Dr. Bart Kosko is a bestselling author and professor of electrical and computer engineering, and law, at the University of Southern California. He is an award-winning pioneer and author in the machine-learning fields of artificial intelligence, neural networks, and fuzzy logic, as well as the recent winner of the International Neural Network Society (INNS) Hebb Award for neural learning. He joined Ian Punnett (Twitter) to discuss the latest developments in AI and neural learning, and a new edition of his classic work, Fuzzy Thinking.

Kosko reported on AI hallucinations, which is when an AI-powered language model, such as ChatGPT, generates false information. According to Kosko, who has tested and interrogated many large language models (LLM), these systems are rife with error and are unable to determine if they have answered questions accurately. "It does not tell you the confidence it has when it answers a question and it really can't," Kosko said, noting AI-powered language models do not know how because the people who trained them do not know how to teach about hallucinatory errors. "Relying on this, deferring to it... is very dangerous," he warned.

Kosko described LLM as a "black box" AI system using an old-fashioned neural network, and cautioned about racing ahead without thinking through the implications, something he called "digital delirium." He spoke about the importance of having a backup or restore point against which a check could be made for errors. "I don't see how you could correct a large-scale AI corruption without a benchmark," he suggested. Since it is a black box there could potentially be issues with bad data fed into the system. "We don't know what it's forgotten when we teach it something new... it's just going to get a lot worse as we go," Kosko said. He also revealed how "hints" could be introduced into the programming that would have a subtle effect on the output and would be difficult to detect.

Police Intuition & Gilgo Beach Murder Case

In the first hour, paranormal investigator Greg Lawson provided his unique insights on the Gilgo Beach murder case and the role of intuition in police work. "We pick up these little things that we feed in on that just aren't quite right... a little red flag that will lead you someplace else, and you can't always explain exactly why you feel that little red flag," Lawson explained. We now know the disgraced Suffolk County police chief botched the Gilgo Beach murder case, but why did no one see it at the time? Lawson suggested police officers sometimes suppress the instinct that leads to intuition in order to prevent emotions from crippling them in heated situations. He also reported on his investigation into Jack the Ripper, pointing to a series of murders in Austin, Texas from 1884-1886. The chief suspect fled to London where murders attributed to Jack the Ripper began in 1888.