Everyone's a loser in Trump's AI Action Plan

On July 23, the Trump Administration released its long-awaited AI Action Plan. Short of copyright exemptions for model training, the administration appears ready to give OpenAI, Anthropic, Google and other major players nearly everything they asked of the White House during public consultation. However, according to Travis Hall, the director of state engagement at the Center for Democracy and Technology, Trump's policy vision would put states, and tech companies themselves, in a position of "extraordinary regulatory uncertainty."

It starts with Trump's attempt to prevent states from regulating AI systems. In the original draft of his recently passed tax megabill, the president included an amendment that would have imposed a 10-year moratorium on any state-level AI regulation. Eventually, that clause was removed from the legislation in a decisive 99-1 vote by the Senate.

It appears Trump didn't get the message. In his Action Plan, the president signals he will order federal agencies to only award "AI-related" funding to states without "burdensome" AI regulations.

"It is not really clear which discretionary funds will be deemed to be 'AI-related', and it's also not clear which current state laws — and which future proposals — will be deemed 'burdensome' or as 'hinder[ing] the effectiveness' of federal funds. This leaves state legislators, governors, and other state-level leaders in a tight spot," said Grace Gedye, policy analyst for Consumer Reports. "It is extremely vague, and I think that is by design," adds Hall.

The issue with the proposal is nearly any discretionary funding could be deemed AI-related. Hall suggests a scenario where a law like the Colorado Artificial Intelligence Act (CAIA), which is designed to protect people against algorithmic discrimination, could be seen as hindering funding meant to provide schools with technology enrichment because they plan to teach their students about AI.

The potential for a "generous" reading of "AI-related" is far-reaching. Everything from broadband to highway infrastructure funding could be put at risk because machine learning technologies have begun to touch every part of modern life.

On its own, that would be bad enough, but the president also wants the Federal Communications Commission (FCC) to evaluate whether state AI regulations interfere with its "ability to carry out its obligations and authorities under the Communications Act of 1934." If Trump were to somehow enact this part of this plan, it would transform the FCC into something very different from what it is today.

"The idea that the FCC has authority over artificial intelligence is really extending the Communications Act beyond all recognition," said Cody Venzke, senior policy counsel at the American Civil Liberties Union. "It traditionally has not had jurisdiction over things like websites or social media. It's not a privacy agency, and so given the fact that the FCC is not a full-service technology regulator, it's really hard to see how it has authority over AI."

Hall notes this part of Trump's plan is particularly worrisome in light of how the president has limited the agency's independence. In March, Trump illegally fired two of the FCC's Democratic commissioners. In July, the Commission's sole remaining Democrat, Anna Gomez, accused Republican Chair Brendan Carr of "weaponizing" the agency "to silence critics."

"It's baffling that the president is choosing to go it alone and unilaterally try to impose a backdoor state moratorium through the FCC, distorting their own statute beyond recognition by finding federal funds that might be tangentially related to AI and imposing new conditions on them," said Venzke.

Igor Bonifacic for Engadget

On Wednesday, the president also signed three executive orders to kick off his AI agenda. One of those, titled "Preventing Woke AI in the Federal Government," limits federal agencies to only obtaining those AI systems that are "truth-seeking," and free of ideology. "LLMs shall be neutral, nonpartisan tools that do not manipulate responses in favor of ideological dogmas such as DEI," the order states. "LLMs shall prioritize historical accuracy, scientific inquiry, and objectivity, and shall acknowledge uncertainty where reliable information is incomplete or contradictory."

The pitfalls of such a policy should be obvious. "The project of determining what is absolute truth and ideological neutrality is a hopeless task," said Venzke. "Obviously you don't want government services to be politicized, but the mandates and executive order are not workable and leave serious questions."

"It's very apparent that their goal is not neutrality," adds Hall. "What they're putting forward is, in fact, a requirement for ideological bias, which is theirs, and which they're calling neutral. With that in mind, what they're actually requiring is that LLMs procured by the federal government include their own ideological bias and slant."

Trump's executive order creates an arbitrary political test that companies like OpenAI must pass or risk losing government contracts — something AI firms are actively courting. At the start of the year, OpenAI debuted ChatGPT Gov, a version of its chatbot designed for government agency use. xAI announced Grok for Government last week. "If you're building LLMs to satisfy government procurement requirements, there's a real concern that it's going to carry over to wider private uses," said Venzke.

There's a greater likelihood of consumer-facing AI products conforming to these same reactionary parameters if the Trump administration should somehow find a way to empower the FCC to regulate AI. Under Brendan Carr, the Commission has already used its regulatory power to strongarm companies to align with the president's stance on diversity, equity and inclusion. In May, Verizon won FCC approval for its $20 billion merger with Frontier after promising to end all DEI-related practices. Skydance made a similar commitment to close its $8 billion acquisition of Paramount Global.

Even without direct government pressure to do so, Elon Musk's Grok chatbot has demonstrated twice this year what a "maximally truth-seeking" outcome can look like. First, in mid-May it made unprompted claims about "white genocide" in South Africa; more recently it went full "MechaHitler" and took a hard turn toward anti-semitism.

According to Venzke, Trump's entire plan to preempt states from regulating AI is "probably illegal," but that's a small comfort when the president has actively flouted the law far too many times to count less than a year into his second term, and the courts haven't always ruled against his behavior.

"It is possible that the administration will read the directives from the AI Action Plan narrowly and proceed in a thoughtful way about the FCC jurisdiction, about when federal programs actually create a conflict with state laws, and that is a very different conversation. But right now, the administration has opened the door to broad, sort of reckless preemption of state laws, and that is simply going to pave the way for harmful, not effective, AI."

This article originally appeared on Engadget at https://www.engadget.com/ai/everyones-a-loser-in-trumps-ai-action-plan-160023247.html?src=rss https://www.engadget.com/ai/everyones-a-loser-in-trumps-ai-action-plan-160023247.html?src=rss