President Donald Trump asserts that “woke AI” presents a considerable danger to truth and independent thought. Detractors contend that his approach to address alleged woke AI undermines free speech and might breach the First Amendment.
The phrase gained traction following the president’s announcement of The White House’s AI Action Plan on Wednesday, July 23, as part of initiatives to secure American dominance in the swiftly growing AI field.
The AI Action Plan consists of three executive directives:
– Promoting the Export of the American AI Technology Stack
– Accelerating Federal Permitting of Data Center Infrastructure
– Preventing Woke AI in the Federal Government
The plan aligns with the interests of major tech companies and incorporates terminology like “truth-seeking” from AI figures such as Elon Musk. The directive concerning woke AI casts large-language models with purported liberal biases as a newly emerging right-wing issue.
Pinpointing woke AI is difficult, as interpretations differ based on viewpoint. A spokesperson from the White House cited a fact sheet associated with the woke AI directive, characterizing it as “biased AI outputs influenced by ideologies such as diversity, equity, and inclusion (DEI) at the expense of accuracy.”
What is Woke AI? Analyzing the White House’s definition
Curiously, the woke AI executive order does not directly employ the term. The definitions section of the order also fails to provide a clear explanation. Nonetheless, the White House’s fact sheet suggests that government officials should utilize only “large language models (LLMs) that comply with ‘Unbiased AI Principles’ as outlined in the Order: truth-seeking and ideological neutrality.”
The fact sheet describes “truth-seeking” and “ideological neutrality” in the following manner:
Truth-seeking entails that LLMs should emphasize historical accuracy, scientific inquiry, and objectivity, recognizing uncertainty when information is incomplete or contradictory.
Ideological neutrality insists that LLMs should serve as nonpartisan instruments that do not endorse ideological doctrines like DEI, and developers should not deliberately input partisan or ideological biases into an LLM’s outputs unless prompted by or accessible to the user.
The White House characterizes woke AI as LLMs that lack truth-seeking or ideological neutrality. The order underscores possible biases, including “critical race theory, transgenderism, unconscious bias, intersectionality, and systemic racism.” A cultural discussion exists regarding whether these subjects carry inherent bias.
AI firms that do not align with the White House’s standards might be barred from lucrative federal contracts. The order’s characterization of widely accepted liberal beliefs as biased could pressure AI companies into modifying their models’ inputs and outputs.
The Trump administration promotes free speech, yet critics argue this directive endangers it.
“The action plan’s ‘Ensure that Frontier AI Protects Free Speech and American Values’ section appears aimed at controlling AI information and might suggest measures that infringe upon the First Amendment,” remarked Kit Walsh, Director of AI and Access-to-Knowledge Legal Projects at the Electronic Frontier Foundation, in a statement to Mashable. “Generative AI affects users’ First Amendment rights to access information and reflects the protected expressive choices of those who develop AI messages. The government cannot dictate ideas shared through AI any more than through newspapers or websites.”
“The government can select services for its use but cannot penalize publishers for delivering AI services that communicate ideas it dislikes,” Walsh stated.
Is Woke AI a genuine concern?
The response is contingent on political perspectives, and “woke” has become a contentious term. It originated within the Black community, referring to awareness of racial bias and injustice. Recently, conservatives have employed it as an insult against politically correct liberals.
Both liberals and conservatives express concern regarding bias in large-language models.
In November 2024, the Heritage Foundation, a conservative legal organization, held a YouTube panel addressing woke AI. Curt Levey, President of the Committee For Justice, shared his insights as a conservative lawyer with experience in the AI industry.
“I find it fascinating that both the left and right are worried about AI bias, yet they concentrate on different dimensions. The left is concerned about AI models discriminating against minority groups in areas like hiring, lending, bail amounts, and facial recognition. The right focuses on discrimination against conservative perspectives in large language models such as ChatGPT.”
Elon Musk contends that AI models adopt a woke perspective from their creators, which conflicts with the pursuit of maximal truth-seeking. Musk claims firms instruct AI to misrepresent for the sake of political correctness.
Levey noted that biased LLMs are not necessarily designed with bias in mind. Researchers constructing AI models make decisions about data, and many reside in liberal regions like the San Francisco Bay area, potentially harboring unconscious biases when choosing data.
A conservative using “unconscious bias” without irony? Unbelievable.
LLMs embody biases because we possess biases
AI models reflect the biases of their training data, replicating our own biases. They function like a mirror with an inclination to distort reality.
To conform to the Executive Order, AI companies could minimize “biased” responses by regulating training data or employing system prompts that dictate model outputs.
However, as xAI has demonstrated, this strategy can be problematic. xAI’s chatbot Grok fixated on “white genocide in South Africa” and