What if machines could reason like humans? We're racing toward that reality. It's called artificial general intelligence. The question is, can we build it safely?

In the second part of our conversation with Dr. Zico Kolter, head of Carnegie Mellon University's Machine Learning Department and OpenAI board member, where he chairs their Safety and Security Committee, we explore the critical challenges facing AI development today.

Dr. Kolter addresses deepfakes and the erosion of trust in media, explaining how AI accelerates existing problems while offering potential technological solutions. We examine privacy concerns, debunking common misconceptions about how chatbots use personal data. The discussion covers data scarcity, infrastructure challenges, and the massive energy demands of AI systems.

We also explore bias in AI models, the psychological impact of human-AI relationships on vulnerable populations, and the concept of artificial general intelligence (AGI). Dr. Kolter shares his optimistic yet cautious vision for the next five years, emphasizing the importance of building AI systems that safely serve humanity's best interests.

This is the Season Two finale of Where What If Becomes What's Next from Carnegie Mellon University.

What if AI is automating the one thing that's always made us human—Intelligence itself? And how do we ensure that it's developed safely?

In this first of a two-part series, we speak with Dr. Zico Kolter, head of Carnegie Mellon University's Machine Learning Department and newly appointed OpenAI board member, where he chairs their Safety and Security Committee, to explore the critical challenge of developing artificial intelligence safely.

Dr. Kolter discusses CMU's pioneering machine learning department and outlines four essential categories of AI safety concerns: immediate security threats like data exfiltration and prompt injection; societal impacts on jobs, economy, and mental health; catastrophic risks from malicious actors wielding AI-powered capabilities; and long-term scenarios of uncontrollable superintelligence.

Unlike previous technological revolutions that automated physical labor or computation, AI represents something unprecedented—the automation of intelligence itself. Dr. Kolter argues this fundamental difference demands collaborative oversight from industry, academia, and government to ensure AI serves humanity's best interests. The conversation emphasizes why getting AI safety right matters more than ever as we integrate thinking machines into our critical infrastructure.

What if AI could help design gene therapies so precise they eliminate chronic pain while leaving other senses completely intact? And what if the secret to treating Parkinson's Disease lies hidden in the genome of a mouse?

In this episode, we explore groundbreaking advances in genetic medicine with Andreas Pfenning, Associate Professor in Carnegie Mellon University's Computational Biology Department and School of Computer Science. Professor Pfenning discusses research in his experimental lab, and how AI is revolutionizing gene therapy by making treatments more targeted while reducing toxic side effects.

The conversation covers Pfenning's work developing precision gene therapies for chronic pain originating in the spine, where AI helps identify exactly which spinal cord cells transmit pain signals while preserving normal touch and movement functions. The episode also explores KGWAS (Knowledge Graph Genome-Wide Association Study), an AI-powered tool developed at CMU in partnership with other research institutions, that helps identify genetic problems that can cause rare diseases or disorders.

Pfenning shares insights from his collaboration on the ambitious Vertebrate Genomes Project, which aims to map genomes from more than 500 vertebrate species. This massive international effort is revealing surprising genetic similarities across species that could unlock new treatments for human diseases including Parkinson's, Alzheimer's, and other neurological disorders.

The discussion also highlights how machine learning is accelerating the path from laboratory discovery to clinical applications, with promising results already emerging in Parkinson's disease treatment trials.