Inside the White House’s Push for Responsible AI
Artificial intelligence has emerged as one of the most consequential technologies of the era, carrying the potential to accelerate solutions to global challenges while posing risks that demand urgent governance. Arati Prabhakar, Director of the White House Office of Science and Technology Policy, underscored this duality in a recent discussion with Carnegie Endowment for International Peace President Tino Cuéllar, framing AI as a force that can either advance human aspirations or undermine them depending on human choices.
President Biden’s recent executive order marks what Prabhakar described as “the most significant action taken anywhere in the world on AI.” It sets new safety and security standards, strengthens privacy protections, addresses discrimination, supports workers, and aims to bolster U.S. leadership internationally. This order builds on the administration’s broader AI strategy, which began with the October 2022 “Blueprint for an AI Bill of Rights,” emphasizing safety, privacy, freedom from discrimination, and transparency.
Prabhakar noted that generative AI’s rapid rise—capable of producing text, images, audio, and video—has captured public attention, but it is just one branch of machine learning. Models trained on diverse datasets, from sensor readings to financial records, are now applied across industries. In automotive engineering, AI can enhance vehicle safety systems but also risk fostering driver complacency. In biotechnology, it can accelerate drug discovery or be misused to design biological threats. In finance and hiring, it can streamline processes or embed bias at scale.
The unpredictability of AI’s evolution makes governance challenging. Even with guardrails, users have found ways to bypass protections in both proprietary and open-source systems. As Prabhakar explained, “This period in AI technology is like medicines before clinical trials.” Robust tools for evaluating safety, trustworthiness, and effectiveness remain in their infancy.
One striking trend is the declining cost of building capable AI models. While frontier systems like GPT-4 required training budgets exceeding $100 million, advances in algorithms and engineering have slashed costs. MosaicML, for example, trained a GPT-3–level model for $450,000. Open-sourcing by companies such as Meta has further lowered barriers, enabling domain-specific models to be built for as little as $100. This democratization expands opportunities for innovation but also accelerates proliferation of potentially harmful capabilities.
Risks vary in timescale. Some, like AI-enabled cyberattacks on critical infrastructure or biosecurity breaches, could unfold within weeks. Others, such as erosion of trust through disinformation, may take years. Bias in automated decision-making can shape inequities over generations. President Biden’s warning resonates here: “America is an idea…of hope and opportunity…If we allow AI to be used to automate and magnify discrimination at scale, it can be catastrophic to the very idea of America.”
The administration’s approach blends domestic action with international coordination. Voluntary safety and security commitments from 15 leading AI companies followed a White House summit where Vice President Harris told executives they have “an ethical, moral, and legal responsibility to ensure the safety and security” of their products. The U.S. is also engaging through the G7 Hiroshima AI Process, the UK’s AI Safety Summit, and discussions at the United Nations.
Domestically, federal agencies are moving to integrate AI responsibly. Draft policy from the Office of Management and Budget would require agencies to assess equity, fairness, and privacy impacts before deploying AI, particularly in sensitive contexts like social safety net eligibility. Safety measures include screening protocols for synthetic biology research and monitoring high-compute AI training runs. Enforcement agencies have reminded industries that AI use does not exempt them from existing laws.
The executive order also directs efforts to protect workers, develop testing standards through the National Institute of Standards and Technology, and expand public-sector AI expertise. Initiatives like the National AI Research Resource pilot aim to provide compute and data infrastructure for research, while DARPA’s AI Cyber Challenge offers up to $20 million in prizes for AI-driven cybersecurity breakthroughs.
For engineers and technologists, the applications under exploration are notable: AI models predicting earthquakes days in advance, forecasting extreme weather at kilometer resolution, decoding neural signals to restore movement in paralysis, and designing novel materials and antibiotics. These examples illustrate AI’s potential to intersect with aerospace, robotics, advanced manufacturing, and other innovation-driven fields.
Prabhakar emphasized that achieving the promise of AI while containing its risks will require sustained effort across government, industry, academia, and civil society. As she put it, “We stand at an inflection point in history.” The trajectory of AI will be determined not by the technology alone, but by the collective choices made about how it is built, deployed, and governed.
