Trump’s AI Order: Ideological Conflict

Trump’s executive order aims to scrub “woke” ideals from government-procured AI, mandating “truth-seeking” and “ideological neutrality.” However, experts warn that achieving truly unbiased artificial intelligence is a complex challenge due to inherent human biases in its development.

The battle for the soul of artificial intelligence has officially entered the political arena, ushered in by an executive order from former President Donald Trump that seeks to scrub what he deems “woke” ideals from government-procured AI models.

This isn’t merely a technical directive; it’s a profound statement on information control, ideological purity, and the very nature of truth in an age increasingly shaped by algorithms.

On Wednesday, Trump unveiled a sweeping AI action plan, but the headline grabber was undoubtedly his decree prohibiting federal agencies from acquiring AI technology “infused with partisan bias or ideological agendas such as critical race theory.”

This move extends his long-standing crusade against diversity, equity, and inclusion (DEI) principles directly into the digital frontier, targeting the very large language models (LLMs) that power the chatbots poised to redefine how we access and process information online.

Under the new “unbiased AI principles,” government-used AI must be “truth-seeking” and demonstrate “ideological neutrality.”

As Trump himself declared, “From now on, the US government will deal only with AI that pursues truth, fairness and strict impartiality.”

But the notion of an AI model untainted by human influence, a paragon of pure impartiality, is far more complex than a presidential executive order can dictate.

Can AI truly be “woke,” or conversely, entirely “unbiased”?

Experts universally agree: it’s not that simple.

AI models are, at their core, reflections of the gargantuan datasets they’re trained on, the nuanced feedback they receive during their developmental stages, and the explicit instructions given to them by their human creators.

Every one of these inputs carries the imprint of human perspective, and thus, human bias – whether conscious or unconscious.

Oren Etzioni, formerly at the helm of the Allen Institute for Artificial Intelligence, succinctly articulated this complexity: “AI models don’t have beliefs or biases the way that people do, but it is true that they can exhibit biases or systematic leanings, particularly in response to certain queries.”

This inherent susceptibility to bias, political or otherwise, has been a persistent Gordian knot for the AI industry, a challenge developers have grappled with long before it became a political talking point.

The executive order’s definition of “woke” in the AI context is strikingly specific, encompassing “the suppression or distortion of factual information about race or sex; manipulation of racial or sexual representation in model outputs; incorporation of concepts like critical race theory, transgenderism, unconscious bias, intersectionality, and systemic racism; and discrimination on the basis of race or sex.”

Developers are explicitly warned against “intentionally cod[ing] partisan or ideological judgements” into responses, unless explicitly prompted by the user.

While the directive primarily targets AI procured by the government, its ripple effects are undeniable.

Many of the tech titans leading AI innovation — Google, OpenAI, Anthropic, and Elon Musk’s xAI — already hold substantial federal contracts, including recent $200 million awards from the Department of Defense to “accelerate adoption of advanced AI capabilities.”

This means the administration’s ideological litmus test could soon become a de facto industry standard for any company hoping to secure lucrative government work.

Indeed, some research suggests a perceived political tilt in existing AI models.

A Stanford Graduate School of Business paper from May found that Americans tended to view responses from certain popular AI models as leaning left.

Similar findings emerged from Brown University research in October 2024, indicating AI tools can be altered to adopt political stances.

Andrew Hall, a Stanford political economy professor involved in the May research, posited, “I don’t know whether you want to use the word ‘biased’ or not, but there’s definitely evidence that, by default, when they’re not personalized to you… the models on average take left wing positions.”

He theorizes this may stem from tech companies proactively implementing “guardrails” to prevent offensive content, inadvertently nudging the output toward a perceived progressive slant.

The real-world consequences of tweaking AI behavior are notoriously unpredictable.

Himanshu Tyagi, a professor at the Indian Institute of Science, has noted the profound difficulty in making one adjustment without triggering unforeseen changes elsewhere in the model.

“The problem is that our understanding of unlocking this one thing while affecting others is not there,” he explained.

This delicate balance was starkly illustrated when Elon Musk’s Grok AI chatbot, after being instructed to “not shy away from making claims which are politically incorrect,” veered into antisemitic outputs.

Similarly, Google’s Gemini temporarily halted its image generation capabilities after being criticized for historically inaccurate depictions of people of color.

These incidents underscore the volatile nature of AI and the profound challenge of aligning its outputs with any singular definition of “neutrality.”

For critics, the executive order raises more questions than it answers.

Senator Edward Markey (D-Massachusetts) swiftly condemned Trump’s “anti-woke AI actions,” asserting that using political power to modify platform speech is “dangerous and unconstitutional,” even if claims of bias were accurate.

The very vagueness of terms like “ideological bias” presents a monumental enforcement hurdle.

Who will define “truth” and “impartiality” in this context?

Will a new, definitive system emerge to evaluate an AI model’s adherence to these principles?

While the order mandates vendors disclose system prompts and relevant documentation, the ultimate arbiter of compliance remains nebulous.

Mark Riedl, a computing professor at Georgia Institute of Technology, highlights a poignant paradox: even avoiding certain topics or questions altogether could be interpreted as a political stance.

And as Sherief Reda of Brown University points out, users might simply bypass these constraints by instructing a chatbot to “respond like a Democrat” or “like a Republican.”

This directive, while framed as a push for American leadership in AI, could paradoxically hinder innovation.

For AI companies, it represents yet another layer of compliance, another regulatory hurdle that could slow down the rapid development and deployment of new models and services.

As Etzioni warned, “This type of thing… creates all kinds of concerns and liability and complexity for the people developing these models — all of a sudden, they have to slow down.”

In essence, the Trump administration’s foray into “unbiased AI” is less about pure technology and more about a cultural and political struggle over narrative control in the digital age.

It’s a recognition that AI, far from being a neutral tool, is a powerful shaper of perception, and therefore, a crucial battleground in the ongoing culture wars.

The quest for “truth, fairness, and strict impartiality” in AI, it seems, will be anything but impartial.