Artificial intelligence (AI) has caught the imagination of some of the grand old men of Wall Street. JPMorgan chairman Jamie Dimon devoted a section of his annual letter to shareholders in April to AI and the bank’s progress in its use.
“AI and the raw material that feeds it, data, will be critical to our company’s future success – the importance of implementing new technologies simply cannot be overstated,” Dimon said, setting a challenge to people who think that it probably can be overstated and that we might even be approaching peak overstatement.
Dimon noted that JPMorgan employs over 900 data scientists to create AI and machine-learning models and 600 engineers to write the code that puts the models into action.
Soon after Dimon’s letter was published, JPMorgan unveiled one of the fruits of this AI drive in the form of a ChatGPT-based language model to interpret the messaging produced by central bankers.
Pranked
The timing of the launch of the model – which produces a Hawk-Dove Score, rating central bankers’ comments – was unfortunate, as it coincided with news that US Federal Reserve chair Jerome Powell had been pranked into discussing monetary policy in an online video with a Russian provocateur who pretended to be Ukraine’s president Volodymyr Zelenskyy.
Powell joined his fellow central bank head Christine Lagarde of the ECB in falling for the Fake Zelenskyy prank, which highlights an obvious problem with overstating the potential benefits of AI.
Impressive as some of the models may seem, they remain susceptible to a principle that is as old as computing: garbage in, garbage out.
One early conclusion is that we should be wary of boasts about AI by financial industry executives
Online conspiracists tried to twist the prank videos with Powell and Lagarde to demonstrate that they had revealed secret agendas when they thought they were talking to Zelenskyy. In fact, they simply repeated standard talking points, and demonstrated the wary patience that central banking technocrats learn to deploy when interacting with politicians.
Fears about AI systems taking over from their human overlords have been revived by the current competition between technology companies to market language models.
Should we worry that JPMorgan’s central bank model could learn to manipulate the language of central bankers to fit its own conclusions about future interest rate direction, or even start making its own fake videos?
AI uh oh
Experts in the field may want to join JPMorgan data scientists in attending a conference in May to discuss 'Certifiably robust policy learning against adversarial multi-agent communication', which is hosted by the International Conference on Learning Representations.
People who are pressed for time can instead deploy Euromoney’s distinctly obvious heuristics (D’Oh!) model to decide if something looks suspicious.
This model is not infallible, but it at least reaches a quick, actionable output.
One early conclusion is that we should be wary of boasts about AI by financial industry executives. Dimon isn’t the only grand old man of Wall Street who has become infatuated with AI. Blackstone’s chairman Stephen Schwarzman and his number two Jon Gray made their own pitch for AI upside in the firm’s recent quarterly earnings call.
“If you think about investing as pattern recognition, connecting dots, I don't think there's any firm in the world who gets to see more dots than we do,” said Gray, noting that Blackstone has more than 40 data scientists.
That might seem to put Blackstone at a disadvantage relative to JPMorgan, with its roster of over 900. But perhaps Blackstone’s in-house humans are simply better at producing models to connect dots. Only time will tell.
While we wait, we should heed the wisdom of Schwarzman.
“When you accumulate large databases, sometimes it's hard to figure out what's going on with them,” he accurately observed.
