Politico: Errors have appeared in U.S. bills due to the use of AI

The U.S. Congress has faced an influx of legislative documents generated by artificial intelligence (AI), many of which are full of errors and poor-quality wording.

This was reported by Politico, citing sources.

According to their information, some members of Congress "are turning to the Claude and ChatGPT models to draft legislative language." The sources note that, as a result, staff of the House of Representatives' legal office are forced to spend "more time checking and rewriting texts generated with the help of AI than on writing bills from scratch."

The publication's interlocutors note that errors in the generated texts could lead to a slowdown of legislative processes and result in lawsuits due to incorrectly cited laws or improper legal definitions. According to Wade Ballou, the former head of the legal office of the lower chamber of Congress, AI often misses nuances of legislative vocabulary, for example, whether a particular sum of money should appear in a document as a "tax credit, tax deduction, tax break or grant."

According to one of Politico's sources, the problem is aggravated by the fact that "people using AI to write laws may not advertise it." He emphasized that the use of neural networks in work is still perceived on Capitol Hill as taboo.

Source: TASS