Today we have an Ask Me Anything episode that focuses on artificial intelligence. For the past year, AI has dominated the headlines. Perhaps because there’s so much media and public interest, as well as paranoia, about AI, listeners have been flooding our mailbox with AI-related questions.
So, today’s AMA is an exclusive focused on artificial intelligence. For our listeners who have been sending questions about ketamine; NASA’s Artemis mission to the Moon; or whether high meat consumption leads to dementia; fear not, because we will follow up today’s episode in a few weeks with a second round of AMA.
Dr. Ken Ford will be answering today’s AI questions. He has been working in AI since the 1980s and is a Fellow of the Association for the Advancement of Artificial Intelligence. As a result, he is well-positioned to help us today to separate real advances in AI from hype, panic and science fiction.
If you have questions for Ken and Dawn after listening to today’s episode, or any episode of STEM-Talk, email your questions to STEM-Talk producer Randy Hammer at rhammer.ihmc.org.
Show notes:
[00:03:15] Dawn opens our AMA episode on AI with a listener question for Ken on whether AI companies should be protected under section 230 of the U.S. Communications Decency Act.
[00:08:03] Following up on the previous question, a listener asks if Ken foresees a day where laws and courts will designate AI as a judicial person, in the same way that corporations have rights under the legal doctrine of corporate personhood.
[00:10:55] A listener asks Ken for his thoughts on a recent article published in the Atlantic titled “The Data-Center Panic is Overblown: Critics are Inflating the Costs.”
[00:16:09] A listener asks Ken a question regarding episode 171 of STEM-Talk, in which Ken discussed a June 2024 report by Leopold Aschenbrenner which predicted rates of growth in both compute resources and power requirements for AI data centers. The listener asks how well these predictions have held up in the following year.
[00:22:19] Another listener question about data centers asks Ken what his thoughts are on the notion of solving the electricity costs and cooling requirements for data centers by building them in orbit and powering them via solar energy.
[00:30:07] Moving on to the applications of AI, Peter Attia recently wrote about a study out of Harvard that suggests that today’s large language models perform extremely well on medical licensing exams and often arrive at the correct diagnosis yet still struggle with differential diagnosis. The listener asks what this means about the use of AI in medicine.
[00:36:53] A listener asks Ken about repeated public statements by Anthropic’s CEO that their AI models, and those expected in the future, are already powerful and dangerous enough to justify concern, while at the same time Anthropic and other leading AI companies are continuing to employ these systems and cautioning governments against regulation that could slow down innovation. The listener asks Ken how to resolve this contradiction.
[00:45:18] A listener writes that in a previous AMA, STEM-Talk episode 184, Ken objected to the use of the word hallucination to describe errors made by large-language models and asks Ken to expound on this.
[00:53:28] A listener asks Ken whether current AI models are evolving from being fundamentally prediction engines to being able to truly understand and reason, or if they are just becoming better and more convincing prediction engines.
[00:59:37] A listener writes to Ken, mentioning that a few years ago, a thousand technology experts and researchers signed a letter urging AI labs and executives to pause the development of highly advanced AI systems and tools. This letter, drafted through the not-for-profit Future of Life Institute, warned that AI developers were locked in an out-of-control race to develop and deploy ever more powerful digital minds that no one, not even their creators, can understand, predict or reliably control. The listener also states that they learned recently that ChatGPT 5 was programmed using an earlier version of ChatGPT. If the concerns of tech leaders are to be taken seriously, the listener wonders if it is a good idea to have earlier versions of ChatGPT programing later versions of ChatGPT?
[01:07:45] A listener asks Ken about the process of training smaller AI models from larger models, a process called distillation. The listener asks how this process works and if this is a way for the cost of training large models to benefit everyone.
[01:14:56] A listener asks Ken what it means for China’s Open Weight GLM 5.2 to be reported to have achieved near parity with Anthropic’s Mythos. On several cyber security and vulnerability discovery benchmarks, it seems alarming that GLM 5.2 is operating at roughly 1/6th the cost and remaining openly available for download and deployment.
[01:22:34] A listener writes that they recently heard a story about someone who asked an AI whether they should walk or drive to a car wash in order to get their car washed when the car wash was close by their starting location. The AI in this condition recommended that they walk instead of drive. The listener asks why the AI was unable to give the obviously correct answer to a simple common-sense question.
[01:29:05] For our final question in this AMA, a listener writes that, recently, nearly 200 researchers and economists signed a statement warning that as AI becomes more powerful it could take over a large share of human work and lead to widespread joblessness. The listener wants to know if Ken would have signed the statement and what he thinks of the statement’s three points: One: AI may become radically more powerful over the next 10 years. Two: This could drive an unprecedented transformation of our economy larger than the industrial revolution and over a vastly shorter timeframe, with large scale job displacement. Three: Economists, policy makers, and tech leaders must act now in order to understand the economic impacts of transformative AI and to build the incentives guardrails and institutions need to steer AI in a direction that compliments humans and benefits society.
Links:
Learn more about IHMC
STEM-Talk homepage
Ken Ford bio
Ken Ford Wikipedia page
Dawn Kernagis bio
