“AI systems require massive amounts of data to work. For recommending an IRA allocation, this system would need plenty of examples of good and bad allocations for every potential client circumstance. For simple models using just age, income, and risk tolerance, this data might be available. But there is simply no data for the different economic and geopolitical environments that are possible. Humans rely on inductive cause-and-effect thinking to provide suitable recommendations for complex scenarios.
As for language models like ChatGPT or GPT-4, there is probably even less data. These models are built using publicly-available text, and I doubt that exists for every type of client. Because finance is rooted in numbers, it would be much more efficient to provide an AI system with specific numerical data to train than text. A language model doesn’t make sense for portfolio decisions. Where a language model might be extremely useful is to help clients understand difficult financial topics. For example, it could explain the difference between a Roth IRA and Traditional IRA.
AI might be described like statistics on steroids, and statistical theory has been applied to the markets for decades. None of the legendary investors like Warren Buffett or Peter Lynch attribute statistics their success.”