We had the pleasure of interviewing Professor Stuart Russell in the latest episode of Silicon Valley Tech Talks.

Professor Stuart is a Distinguished Professor of Computer Science at UC Berkeley—his contributions to AI span across more than four decades. He has co-authored multiple AI books that are being taught at hundreds of prestigious universities around the world. Professor Stuart is highly regarded as one of the leading voices on AI worldwide. He has also served as co-chair of the World Economic Forum’s AI and Robotics Council.

During our chat, Professor Stuart first recounts how his interest in AI began in the 1970s, stemming from early programming and chess experiments, which eventually led him to Stanford to study how intelligence works. He explains how early AI was an engineering discipline built on logic and probability.

He identifies two major misconceptions today:

Language models are not merely “next-word predictors” — they imitate humans and can absorb unsafe human-like motivations.

Deep learning is fundamentally limited — scaling up models only hides the fact that they cannot truly learn or reason efficiently.

Professor Stuart warns that future evolution of more intelligent AI systems could lead to existential risks, as superhuman AI might escape human control or cause systemic collapse. He urges governments to set red lines (e.g., AI must not replicate or surveil on its own) and require extreme safety standards before deployment, much like nuclear regulations.

Finally, he describes his work at the Center for Human-Compatible AI, which focuses on building “assistance game” systems — AIs that aim to further human interests while remaining uncertain about what those interests are. This makes them cautious, deferential, and willing to be switched off — the foundation, he believes, for truly safe and beneficial AI.