In a significant move regarding artificial intelligence, President Donald Trump has signed an executive order aimed at addressing ideological biases found in AI models, particularly in the context of federal procurement.
This order emphasizes the necessity for AI to be reliable and objective, asserting that biases can compromise the accuracy of AI-generated outputs.
The announcement underscores a growing concern that biases related to concepts such as ‘diversity, equity, and inclusion’ (DEI) threaten the integrity of AI technologies.
According to the order, DEI can lead to the distortion or suppression of factual information, which can ultimately undermine the truths that AI models are meant to present.
Among the practices criticized are alterations of historical facts, including the misrepresentation of well-known figures based on their race or gender, as well as refusal to fulfill requests based on elements of a person’s identity such as race or gender.
The executive order cites examples where AI outputs have been manipulated in favor of ideological narratives rather than sticking to factual representations.
The order expresses that recent instances showcase the perils these ideological biases pose to the reliability of AI technology.
In cases highlighted by the order, certain AI models exhibited biased behavior—for instance, altering the race or sex of historical figures upon user requests, deviating from the principles of accuracy.
Another troubling example involved an AI system that actively refused to acknowledge certain achievements or contributions based on race, revealing a clear inconsistency in criteria applied to different racial groups.
This executive action builds upon earlier efforts initiated under Executive Order 13960 from December 3, 2020, which aimed to promote trustworthy AI usage within the Federal Government.
The objective is to ensure that federally procured AI technologies are dependable and grounded in facts, rather than influenced by social agendas or political ideologies.
In the order, Trump lays out the definition of an ‘agency’ and delineates the responsibilities of agency heads when it comes to AI procurement, particularly concerning large language models (LLMs).
Two key principles have been established, termed the ‘Unbiased AI Principles,’ which agencies must adhere to when procuring LLMs.
The first principle, ‘Truth-seeking,’ asserts that AI models must respond truthfully to factual queries, prioritizing accuracy and acknowledging uncertainties when information may be incomplete.
The second principle is that LLMs should embody ‘Ideological Neutrality,’ refraining from incorporating partisan biases or ideological agendas in their outputs.
Agency heads will be guided by this order to only procure AI technologies aligning with these principles, thereby safeguarding the integrity of information disseminated through AI systems.
To facilitate the implementation of these principles, the Director of the Office of Management and Budget (OMB) is instructed to draft and issue guidance to federal agencies within 120 days.
This guidance shall address several considerations, including allowing vendors to maintain transparency regarding the ideological judgments of their models without disclosing sensitive technical details that may compromise proprietary information.
Moreover, the guidance aims to be flexible, permitting various approaches to meet the Unbiased AI Principles, while also specifying factors for agency heads to assess compliance and applicability of these principles to different types of AI models.
In addition, within contract terms, future LLM procurements will mandate compliance with these principles, with terms including consequences for noncompliance.
Agencies are also encouraged to revise existing contracts where possible to ensure adherence to the newly instituted policies.
Within 90 days of receiving guidance from the OMB, agencies are expected to adopt procedures guaranteeing that all new LLM procurements will comply with the Unbiased AI Principles.
The order concludes with provisions intended to preserve the authority of executive departments and agencies while emphasizing that this order does not establish any private rights enforceable in courts against the federal government or its employees.
The overarching goal of this executive order is to foster an environment in which reliable and accurate AI technologies can thrive, unaffected by ideological biases, ultimately leading to more trustworthy interactions between the government and its citizens.
President Donald Trump’s order marks a decisive step towards securing the integrity of AI systems while acknowledging the ongoing challenges posed by ideological influences within this rapidly evolving technological landscape.
image source from：whitehouse
