The White House has rolled out its AI Action Plan
A sweeping, 90+ point blueprint designed to accelerate U.S. AI development, cut through infrastructure red tape, and sharpen America’s competitive edge on the world stage. Tucked between the big-ticket items like R&D funding and deregulation is one provision grabbing most of the headlines: a new executive order requiring AI systems used by federal agencies to be ideologically neutral.
On paper, that means preventing “ideological distortion,” prioritizing historical accuracy, and steering clear of diversity, equity, and inclusion (DEI) frameworks like unconscious bias or systemic racism. Critics and supporters alike have labeled this push “Preventing Woke AI”, but the underlying question is serious: can AI ever truly be neutral, or will it always carry the fingerprints of the humans who built it?
Defining “Woke AI”
The order defines “woke AI” as systems embedding ideological bias, often through a DEI lens. The mandate is simple: AI should seek truth without filtering it through a political worldview. But that’s easier said than done—because AI is built by humans, and humans aren’t neutral.
The Roots of the Bias Debate
Concerns about bias in AI have existed as long as modern machine learning. AI models learn from human-created data, and deciding what data to include (or exclude) is already a value judgment. Even choosing a history textbook is political; every textbook reflects its own perspective, shaping how people understand the past (Wikipedia).
Politically, conservatives often cite AI outputs they feel align with progressive values as evidence of bias. Supporters of the neutrality mandate see it as a corrective measure. Many progressives counter that removing terms like “systemic racism” risks whitewashing history, where “neutrality” becomes a stand-in for erasure.
Evidence of bias is well-documented. Research by Joy Buolamwini and Timnit Gebru found facial recognition error rates as high as 34.7% for darker-skinned women, compared to 0.8% for lighter-skinned men (Wikipedia). Amazon’s now-defunct hiring tool penalized resumes containing “women’s,” showing how biased datasets create biased systems. Large language models have reinforced stereotypes, associating “nurse” with women and “engineer” with men, echoing decades of online prejudices.
To address this, companies added human oversight: annotators, policy teams, guardrails. But every intervention involves human judgment, which is inevitably shaped by the politics and culture of its time.
Overcorrection vs. Under-correction—Navigating a Tightrope
Bias mitigation can go too far. Google’s Gemini faced backlash after producing racially diverse images of Nazis—a well-intentioned inclusivity effort that crossed into historical inaccuracy (The Verge).
On the flip side, Elon Musk’s xAI dialed back moderation in the name of free expression. Within weeks, The Washington Post found its chatbot, Grok, generating antisemitic and harmful responses, showing how removing constraints can revive old prejudices (The Washington Post).
The balance is tricky: overcorrect and you distort reality; undercorrect and discrimination comes roaring back.
When “Neutral” Isn’t Neutral—Historical Echoes
History is full of “neutrality” used as propaganda. Stalin’s USSR erased dissenters from official records, replacing them with doctored narratives. Post–Civil War Southern textbooks glorified the Confederacy while minimizing slavery—a curated “objectivity” that served political ends (The New Yorker)(Time).
In both cases, neutrality wasn’t truth—it was the story those in power wanted told.
A Nation Divided on AI’s Future
Reactions to the neutrality mandate vary:
- Rumman Chowdhury, executive director of Humane Intelligence, told The Washington Post that phrases like “free of ideological bias” may sound appealing—but are “impossible to do in practice (Washington Post)..”
- Fabio Motoki, a lecturer at the University of East Anglia, echoed that it’s “very, very difficult to steer these models to do what we want. (Washington Post)”
- Kit Walsh at the Electronic Frontier Foundation warns that while the government can buy AI that meets neutral standards, it cannot stop developers from offering more ideologically expressive versions elsewhere (The Guardian).
- Neil Chilson, AI policy head at the Abundance Institute, views the order as moderate—requiring only transparency—not the absence of agenda (Washington Post).
- Critics such as Jim Secreto warn it may function as soft coercion, pushing firms toward self-censorship to win federal contracts—raising red flags over democratic norms (AP News).
- Even Marc Andreessen weighed in, alleging that tech lawyers and policy teams intentionally encoded social agendas into AI systems (AP News).
The Enforcement Problem: Who Defines Neutrality?
The order leans on procurement: only vendors passing “neutrality” tests, criteria expected within 90 days, can sell AI to the government. But “neutral” could mean evaluating outputs, auditing training data, or policing developer rules. And as political leadership changes, so could the definition.
That makes neutrality not just a standard, but potentially a political weapon.
The Bigger Questions Await
Beneath the partisan sparring are unresolved issues:
- Is bias inevitable, or can it be engineered out?
- Who decides what’s neutral?
- Does neutrality risk becoming censorship?
- Should AI engage with fringe or false ideas to show it can reason through them?
Our Take
The AI Action Plan isn’t just a policy, it’s a reset for how AI will be built, funded, and regulated in the U.S. The bias debate isn’t new, but the neutrality mandate forces it into the spotlight. At IP.com, we believe AI should be grounded in verifiable data and transparent processes—not shifting political winds.
Whether total neutrality is possible, or even desirable, remains unsettled. But the decisions made now will shape both the direction of AI and the pace of innovation.
Because in the end, “woke AI” isn’t only about algorithms—it’s about who gets to define truth.
