Mr. Griffith presses witnesses on AI political bias, cites Washington Post study

Mr. Griffith raised concerns about political bias in AI chatbots during a committee hearing, citing a Washington Post article and asking witnesses whether the political leanings of people who train models could explain left-leaning outputs. He also thanked the Congressional Research Service (CRS) for a jurisdictional paper that clarified committee responsibilities.

Griffith said he values AI as a research assistant but warned the committee to be skeptical after a Washington Post test, which the congressman summarized as finding system outputs skewed toward left-leaning answers on political questions. "If you ask a political question, OpenAI came back 80% of the time with a left leaning argument," he said, and recited other percentages he attributed to the Post's testing of specific chatbots. Griffith asked that the Post article by Kevin Shaw be submitted for the record and invoked statements from Dr. Don Fritz urging caution.

The witness, identified only in the transcript as a witness, agreed with the need for skepticism and said the central challenge is not that AI will make research obsolete but that AI can make "plausible looking" research cheap and ubiquitous. The witness said that the technology should be "married" to human expertise and experience, arguing that human review remains critical.

Griffith framed his line of questioning around how training and labeling choices might shape outputs, asking whether hiring patterns among people used to train models could yield systematic political leanings. He urged clarity about how additional funding to train models would be used and whether that would affect answers produced by AI systems.

The hearing record, as summarized from the transcript, shows no formal vote or committee action on the matter. Mr. Griffith submitted the Washington Post article for the record and asked for additional discussion about training practices and oversight. The witness recommended combining AI tools with human subject-matter expertise to avoid overreliance on automated outputs.

The committee's discussion continued beyond the transcript excerpt provided; no formal decision on oversight or funding for AI training was recorded in the available text.