This eleven minute and twenty second TEDx talk explores the accelerating risks of artificial intelligence and the urgent need to redesign how it is built. Karen Hao opens with the paradox of technological progress: we are living through unprecedented change but remain profoundly unprepared for its consequences. Using compelling visuals of AI-generated deep fakes, she illustrates how powerful tools once limited to professional studios can now be created by a single person, amplifying both creativity and harm. She recounts the case of Indian journalist Rana Ayyub, whose face was deep-faked onto pornographic footage, triggering widespread threats and forcing her into hiding. This example underscores how easily AI innovations can become weapons against vulnerable communities.

Karen explains that unintended consequences are not unique to AI, yet historically societies have relied on a feedback loop of invention, impact, public debate, regulation, and redesign to make technologies safer. Cars, for instance, did not include seat belts until decades of accidents prompted change. But AI breaks this timeline. Modern algorithms scale globally in months, not decades, leaving society no time to evaluate risks before widespread harm occurs. Deep fakes, Facebook’s discriminatory ad-targeting algorithm, and mass deployment of facial recognition all exemplify systems that became ubiquitous before their implications were understood.

The core argument of the talk is clear: the way we develop AI is fundamentally broken because social expertise and technical expertise remain siloed. Regulators cannot keep pace, and developers often lack context about societal harm. Karen argues for a redesigned process where debate begins before deployment, with social scientists and technical builders collaborating from the outset. She stresses the need for interdisciplinary teams and, ultimately, interdisciplinary people capable of understanding both technology and society.

The interview begins with Karen Hao reflecting on her background at MIT, where she studied mechanical engineering alongside many people who now lead or build modern AI systems. This proximity demystified the industry for her and taught her that technology is not magic but the cumulative result of human choices, biases and blind spots. She explains that concentrating the power to shape society into the hands of a small, homogeneous group of technologists is structurally unsound, because their worldviews inevitably limit what they imagine, build, and safeguard against. This sets the tone for a conversation that scrutinizes the ideologies, business incentives and geopolitical consequences behind the current AI boom.

Aaron Bastani introduces AI as the focal point of contemporary political and corporate discourse, often framed as a revolutionary force akin to the steam engine. While the public debate gets stuck between techno-optimism and doomerism, Karen positions herself in the present, grounded in empirical reporting rather than speculation. Her new book, Empire of AI, draws on more than 300 interviews and unprecedented access to insiders at OpenAI and other companies. Instead of portraying AI leaders as visionaries or villains, she reconstructs their motivations, contradictions and myth-making tendencies through firsthand accounts.

Karen explains why defining AI is difficult and historically contingent. The term was coined in 1956 by John McCarthy largely as a marketing tool to secure funding, and much of the field’s language has continued to evolve via rebranding rather than conceptual clarity. Today “AI” most commonly refers to deep learning systems trained on vast datasets to compute statistical patterns that allow them to generate text, images or predictions. She outlines the nested relationship between AI, machine learning and deep learning, emphasizing that people often conflate them in ways that obscure the actual technologies.