Liability, regulation, and AI’s new false dichotomy

Gary Marcus published a post arguing against treating liability for AI companies and regulation as alternatives, describing the framing as a "false dichotomy." He said he agrees with Senator Josh Hawley (R-MO) that AI companies should be liable for harm their systems cause.

Marcus pointed to what he called a growing trend of people on the technolibertarian right calling for AI liability while using it as an argument against regulation. He cited an interview with Joe Lonsdale that was amplified by David Sacks, saying the liability portion was true but that the attack on regulation does not follow. He also cited Mark Cuban's reply to Sacks asking who should be trusted to pass new AI-specific laws or apply current laws to AI, and said that without regulation suited to AI, liability is pointless.

Karen Kornbluh, described as former Director of the National AI Office and former Deputy Principal Deputy CTO to the White House, amplified Marcus's post, writing that for other powerful industries such as autos, planes, drugs and financial services there are government standards, verification, incident reporting and investigation, and liability for violations and harms, and that none of these should be sacrificed in a rush.

Marcus recalled that Hawley questioned him on whether liability alone might suffice when he testified before the U.S. Senate in May 2023. Marcus said he answered no then, as he does now, arguing for both liability and regulation. He recalled Hawley raising concerns about agency capture and proposing a broad private right of action instead, asking why people could not simply sue; Marcus responded that current laws predate AI and do not give enough coverage, that litigation can take a decade or more, and that gaps exist, including uncertainty about how copyright law applies to AI and whether Section 230 applies.

Marcus wrote that Hawley has since become a champion of both liability and regulation, citing his bipartisan AI framework. Marcus said investors like Sacks may want to shield AI companies from regulation, but that others should not be fooled. In a postscript, Marcus wrote that AI is getting harder to predict, that a lot of bad things will happen, that there will be a lot of litigation, and that tech companies can afford very expensive lawyers.