“If insolvency is not transparent or well understood, and if illiquidity is backstopped by the Federal Reserve, then why do bank runs commence? The answer is psychology. Some customers or counterparties come to believe a bank will not repay them so they pull their money out or close transactions as quickly as possible. They are not reassured by ... press releases or positive comments by management. Word spreads, the withdrawals accelerate, and within days, sometimes hours, the bank closes its doors. From there it's an open issue whether the lost confidence spreads to other banks, in a process called contagion. No amount of capital or comment can stop a bank panic; it has a life of its own.
...
Enter AI. The next bank run may be triggered not by human panic but by AI imitating human panic. An AI bank analysis program with deeply layered neural networks and machine learning capability (perhaps complimented by a GPT capacity to speak with human analysts) Could read millions of pages of financial data on thousands of individual banks, far more than any team of human analysts could review. It's training set of materials provides familiarity with the dynamics of bank runs, basically an emerging property of a complex dynamic system, along with historical examples, worst case scenarios, and defensive moves. Events like the gold corner of 1869, the panic of 1907, the Great Depression of the 1930s, and the S&L crisis of the 1980s would all seem as fresh as today's news. This system would reach the same conclusion as a human analyst — move first, get your money out fast, don't be the last in line.
The true danger is not that the machine thinks like a human — it's supposed to. The danger is that it can act faster and communicate with other machines.”