Blocking AI’s Information Explosion Hurts Everyone

Nearly two million Americans have been looking for a job for at least 27 weeks; they’re the so-called long-term unemployed. This subset of the total 7.5 million unemployed has had a particularly hard time finding their next job. Yet, there are around seven million job openings. Why does that mismatch persist between workers looking for jobs and jobs being open?

At least part of this disconnect can be explained by lack of information. On the one hand, firms cannot find workers with the right skills. On the other hand, workers struggle to locate jobs with the right pay and schedule.

Many older Americans experience financial hardship and require additional support to cover basic expenses and health care costs. Yet, around 16 million low-income older adults eligible for support through the U.S. Supplemental Nutrition Assistance Program (SNAP) did not claim it in 2022. For many such seniors, this means missing out on about $300 per month in SNAP support. According to the AARP, a lack of awareness among eligible adults may be to blame. Presumably, with better information, the U.S. Department of Agriculture — the agency that administers SNAP — could more easily and efficiently identify eligible participants and provide them with a streamlined application process.

One final example. Children in foster care often wait years to find a permanent home. This delay is unsurprising for several reasons related to a shortage of information or of time to process it. Caseworkers often manage 30 or so families, which presses them to move at a rapid pace. Families may not know in advance which child may be the right fit for their home, neighborhood, and community. Children, too, may be unsure of the environment in which they will thrive. In short, placements often fall through because the system does not easily provide the right actors with the right information, nor sufficient time to act on the information they do receive.

Problems caused by information friction, such as prolonged joblessness and delays in foster placement, may soon be solved or at least become less common. Technological tools being developed by companies across the United States and around the world are adept at aggregating vast troves of data and identifying previously unobserved patterns. What’s more, these tools are fast, easy to use, and affordable.

Governments, nonprofits, and private stakeholders are finding novel and significant uses of such tools. Washington, D.C., for example, created a jobs portal using this technology, easing the process of pairing workers with employers. Within just two weeks, the portal facilitated 84,000 job matches. Related efforts are underway in Belgium and Greece. The National Institutes of Health has deployed similar tools to match potential volunteers to clinical research trials. A study by the National Fair Housing Alliance found that this technology may reduce mortgage underwriting disparities driven by human bias against certain underserved groups.

More generally, consumers increasingly rely on this tool to analyze terms of service, contracts, and privacy policies that may contradict their preferences and interests. "The information advantages that sellers, service providers and intermediaries enjoy over consumers" may soon come to an end as this technology "goes mainstream," according to The Economist.

These technologies are powered by artificial intelligence (AI), as you may have guessed. Their algorithms can be trained to improve some of the most enduring information asymmetries, as demonstrated by the examples above and myriad other use cases. Yet this is the very technology targeted by state legislatures across the country. Bills broadly prohibiting “algorithmic discrimination” threaten to undermine some of AI’s most valuable use cases.

House Bill 2157 in Washington State, for instance, mandates that AI developers use “reasonable care to protect consumers from any known or reasonably foreseeable risks of algorithmic discrimination.” That said, the bill does include a carve out for the use of AI to expand “an applicant, customer, or participant pool to increase diversity or redress historical discrimination.” But developers will not risk designing and deploying AI tools that achieve such positive outcomes if they are subject to vague “reasonable care” requirements. HB 2157 broadly defines algorithmic discrimination as “use of an AI system that results in an unlawful differential treatment or impact that disfavors an individual or group of individuals" based on a long list of characteristics. “Reasonable care” is undefined.

This is one of many such AI bills across the country. They contain similarly expansive and even contradictory definitions of AI. They often target behavior, such as discrimination in housing and hiring decisions, that is already illegal under state and federal law. And, they impose unclear standards on AI developers.

From the vantage point of an AI developer, such laws will inevitably have a chilling effect. If HB 2157 and related bills become law, startups poised to help address some of the aforementioned information asymmetries may instead pursue “safer” routes — building tools that introduce a lower risk of litigation. Investors, too, may avoid supporting innovators with grand ambitions to improve labor matching between firms and workers or help individuals find scholarships, benefits, and other programs.

We know what the world looks like when decision-makers lack access to such AI tools. Unemployment needlessly persists. Seniors miss out on the benefits they need to achieve financial security. And foster children wait longer to find the right home. Legislators who single out AI as a technology that requires “special” treatment are effectively perpetuating this status quo.

The proper path forward is to treat AI as we do any other tool and focus legislative attention and regulatory resources on enforcing existing laws that prohibit specific outcomes, regardless of the technological means used to achieve those ends. Housing discrimination, employment discrimination, and benefits fraud are already illegal. When they occur, regulators and courts should respond forcefully — whether the decision-maker was a human using a spreadsheet, a word processor, or an AI tool.

Singling out AI for special, technology-specific prohibitions reverses this logic. It shifts attention away from harms and toward methods, discouraging the very innovations that could help reduce persistent and inefficient information gaps.

Kevin Frazier directs the AI Innovation and Law Program at the University of Texas School of Law. He is also a Senior Fellow at the Abundance Institute and an Adjunct Research Fellow at the Cato Institute.