November 02, 2026In Math We TrustHave you ever checked the wrong box in a wire transfer?
Citibank once lost nearly a billion dollars by messing up a transfer form. On August 11, 2020, Citibank was supposed to disburse $7.8 million of loan payments for Revlon, but instead it sent about $900 million of its own money to hundreds of lenders by mistake. It was because three checkboxes needed to be checked on the forms, and a subcontractor had checked only one. Two supervisors approved it anyway. The form was so confusing that all three believed it was right. Citibank asked for the money back the next day. But ten investment firms did not return it. They knew Revlon was on its way to bankruptcy and not good for the money, and they seemed to have lucked out on a chance to recover their principal. Under New York law, a judge ruled, a creditor may keep money it is actually owed if it had no reason to suspect a mistake, so they could keep it! But in September 2022 the Second Circuit reversed. A reasonable lender, it held, would have known that an unannounced early payoff by a near-bankrupt company, at triple the market price, was an error. The money, frozen during the appeal, went back to Citibank. The argument that cryptocurrency beats fiat currency in security has always struck me as stupid. The beauty of banking is that it is about human trust. In regular banking, mistakes are common and are routinely reversed. I am not sure who really believes we are best served by a cryptographic system where any tiny error is fully irreversible and might mean that you lose billions of dollars. We humans are imperfect, so we need a banking system that is robust to our foibles. I was reminded of this again when on October 29, Binance, totally unsurprisingly, lost more than $6 billion to hackers. Yes, in math we trust. But we must recognize that humans are not creatures of mathematical precision. In the era of superhuman AI and lightning-fast code, we will increasingly grapple with the issue that perfection is not actually good enough for human use. A formally verified program can still do exactly the wrong thing if no human can understand what it does. We need our systems to be understandable. We must recognize and accommodate human limitations. Earlier: In Code We Trust Posted by David at November 2, 2026 07:29 AMComments
Hi Prof, I know from a high level perspective that languages like Lean are used for Formal proofs. Formal proofs are completely precise and deterministic, but quantum physics says the real world is probabilistic at its foundation implying that if something could end up happening irrespective of the formality steps we take to prevent something from happening, does not gaurantee on it not happening. Does that mean the guarantees from methods like Lean can never fully carry over to real-world systems? Posted by: Soham at November 4, 2026 10:40 PMPost a comment
|
| Copyright 2026 © David Bau. All Rights Reserved. |