Does AI need a reality check? An interview with Gary Marcus

From the release of AI agents to claims that artificial general intelligence has (finally!) been achieved, 2025 will probably be another blockbuster year for AI. That sense of continuous progress is not shared by everyone, however. Generative AI, based on large language models (LLMs), struggles with reasoning, reliability and truthfulness. While progress has been made in those domains, sceptics argue that the limitations of LLMs will fundamentally restrict the future of AI.

In this episode, Alok Jha, The Economist’s science and technology editor, interviews Gary Marcus, one of modern AI’s most energetic critics. They discuss what to expect in 2025 and why Gary is pushing for researchers to work on a much wider range of scientific ideas (in other words, beyond deep learning) to enable AI to reach its full potential.