Should Donald Trump emerge victorious in the US presidential election this November, we may witness a significant relaxation of the constraints placed on artificial intelligence development, despite the escalating risks associated with flawed AI systems.
Trump’s potential second term could drastically transform—and potentially undermine—efforts aimed at safeguarding the American public from the various hazards posed by inadequately designed artificial intelligence. These dangers include disinformation, bias and discrimination, and the contamination of algorithms utilized in technologies such as self-driving vehicles.
The federal government has initiated oversight and guidance for AI firms, following an executive order issued by President Joe Biden in October 2023. However, Trump has expressed intentions to revoke that order, arguing that it “restricts AI innovation” and enforces “Radical Leftwing ideas” on AI development, as outlined in the Republican Party platform.
Trump’s commitment has energized opponents of the executive order who perceive it as not only unlawful and hazardous, but also as a hindrance to the US’s digital competition with China. Among these critics are many of Trump’s key supporters, including X CEO Elon Musk, venture capitalist Marc Andreessen, Republican members of Congress, and almost two dozen GOP state attorneys general. Trump’s running mate, Ohio senator JD Vance, firmly opposes any form of AI regulation, as noted in a report by The New York Times.
“Republicans are cautious about rushing to impose regulations on this industry,” remarks Jacob Helberg, a tech executive and AI advocate who has been referred to as “Silicon Valley’s Trump whisperer.”
However, technology and cybersecurity experts caution that removing the executive order’s safety and security measures could erode the reliability of AI models that are increasingly present in various facets of American life, including transportation, healthcare, employment, and surveillance.
The upcoming presidential election may play a critical role in determining whether AI evolves into an extraordinary tool for productivity or spirals into an unmanageable source of chaos.
Biden’s order encompasses a range of initiatives, from enhancing health care for veterans through AI to establishing safeguards for its application in drug discovery. Nevertheless, much of the political debate surrounding the executive order originates from two provisions related to digital security risks and the tangible safety implications involved.
One provision mandates that operators of advanced AI models must inform the government about their training methods and the measures taken to safeguard against tampering and theft. This includes sharing outcomes from “red-team tests” that are aimed at identifying weaknesses in AI systems by mimicking attacks. The other provision instructs the National Institute of Standards and Technology (NIST), which is part of the Commerce Department, to create guidelines that assist companies in building AI models that are protected from cyberattacks and free from biases.
The government is already making significant strides on these initiatives. They have proposed quarterly reporting requirements for AI developers, and NIST has released guidance documents aimed at AI, which cover topics such as risk management, secure software development, synthetic content watermarking, and tackling model abuse. They are also initiating various efforts to enhance model testing.
Proponents of these regulations argue they are crucial for ensuring basic governmental oversight of the swiftly growing AI sector and for encouraging developers to adopt better security protocols. Conversely, critics from conservative circles believe that the reporting obligation constitutes unlawful government intrusion, potentially stifling AI innovation and compromising the confidentiality of developers’ trade secrets. They view the NIST guidelines as a liberal strategy that could infuse AI with extreme leftist ideas regarding disinformation and bias, effectively censoring conservative perspectives.
During a rally in Cedar Rapids, Iowa, last December, Trump criticized Biden’s executive order, claiming without substantiation that the Biden administration had already misused AI for malicious ends.
“Once I am reelected,” he stated, “I will revoke Biden’s executive order regarding artificial intelligence and prohibit the use of AI to restrict the speech of American citizens starting on Day One.”
Biden’s initiative to gather information about the methods companies employ in developing, testing, and securing their AI models sparked significant controversy on Capitol Hill as soon as it was announced.
Republicans in Congress quickly criticized Biden’s rationale for the new mandate, which cited the 1950 Defense Production Act, a wartime statute allowing the government to direct private-sector actions to ensure a stable supply of goods and services. GOP lawmakers labeled Biden’s action inappropriate, illegal, and unnecessary.
Conservative voices have also condemned the reporting requirement as an imposition on the private sector. During a March hearing she led on “White House overreach on AI,” Representative Nancy Mace remarked that the provision “could deter potential innovators and hinder further breakthroughs akin to ChatGPT.”
Helberg expresses concerns that onerous requirements could favor established businesses while putting startups at a disadvantage. He also mentions that critics from Silicon Valley worry that these regulations might be a precursor to a licensing system where developers would need governmental approval to test their models.
Steve DelBianco, the head of the conservative tech organization NetChoice, argues that the obligation to disclose red-team testing results can be viewed as a form of censorship. He suggests that the government will likely focus on issues such as bias and misinformation. “I have significant concerns about a left-leaning administration … whose red-teaming evaluations might lead AI to restrict its outputs for fear of raising these issues,” he asserts.
Critics from the conservative side contend that any regulatory measures which hinder AI progress could have grave repercussions for the U.S. in the ongoing technology race with China.
“They are incredibly assertive, and they have made it a top priority to take the lead in AI as part of their strategy for succeeding in warfare,” Helberg states. “The disparity between our capabilities and those of the Chinese continues to narrow with each passing year.”
By incorporating social harms into its AI security guidelines, NIST has sparked outrage among conservatives and ignited a new front in the ongoing culture war regarding content moderation and free speech.
Republicans are decrying the NIST guidance as a covert form of government censorship. Senator Ted Cruz recently criticized what he referred to as NIST’s “woke AI ‘safety’ standards,” claiming they are part of a Biden administration “strategy to control speech” based on “vague” social harms. The organization NetChoice has cautioned NIST that it is overstepping its mandate with quasi-regulatory guidelines that disrupt “the proper balance between transparency and free speech.”
Many conservatives outright reject the notion that AI can contribute to social harms and should be designed to avoid them.
“This is a solution in search of a problem that really doesn’t exist,” Helberg states. “There hasn’t been substantial evidence of issues in AI discrimination.”
Multiple studies and investigations have consistently indicated that artificial intelligence models harbor inherent biases that facilitate discriminatory practices across various sectors such as employment, law enforcement, and health services. Research indicates that individuals who experience these biases might unconsciously integrate them into their beliefs.
Concerns among conservatives are often directed more towards the oversensitivity of AI companies in addressing these issues rather than the issues themselves. “There is a direct inverse correlation between the degree of wokeness in an AI and the AI’s effectiveness,” claims Helberg, referencing a previous problem encountered with Google’s generative AI platform.
On the other hand, Republicans are urging NIST to prioritize the physical dangers associated with AI, including its potential to assist terrorists in the creation of bioweapons (a concern noted in Biden’s executive order addressing these threats). Should Trump be elected again, it is anticipated that his appointed officials may downplay the importance of government studies on the social repercussions of AI. Helberg points out that the “substantial volume” of research into AI bias has overshadowed investigations into “more significant hazards linked to terrorism and biowarfare.”
AI professionals and politicians provide strong support for Biden’s agenda concerning AI safety.
These initiatives “ensure that the United States stays at the forefront” of AI development “while safeguarding Americans from possible risks,” states Representative Ted Lieu, the Democratic co-chair of the House’s AI task force.
The necessity of reporting requirements lies in their role in alerting the government to potentially hazardous new capabilities in increasingly advanced AI models, according to a U.S. government official involved in AI matters. Speaking on condition of anonymity to express candid thoughts, the official references OpenAI’s acknowledgment regarding its newest model’s “inconsistent refusal of requests to synthesize nerve agents.”
The official emphasizes that the reporting obligation is not excessively burdensome. They contend that, unlike AI regulations in the European Union and China, Biden’s executive order signifies “a very broad, light-touch approach that promotes innovation.”
Nick Reese, the first director of emerging technology for the Department of Homeland Security from 2019 to 2023, dismisses claims from conservative circles that the reporting requirement will endanger companies’ intellectual property. He believes it could actually be advantageous for startups by motivating them to create “more computationally efficient,” less data-intensive AI models that are under the reporting threshold.
According to Ami Fields-Meyer, who played a key role in drafting Biden’s executive order as a technology official in the White House, the formidable capabilities of AI necessitate government supervision.
“We are dealing with companies claiming to create some of the most influential systems ever seen,” Fields-Meyer explains. “The government’s primary responsibility is to safeguard individuals. Simply saying, ‘Trust us, we know what we’re doing,’ is not a particularly persuasive stance.”
Experts highlight the importance of NIST’s security guidelines as crucial for integrating safeguards into emerging technologies. They emphasize that flawed AI systems can lead to significant societal issues, including biases in housing and lending, as well as wrongful denial of government assistance.
Trump’s initial AI executive order mandated that federal AI systems uphold civil rights, a requirement that will necessitate investigation into social repercussions.
The AI industry has predominantly embraced Biden’s safety agenda. “What we’re hearing is that it’s broadly useful to have this stuff spelled out,” a US official comments. For emerging companies with smaller teams, “it enhances the capability of their members to tackle these issues.”
Reversing Biden’s executive order would convey a troubling message that “the US government is opting for a hands-off stance regarding AI safety,” states Michael Daniel, who previously served as a presidential cyber adviser and now directs the Cyber Threat Alliance, a nonprofit focused on information sharing.
Regarding competition with China, supporters of the executive order argue that safety regulations will ultimately help the US to succeed by ensuring that American AI models outperform their Chinese counterparts while safeguarding them from Beijing’s economic espionage efforts.
If Trump returns to the White House next month, anticipate a significant shift in the government’s approach to AI safety.
Republicans are looking to address potential AI risks by utilizing “existing tort and statutory laws” instead of implementing extensive new regulations on the technology, according to Helberg. They emphasize a “much greater focus on maximizing the opportunity offered by AI, rather than concentrating excessively on risk mitigation.” This approach may jeopardize the reporting requirement and could impact some of the guidance provided by NIST.
The reporting requirement might also encounter legal challenges following the Supreme Court’s recent decision to weaken the deference historically granted to agencies while courts assess their regulations.
Moreover, Republican resistance could threaten NIST’s voluntary AI testing partnerships with prominent companies. A US official poses the question, “What happens to those commitments in a new administration?”
This division over AI issues has left technologists exasperated, particularly those who fear that Trump may hinder progress towards creating safer AI models.
“Alongside the promises of AI are perils,” says Nicol Turner Lee, the director of the Brookings Institution’s Center for Technology Innovation, “and it is vital that the next president continue to ensure the safety and security of these systems.”