Episode Details

In this episode of The Neil Haley Show, Neil “The Media Giant” Haley welcomes AI pioneer and author William Meiselto discuss his book, The Limits of AI: The True Role of Computers in Our Future. Meisel brings decades of experience in neural networks, pattern recognition, speech recognition, and computer science to a conversation about what artificial intelligence actually is, where today’s systems succeed, and why he believes the biggest AI breakthroughs may not come from simply building larger language models.Meisel explains that the term artificial intelligence dates back to the 1950s, but today’s AI boom is largely powered by machine learning and deep neural networks. These systems learn patterns from enormous amounts of human-generated data and have produced impressive results in areas including image recognition, speech recognition, medical analysis, protein research, and large language models such as ChatGPT.But impressive does not mean unlimited.That distinction is at the heart of Meisel’s book.Meisel’s own history with AI goes back decades. His Ph.D. dissertation focused on neural networks, and he later wrote about computer pattern recognition while teaching at USC. He also worked on defense applications involving large amounts of data and eventually founded a speech-recognition company.Many of his early ideas worked technically, but computing power was too slow or expensive for widespread adoption.Today, the hardware finally exists.The question, Meisel says, is whether simply making neural networks larger will continue producing dramatically better intelligence.He is skeptical.Neural networks are statistical models trained on data. As the models become larger and contain more parameters, they require enormous amounts of diverse, high-quality information. Meisel argues that obtaining enough truly independent and reliable data could become increasingly difficult.That problem becomes even more complicated when AI-generated material begins appearing throughout the internet.If future AI systems train heavily on content that was itself created by previous AI systems, Meisel worries that models may increasingly recycle summaries of existing information rather than learning genuinely new knowledge.Neil connects that argument directly to his own experience using AI agents.He explains that the strongest results come when he gives AI his own business information, knowledge documents, workflows, instructions, and context rather than expecting a model to operate independently.Both agree on a major principle: good input still matters.Neil describes AI as something that can accelerate his work, analyze information, organize projects, function as an executive assistant, and help manage business operations—but only when a human remains actively involved.The conversation then explores whether specialized AI systems may ultimately be more valuable than enormous general-purpose models.Meisel points to applications where neural networks are trained on focused, high-quality datasets, including medical research. A specialized system working with a well-defined body of information may achieve far greater reliability than a model expected to answer virtually any question imaginable.He also discusses unsupervised learning, where computers identify patterns and clusters without being provided a predetermined label for every example.But one of the most exciting areas for both Neil and Meisel is conversational AI.Meisel spent years working on speech recognition and believes increasingly capable digital assistants could become one of computing’s most useful future applications.Neil strongly agrees.He describes using conversational AI as an ongoing business partner—talking through ideas, giving instructions, brainstorming, reviewing information, and refining decisions through dialogue rather than simply typing one prompt and accepting the first response.That human-AI conversation may be where artificial intelligence becomes most practical.Instead o