Artificial Intelligence (AI) has been a topic of significant hype, yet recent surveys indicate that a majority of the U.S. public believes AI is more likely to cause harm than benefit. The potential for harm extends across various domains, including threats to workers’ rights, climate impacts, and biases in hiring or medical diagnoses. One area that has not been extensively examined is AI’s ability to influence people without their knowledge, thereby threatening their agency, autonomy, and decision-making capabilities. For instance, the rise of ChatGPT has been shown to correlate with changes in the words people use in conversation. As models are trained on data exhibiting these shifts, they adopt these changes to an even greater extent. While language change might seem benign, similar feedback loops can exist in domains that govern access to opportunities and resources, such as policing, education, and hiring. These loops can cause widespread societal harm without the knowledge or consent of decision-makers.
Impact of AI on Decision-Making
A large-scale experiment was conducted where human subjects screened resumes in collaboration with racially biased AI models. The findings revealed that people cannot adequately identify and mitigate traces of AI biases that propagate into their decision-making. This aligns with similar findings from experiments in emotional perception and medical diagnostics, suggesting that AI’s threat to autonomous human decision-making is a widespread problem. Current AI policy recommendations and regulations that call for humans-in-the-loop for high-stakes decisions fall short of addressing this impact. Additional solutions are needed, such as financial support for AI impact assessments, regulations encompassing both unintentional and intentional AI harms, and incentives for developing responsible AI systems.
Study on AI’s Influence on Decision-Making
AI’s influence on decision-making and autonomy was further explored in a study where participants made hiring decisions in collaboration with a simulated AI system. Over 500 people completed cognitive assessments of their explicit and implicit biases and decision-making scenarios. Participants were exposed to different AI conditions: no AI, biased AI that reinforced or contradicted common stereotypes, and unbiased AI. The results showed that interacting with biased AI made respondents more likely to make biased decisions themselves, regardless of whether the bias aligned with or contradicted common stereotypes. When the AI reinforced race-occupation stereotypes, respondents selected majority-white candidates 90.4% of the time. When the AI contradicted those stereotypes, respondents selected majority-non-white candidates 90.7% of the time. In contrast, without any AI recommendations, respondents chose majority-white candidates 49.3% of the time and non-white candidates 50.7% of the time. These results demonstrate how strongly AI can shape human choices, posing a serious threat to people’s ability to make autonomous decisions free from hidden influence or coercion.
Legislative Proposals to Address AI Harm
In recent years, lawmakers have introduced numerous legislative proposals aimed at reducing the harmful impacts of AI for high-stakes decision-making. In 2024, the Colorado AI Act was introduced to prevent intentional and unintentional discrimination caused by AI systems making high-impact decisions. Though set to take effect in 2026, legislators are still debating terms of the bill, primarily concerning liability in cases of disparate impact. In 2025, Texas enacted legislation regulating AI with provisions for nondiscrimination in AI use, but developers and deployers are only liable if they developed or used AI systems with the intention of causing discrimination. The Virginia legislature also passed regulation aimed at curbing AI discrimination in 2025, though it was vetoed due to concerns about burdens on smaller firms and threats to AI innovation. Fewer proposals have been made to prevent AI or other technologies from interfering with individual autonomy. Historically, protecting individual autonomy from technological interference through regulation has been challenging. For example, social media’s impact on both children’s and adults’ autonomy remains largely unregulated in the United States. The European Union, however, has prohibited social media platforms from using non-transparent techniques to change users’ behavior and declared that using AI for cognitive behavioral manipulation is unacceptable.
Need for Comprehensive Legislation
As AI capabilities and applications advance rapidly, developing comprehensive legislation to prevent discrimination and protect individual autonomy and safety will require building upon existing regulation and looking to related concepts such as intentionality, privacy, and freedom. Supporting efforts to improve AI impact assessments is crucial. Without understanding the human and societal effects of AI proliferation, it is difficult to craft policies that aim to curb harmful impacts. The speed at which new AI technologies are deployed far exceeds the rate at which reliable, valid, and generalizable AI evaluations are developed. To address this gap, a greater portion of funding for general AI development should be allocated to projects that specifically aim to improve AI evaluation standards.
Importance of Addressing Unintentional Impacts
Including provisions for unintentional impacts in AI regulation is also essential. The Equal Employment Opportunity Commission (EEOC) halted investigations into disparate impact claims, which could undermine efforts to prevent algorithmic discrimination. Comprehensive AI legislation should address both disparate impact and disparate treatment. While some proposals treat unintentional harms as less severe than intentional ones, these harms must be evaluated contextually. Unintentional disadvantaging bias can produce serious real-world consequences, such as longer wait times for organ transplants for Black patients, while intentional corrective bias can improve diagnostic accuracy or support more equitable resource allocation.
Incentivizing responsible AI development is another critical step. Regulation can drive positive innovation that doesn’t interfere with users’ privacy and preserves their rights. Approaches to encourage this include tax incentives to offset costs of responsible AI development, increased industry-academia cooperation, and educational curricula that teach computer science as a value-laden, sociotechnical discipline. Achieving an appropriate balance of AI regulation and innovation will require guidance and action from stakeholders across industry, government, academia, and the public.
Desarrollo responsable de la inteligencia artificial
Estos hallazgos destacan cómo los daños relacionados con la discriminación y la autonomía pueden intersectar y reforzarse entre sí en el ámbito de la contratación. Estos efectos tienen el potencial de causar un daño socializado amplio pero a menudo se omiten de las percepciones comunes de los impactos negativos de la IA. Además de fortalecer las protecciones legales contra la discriminación en el empleo, los formuladores de políticas deben considerar cómo la IA influye en la autonomía y priorizar el desarrollo de sistemas de IA que mejoren la agencia y la equidad en lugar de limitarlas.
