Episode Details

For the first time ever, parents going through IVF can use whole genome sequencing to screen their embryos for hundreds of conditions. Harness the power of genetics to keep your family safe, with Orchid. Check them out at orchidhealth.com.

On this episode of Unsupervised Learning Razib  talks to Nick Cassimatis, erstwhile artificial intelligence researcher and currently an entrepreneur. Cassimatis has undergraduate and doctoral degrees in cognitive and computer science from Massachusetts Institute of Technology, and a master’s degree in child psychology from Stanford. He studied for his Ph.D. under Marvin Minsky, arguably the most prominent and influential artificial intelligence researcher of the second half of the 20th century. Later, Cassimatis was a professor at Rensselaer Polytechnic Institute, founder of a successful startup, and a researcher at Yahoo and Samsung.

Because of the explosion of large language models as implemented in OpenAI’s ChatGPT and Google’s Gemini, we are now living through an artificial intelligence “hype cycle.” But Cassimatis observes that this is not the first time this has occurred. The 1960’s saw enthusiasm triggered by the ELIZA therapist chatbot. Then, in the 1990s another wave of interest crested because of the mastery of chess by Deep Blue. Finally, there was a boom of excitement around artificial intelligence after Watson’s victory in Jeopardy in the early 2010s. But these hype cycles also have their equivalent troughs; Cassimatis recounts that when he went to study artificial intelligence in the early 2000s, many people discouraged him because the field’s allure had cooled considerably. And yet, under Minsky he developed an interest in how computers could learn, writing papers like A cognitive substrate for achieve human-level intelligence. This background makes Cassimatis a particularly well-informed and trenchant observer and analyst of the current arguments about the possible emergence of artificial general intelligence in the next decade, and what it means for the future of humanity.

But first, Cassimatis and Razib step back and address some basics. What is machine learning? How does this relate to deep learning and natural language processing? What are transformers and what is a neural network? These are terms that are thrown around casually in the technology press, but these concepts emerge from over fifty years of research in computer science. With those preliminaries out of the way, Razib probes Cassimatis’s opinions about the past and future of large language model-driven artificial intelligence, and the probability of Ray Kurzweil’s “intelli