The legal pushback that could complicate Trump's AI plan

Yahoo

2 days ago

The Trump administration's new 'Action Plan' to support US dominance in artificial intelligence is all but certain to be met with legal pushback, especially from civil rights advocates and copyright holders.
President Trump and administration officials spent Wednesday promoting the plan and three accompanying executive orders after repealing much of President Biden's comparatively protectionist AI policies with a prior executive order in January.
Key pillars of the new Trump plan include accelerating permitting to build the data centers needed to create AI technologies, expanding the export of US-made AI, and freeing foundational AI models of ideological bias.
'From now on, the US government will deal only with AI that pursues truth, fairness, and strict impartiality,' Trump said on Wednesday in remarks from an AI Summit at Andrew W. Mellon Auditorium in Washington, D.C.
The plan also seeks to clear away legal headwinds, such as state-specific AI restrictions, and environmental and copyright laws.
Constitutional law scholars say the administration can expect litigation to target his directives that are on less solid legal footing, such as his executive order requiring bias-free algorithms.
That order bans the federal government from procuring AI technology infused with partisan bias and ideological agendas.
The requirement may be impossible to fulfill, the lawyers said. And even if eradicating bias were possible, government discrimination against developers may not hold up under a First Amendment challenge, they added.
'Any executive order that purports to require neutrality among AI doesn't understand how AI works,' said Stanford University technology law professor Mark Lemley.
He cautioned that if the government were to deny developers from securing government contracts based on viewpoints expressed in their technologies, it could draw constitutional challenges.
Technology lawyer Star Kashman agreed, and added that the concept of unbiased AI is 'nice,' but in practice would limit the use of all AI systems in this early stage of development.
'All AI systems carry inherent bias,' Kashman said.
Further complicating the idea, Kashman said, is that the orders seem to penalize companies based on the perceived ideological leanings of their AI systems or their training data sets.
'What would it mean for an AI system to be 'biased'...and who determines that standard?
'Who decides what the truth is?'
AI systems are trained on content across the internet, using data written by humans. Every person has their own bias, which then slips into AI models that process information pulled from the web.
'The way I see it is that there is definitely bias in AI," Cornell University assistant professor of information technology Aditya Vashistha told Yahoo Finance.
"I ask people, do you see bias on the internet? Do you see that people have some thoughts around certain demographics, identities, cultures, and so on? And usually, the answer is 'Yes'.'
Even evaluating AI systems for bias is a fraught exercise, he added.
'If you want to design an AI technology which is telling the truth, who decides what the truth is, and how do you even measure it?'
Vashistha pointed to language bias in AI models, explaining that while Hindi is spoken by more than 500 million people, far more than the population of the US, roughly 60% of the data online is English and just 0.6% is in Hindi.
That leads to AI systems that focus on English first, while leaving other languages out.
Another hurdle ripe for challenge is the administration's plan to provide some level of insulation to large language models from copyright infringement claims.
Dozens of copyright holders have sued large language model developers, including Meta (META) and Anthropic (ANTH.PVT), arguing that the developers must pay rights holders before allowing generative AI software to interpret their works for profit. Rights holders also argue that the AI output cannot resemble their original works.
"There is no predicting what's going to come out the other end of those cases," Courtney Lytle Sarnow, an intellectual property partner with CM Law, told Yahoo Finance in June.
Sarnow and other intellectual property experts said they expect the disputes will end up in appeals to the US Supreme Court.
'Of course, you can't copy or plagiarize an article,' Trump said. 'But if you read an article and learn from it, we have to allow AI to use that pool of knowledge without going through the complexity of contract negotiations, of which there would be thousands.'