Artificial intelligence could supercharge Trump’s deregulatory push, but experts flag shortfalls

As the Trump administration has reportedly used artificial intelligence to inform governmental decisionmaking, former public officials and academics have spotlighted opportunities — and pitfalls — to incorporating the technology into the regulatory process. 

The president is pursuing a deregulatory agenda, directing agencies to identify rules for elimination and issuing orders, which have been challenged in court, to streamline the revocation. Officials generally must write a new regulation in order to overturn an existing one, and the entire process typically takes a year. 

Bridget Dooling, a law professor and former career official at the Office of Information and Regulatory Affairs, said during remarks at a Tuesday event sponsored by George Washington University’s Regulatory Studies Center and the regulatory-focused company Norm Ai that it makes sense for agencies to utilize AI to help ensure regulated entities are following current rules. But she warned against relying on AI to speed up the slow process for writing regulations — whether it be for enacting new rules or axing ones selected for repeal. 

“You can wrap yourself up in a rhetoric of efficiency on this, but if you know about the regulatory process, this is making you nervous…because you know that agencies have a legal duty to engage in ‘reasoned decision making’ [and] to not make ‘arbitrary’ and ‘capricious’ decisions’,” she said. “When it comes to reasoning, we can't outsource it.”  

She went on to say that requiring AI to draft regulations while simply keeping “humans in the loop” would degrade government analysis. 

“There are going to be some tasks where a ‘human in the loop’ is really, really weak sauce for the nature of what's going on underneath,” she said. “I think it's a mistake because it subordinates the beautiful complexity of human judgment and thought, it violates the duties the law confers upon agencies to do their own thinking and, frankly, we just deserve better.”

Dan Berkowitz, a former Democratic commissioner of the Commodity Futures Trading Commission during the first Trump term, said during a panel at Tuesday’s event that including AI in agency reasoning could prompt legal scrutiny. 

“To what extent would it be the agency's decision or to what extent are you relying on some external input?” Berkowitz said, while noting that he himself uses AI for legal research. “I'm sure that if somebody didn't like the rulemaking, they might bring a challenge and say ‘Well, what was the input? Who had input into the model that you're using? Was that an industry-funded AI or what?’ So that would raise some additional questions about the integrity of the decision process….” 

Public policy nonprofit Democracy Forward on June 27 filed a lawsuit to obtain records related to the Trump administration’s use of AI, including to potentially rewrite or repeal regulations. 

“Emerging technologies like artificial intelligence are being used to make sweeping decisions across government, including targeting workers and rewriting our laws — but the American people are being kept in the dark,” said Skye Perryman, president and CEO of Democracy Forward, in a statement.

ProPublica reported in June that a Department of Government Efficiency engineer created an AI tool to identify Veterans Affairs Department contracts to cancel that “produced results with glaring mistakes,” including erroneously inflating the cost of some from $35,000 to $34 million. 

Sen. Jon Husted, R-Ohio, in March introduced legislation (S. 1110), which is based on an initiative he led in his home state as then-lieutenant governor, that would set up an annual process for an AI system to identify “redundant or outdated” regulations as well as an expedited timeline for officials to decide whether to repeal such rules. 

“As the federal government strives to serve citizens better and at a lower cost to taxpayers, Congress can help by taking a lesson from Ohio’s work using an AI tool to cut useless and burdensome pieces out of our state code. We estimate that this tool helping experts streamline Ohio’s code saves $44 million and 58,000 manhours over just a decade,” Husted said in a statement. “I wrote this bill to give government a tool that helps them reduce waste and save time — and to give job creators and taxpayers a look at just how much Washington could do to get out of their way and siphon less money from their pockets.” 

Cary Coglianese — the director of University of Pennsylvania’s Program on Regulation, who at Tuesday’s event contended that the government is underutilizing AI — emphasized that some large language models are “no more informed than asking one’s own teenager a question.” 

“These tools are not, generally speaking, good at making causal judgments. So evaluating whether a regulation is or is not working is not something that, generally speaking, AI tools are going to answer for us,” he said. “[Officials cannot] just simply assume that ‘Oh, I've asked it a question. It's given me an answer. It seems confident in the answer, and so therefore I should rely on it.’ That's deeply problematic.”

Experts who spoke at Tuesday’s event also stressed that agencies need workers with the necessary skills to take full advantage of AI, but many public tech employees have lost their jobs or otherwise left government due to Trump’s reductions in the size of the civil service. 

“Unfortunately, we're at a current period where there's more of an exodus, to put it mildly, than an inflow into government,” Berkowitz said. “Government needs individuals who are conversant in AI — the younger generation, people who are now growing up with AI [and] senior people who are open to it and know how to use it — we need to encourage those people to come into government and bring these new tools in.”