posted: March 9, 2025
tl;dr: A cogent warning of the dangers of artificial intelligence in the realms of finance and national security...
Jim Rickards and I share a skeptical view of artificial intelligence (AI). We both think that AI should not be connected to critical systems, a point I made in Mind the gap. Yet as I concluded in that post people are going to do it anyway, in which case we will learn the hard way why this should not be done. Even when an airgap exists, Rickards points out that the airgap may consist of a person who learns to trust the output of the AI and follows its instructions mindlessly, without applying critical thought. In this case there is effectively no airgap, the AI is in control, and disaster may result.
Rickards opens MoneyGPT: AI and the Threat to the Global Economy with a hypothetical example of what can go wrong, and how it can go wrong, when AI is used in the realm of finance. Because of the decentralized nature of the Internet and the many people who use it, it does not require that a single rogue AI be in charge in order for catastrophe to result. Rickards has a strong understanding of complex systems theory and chaos theory, and shows how the various components of a complex system can interact to produce an outcome that is far beyond the norm. Chapter one of MoneyGPT would be an excellent hypothetical case study for budding investment professionals to study in business school.
Rickards and I are also aligned on the bailout of Silicon Valley Bank (SVB) depositors: neither of us thinks the government should have done so. In chapter two of MoneyGPT Rickards explains how severe the SVB failure was, and how the response to it violated established norms and set new precedents. Rickards points out how this introduces more risk into an already risky, complex system.
Rickards other passion besides finance is national security. He has done work for the CIA and participated in game-playing exercises in which participants respond to challenging hypothetical scenarios. Chapter four of MoneyGPT uses historical examples in addition to a fictional example to show how a greater reliance on AI can lead to a catastrophic outcome.
Rickards understands the flaws of AI. The AI’s training material, since it comes from the Internet, is often just plain wrong, fictional, biased, and filled with propaganda. This leads to AI output that mimics mainstream sources without necessarily being correct or truthful. AI systems will hallucinate, or fill in gaps in knowledge with purely invented material. As more AI output makes its way onto the Internet, AI ends up training on its own prior output, leading to a doom loop in which the AI gets dumber, not smarter. This problem may be unsolvable.
Rickards is a smart man. MoneyGPT contains references to Greek philosophers, ancient texts, and works of art that may be unfamiliar to some readers. I was not aware of everything that he referenced yet I still got the gist of his arguments. Some readers may be put off by these citations, and I wouldn’t call MoneyGPT an accessible book for the masses. I thoroughly enjoyed it, however, and am looking forward to Rickards’s next book, whatever the subject may be. As for MoneyGPT, I consumed the audio version of the book, which Rickards himself reads, and plan to listen to it again.