Descriptions:
This episode of the Cognitive Revolution’s “AI in the AM” weekly highlights series compiles three conversations from the frontier of AI research and practice. The episode opens with Cameron Berg, founder of Reciprocal Research, presenting a “dimmer switch” model of AI consciousness — arguing that sentience exists on a continuum rather than as a binary state. Berg describes an experimental program that deploys large language models at scale to evaluate whether current AI architectures satisfy the predictions of major consciousness theories, including Global Workspace Theory, turning what was previously an expert-bottlenecked process into a scalable empirical research pipeline.
The second segment features researcher David Duvenaud on what he calls “gradual disempowerment” — the concern that humanity may cede meaningful control to AI systems through a long series of individually reasonable-seeming decisions rather than any single dramatic moment. Duvenaud challenges successionist arguments that treat any conscious successor civilization as acceptable, contending that the values and character of future dominant systems matter as much as the fact of consciousness itself. He also describes an active technical project using AI to perform historical super-forecasting: training simulations on past data and testing whether they can predict subsequent decades, as a calibration method for forward-looking AI forecasting.
The episode closes with swyx’s observations on where AI engineering alpha is being found in 2026, covering shifts in how development teams are structured and where productivity gains are and are not materializing. Host Nathan Labenz frames all three conversations around a single orienting question: how much do humans actually understand about what is inside these systems, and where the trajectory is heading?
📺 Source: Cognitive Revolution “How AI Changes Everything” · Published June 27, 2026
🏷️ Format: Podcast
