“Wildly irresponsible”: DOT’s use of AI to draft safety rules sparks concerns

The U.S. Department of Transportation (DOT) is facing scrutiny for its decision to employ artificial intelligence in drafting critical safety regulations for aviation, automotive, and pipeline sectors, according to a recent investigation by ProPublica. This innovative approach raises alarms among staff and experts, as the DOT could become the first federal agency to rely on AI for such an important task. The potential risks are significant, given AI's propensity for inaccuracies and its tendency to generate misleading information, often referred to as 'hallucinations.' Staff members worry that failing to identify AI-generated mistakes could lead to flawed regulations, resulting in legal disputes, injuries, or worse, fatalities within the transportation system. Despite these concerns, Gregory Zerzan, the DOT's chief legal advisor, appears unfazed. Meeting notes from December revealed that he believes the primary goal is not to achieve perfection in AI-generated rules. Instead, the focus is on expediting the rule-making process, allowing regulations that typically take weeks or even months to draft to be completed in as little as 30 days. Zerzan noted that their AI tool, Google Gemini, can produce draft rules in under half an hour. At the meeting, he stated, "We don’t need the perfect rule on XYZ. We don’t even need a very good rule on XYZ. We want good enough." ProPublica's findings include insights from experts and anonymous comments from six DOT employees who expressed their unease about the reliance on Google Gemini for rule drafting. Some specialists monitoring AI's implementation in government have suggested that while the tool could be beneficial as a research assistant, it should be used with extensive supervision and transparency. During a presentation, DOT staff were informed that much of the content in regulatory preambles is often just 'word salad,' implying that Gemini could effectively produce such content without substantial oversight.