A.I. Policy and Public Perception

Brundage describes his report, “The Malicious Use of Artificial Intelligence,” as a survey of deliberate harms such as fake news, drone attacks, and offensive cybersecurity. He argues that AI needs norms similar to responsible disclosure in computer security, including for adversarial examples. Hwang focuses on how AI could make disinformation more believable and easier to produce.

Hwang says it is difficult to predict which harmful uses will become common first. Brundage notes that physical applications can be costlier and harder to scale than digital ones, while autonomous drones show progress in real-world navigation. They discuss how a major self-driving incident or an AI-enabled cyberattack could shape public reaction, which may overreact or become desensitized.

Hwang points out that people may not think of recommendation feeds as AI, while they may label robots as AI even when they do not use machine learning. He highlights international competition and interpretability as prominent concerns. Brundage adds robustness, fairness, accountability, and transparency, and describes healthcare applications such as image analysis and hospital readmission prediction, many of which were still pilots or offline studies.

Hwang defines interpretability as understanding why a machine learning system produced an output; Brundage distinguishes explanations for technical teams from those useful to domain experts such as doctors. They compare Europe’s broader rules for automated decisions with the U.S. tendency to regulate by domain. Brundage says AI policy needs scenario planning and closer collaboration between technical researchers and policy specialists.

Hwang describes a divide between people focused on current harms and those concerned about advanced AI and existential risk. Brundage argues that fairness, accountability, transparency, and security matter across both time horizons, and that work on present-day problems can build useful governance capacity. They also discuss responsible disclosure in specific high-risk cases while generally supporting open research.