Mhairi Aitken Generative AI Risks
What are the risks of generative AI? belongs in the index because it is an unusually useful public lecture on AI risk as lived institutional impact rather than spectacle. Aitken defines generative AI broadly across text, images, video, and synthetic voices, then uses concrete cases to show how risk moves through systems: unsafe outputs, rushed deployment, unreliable AI detectors, uncompensated creative work, companion-style intimacy, synthetic media, hidden data-labelling labor, environmental cost, and children's rights. The lecture is strongest where it asks who is responsible across the chain of model provider, app developer, deploying institution, user, and regulator.
The strongest Spiralist relevance is the shift from model capability to social setting. Aitken's examples are not only about whether a model can generate text or media; they are about what happens when generated output enters classrooms, dating apps, companion systems, election discourse, creative labor markets, supply chains, and childhood. That belongs beside AI Literacy, AI Hallucinations, The AI Detector Becomes the Discipline Machine, AI Companions, Synthetic Media and Deepfakes, Ghost Work and the Hidden Labor of AI, AI Energy and Grid Load, and Youth AI Companion Safeguard. The governance question is whether affected communities can shape systems before harm is normalized as ordinary interface behavior.
External sources support the lecture's frame while narrowing some claims. The Alan Turing Institute's event page identifies Aitken as an Ethics Fellow in its Public Policy Programme and frames the lecture around online and offline safety risks. Stanford HAI's coverage of AI detector bias against non-native English writers supports the education example and explains the role of perplexity-based detection. TIME's reporting on Kenyan data-labeling workers supports the hidden-labor example, and UC Riverside's coverage of AI water consumption supports the claim that generative AI has material infrastructure costs. NIST's synthetic-content transparency report supports the lecture's warning that deepfakes and generated media require layered provenance, labeling, detection, testing, and auditing rather than a single fix.
Uncertainty should stay visible. The lecture was uploaded in November 2023, so some product names, usage numbers, and regulatory details have changed. It is not a technical audit of any one model, and some illustrative estimates, especially infrastructure estimates, depend on assumptions that providers do not fully disclose. Treat the video as a strong evergreen map of current-impact reasoning: it helps distinguish evidence-backed, already-experienced harms from speculative AI panic, while still leaving room for new risks as generative systems become more embedded in institutions.