Angshuman Guha

This article presents an extended dialogue between Angshuman Guha, a researcher with hands-on experience in neural networks dating back to 1993, and Google’s large language model (LLM), Gemini. It explores a set of closely connected questions at the intersection of AI and cognitive science. Are modern LLMs genuine reasoning systems, or are they sophisticated Stochastic Parrots that recombine language without understanding? Do the so-called “emergent abilities” of large models reflect a real shift in machine capability, or are they artifacts of how we measure performance? And perhaps most fundamentally: can a system trained entirely on human-generated text be said to produce anything like “original thought”?