seen before,” she said. “The result could be unprecedented job losses. That could also lead to unprecedented numbers of long-term unemployed, because many of the displaced workers will lack the requisite skills in an economy where AI is increasingly prevalent.”
It would cause a “major shock to the financial system,” with record numbers of unemployed workers unable repay their debts. However, that would be only one of the hits to the global financial system.
Unfamiliar Patterns
The financial services industry is quickly shifting to AI and automation, bringing in complex models with more complex ones that not only learn on their own but also are more difficult to understand. AI models also are increasing in client-facing investment businesses. In normal financial times, this works, given AI’s ability to crunch through massive amounts of information to make predictions to improve resource allocation and inclusion in the financial markets.
But downturns bring unfamiliar patterns – like job losses – which means AI systems could have difficulty responding, Gopinath said. AI tends to perform badly when events differ greatly from the data the models were trained on.
“As a result, they might quickly and simultaneously become overly conservative and rebalance portfolios toward safe assets,” she said. “The models’ decision to leave other assets will then be rewarded as their prices fall, and a self-confirming spiral of fire-sales and collapsing asset prices across different financial markets could ensue. The ‘black box’ nature of AI would make managing such an event particularly challenging.”
Supply Chains Could Take a Hit
A similar fate could await global supply chains. Companies adopting newer AI models will see benefits during normal times, “but in a future downturn, AI algorithms trained on stale information could trigger a series of forecasting errors, creating more rapid swings in production and inventories,” she said. “This could cause crippling delays and shortages of critical supplies across the global economy.”
“Taken together, these forms of risks could turn a regular downturn into an AI-amplified economic and financial crisis and present immense challenges for policymakers,” she said. “And if this AI crisis were to strike in the next few years, it would be hitting countries at a time when they are already dealing with high debt and low growth, seriously constraining their ability to support workers and firms.”
Softening the Blow
Some policies countries should consider adopting are ensuring that tax breaks don’t benefit hardware and software that drive automation over those that complement workers and that heavy investments are made in education and training to help workers adjust to the influence of AI.
“But such upskilling, though necessary, will not be sufficient to prepare workers for AI’s potentially destabilizing effects,” Gopinath said. “We will also need to protect them by adapting social safety nets to a world where AI could create prolonged job losses. Forthcoming research by the IMF shows that more generous provision of unemployment insurance can help workers adapt to job market changes.”
Also, regulators will need to better supervise and regulate AI, financial institutions will have to give greater visibility in how they use AI and the source of their models, and companies inside and outside of the financial services sector will have to stress test their AI models more than they have other systems in the past.
“To be clear: the approach I am advocating would represent a significant shift,” she said. “Most international efforts to mitigate AI risk are currently directed toward concerns around security, privacy, ethics, and disinformation – and those efforts are important. But we also need a serious international effort to AI-proof the economy.”
