Arati Prabhakar on AI Policy, Public Trust, and Human Choices

Shira Lazar introduces Arati Prabhakar, a former White House science and technology adviser who helped shape the Biden administration’s 2023 AI executive order. This conversation was published in March 2025; its policy descriptions and political assessments reflect that moment.

Prabhakar says she has returned home to Palo Alto after leaving the White House on January 20. She reflects on work involving AI, the cancer moonshot, climate, and the innovation system, while describing the political situation at the time as tumultuous.

Asked where the country stood on AI governance, Prabhakar says she felt good about the direction set but considered it the beginning of a long journey. She recalls becoming science adviser shortly before ChatGPT’s release in 2022.

She says AI was already influencing feeds, search, advertising, prices, and aspects of healthcare before generative AI drew wider attention. In her account, the administration sought to manage risks and pursue benefits together.

Prabhakar describes an executive order directing action within existing law, alongside work with Congress, other countries, and industry. She emphasizes that this was a beginning rather than a completed solution.

As an example, she says consumers denied credit should receive an explanation rather than an opaque decision. She also expresses concern about the future of the Consumer Financial Protection Bureau under the incoming administration.

In healthcare, Prabhakar argues that AI should improve diagnosis and treatment without reproducing historical bias. She cites disparities affecting Black women and describes a health-department rule intended to address discrimination; these are her account of policy and conditions at the time.

She also discusses fraud involving impersonation of businesses or government agencies. Her broader argument is that trust and protections are necessary if AI is to help tackle difficult problems.

Lazar asks how much influence leaders such as Sam Altman had on policy. Prabhakar says AI is larger than any one executive: her team heard from major companies but also sought out startups, investors, universities, and national laboratories.

She says those conversations helped identify risks as well as opportunities. Her answer to the question of whether to regulate or innovate is both: manage harms and use technology to solve meaningful human problems.

The speakers compare AI’s arrival with the social-media era. Prabhakar recalls early optimism about connection, followed by greater attention to bullying, abuse, and synthetic intimate images.

Prabhakar says the 2023 executive order had been rescinded and assesses the new administration’s approach as uncertain. She criticizes federal staffing and funding actions, arguing that they could weaken government capacity; this is her political assessment in early 2025.

Asked about a major announced private AI investment plan, she notes the scale of data-center spending. She then points to DeepSeek as a reminder, in her view, that research and algorithmic approaches can matter alongside brute-force computing.

Lazar raises questions about claims, competition, and privacy. Prabhakar argues that researchers in many countries are exploring different approaches to improving capability and that the landscape is broader than the most prominent firms.

Prabhakar contrasts access to proprietary hosted models with models made available for others to run and build upon. She sees shared research as a potential source of innovation, though model openness can vary in practice.

She distinguishes using a model hosted by a separate provider from sending information directly to the model developer’s own service. Her example illustrates why people should consider which service receives their data, rather than assuming that model origin alone determines data handling.

The discussion moves to TikTok. Prabhakar describes a difficult balance between free expression and concerns about access to data under Chinese law. Her legal and political descriptions are part of the historical interview, not a statement of current law.

Lazar asks about privacy risks from American companies as well. Prabhakar acknowledges company data access as a real privacy issue while drawing distinctions between commercial collection and government access.

Both speakers express concern about the political and technology environment. Prabhakar says practical approaches require cooperation with industry but also public action; she criticizes the administration’s direction and worries about the country’s research capacity and international leadership.

On deepfakes, Prabhakar says nonconsensual synthetic intimate images were among the immediate harms her team anticipated. She describes work with industry, including payment processors enforcing their terms against abusive services.

The speakers also discuss a fabricated video involving public figures. They worry that people may struggle to know what is real. Prabhakar says distorted media can weaken a shared basis of fact and leave people in separate fictional realities.

Asked about federal, state, and company responsibilities, Prabhakar expects states to play an important role when federal action is uncertain. She acknowledges that different state approaches can create complexity for companies.

She encourages people to contact elected representatives and support organizations working on issues they value. She frames these ordinary civic actions as a way to move beyond helplessness and influence the direction of AI.

Prabhakar turns to potential benefits, imagining AI helping researchers develop medicines faster for diseases that lack effective treatments. These are aspirations about future research, not a claim that AI can already deliver a particular treatment.

She argues that ambitious advances need universities, companies, and public institutions, including regulators who can protect people while enabling progress. She also hopes AI can help with education and weather forecasting.

Asked which countries are doing well, Prabhakar recognizes research talent in different regions. She criticizes authoritarian surveillance in China and regards Europe as comparatively cautious about regulation, while emphasizing what she sees as American strengths.

Lazar asks how people can feel more control. Prabhakar says people, not technology alone, decide which systems to build, how to use them, and what society will tolerate.

For individuals, she suggests experimenting with chatbots for brainstorming while checking references and grounding claims in evidence. She also recommends noticing where AI shapes decisions, advertisements, feeds, or services and considering what a system may know about its users.

Prabhakar says people can stop using tools they distrust, complain to companies, and communicate with policymakers. She points to an open policy consultation at the time; its availability should not be assumed today.

Lazar asks about job losses and universal basic income. Prabhakar is skeptical of a future in which no one works and instead wants workers to use AI to become more productive and earn more, while acknowledging the danger of rapid displacement.

She describes hearing about workers whose keystrokes or customer conversations are monitored and corrected by AI systems. In her view, maximizing efficiency without preserving the human contribution can make work worse.

Lazar explains that she uses AI as a thought partner and to help prepare messages and presentations for her small business. She says the tools help her move faster while she remains responsible for the ideas.

Both discuss the difficulty of transitions. Prabhakar says earlier waves of automation caused painful disruption and argues for investment in training and pathways to better work rather than assuming a positive outcome will happen automatically.

On content moderation, Prabhakar says human-only and machine-only approaches can each fail in different ways. She is more optimistic when people direct automated tools and bring human judgment to difficult decisions.

She reflects on the difficulty for companies to recognize the harms that can accompany products they believe are beneficial. Ultimately, she says, these are societal problems as well as company problems.

Asked to look ahead, Prabhakar voices concern about the administration’s direction and its effect on national capacity. She calls for a correction and urges people to learn about issues, contact representatives, and support organizations they believe in.

Prabhakar says she remains motivated by work that makes the future better. She hopes the country can combine its capabilities and use trustworthy AI to pursue ambitious goals.

For a practical example, she describes using Perplexity to locate an obscure historical account about researchers observing Sputnik, a story she connects to the early history of GPS. She says the tool helped her find source material from a partial memory.

Lazar thanks Prabhakar for the conversation and points listeners toward her work. The episode closes with sponsor and production acknowledgments.