When AI goes wrong (with all your money)
Exciting stories abound of the potential for AI in finance, well here’s the other side of that trade…
A couple of years ago we sunk our life savings into developing the first, exclusively, AI-powered Asset Manager.
Working with a crack (cheap) team of outsourced Russian engineers and data scientists we quickly realised we’d developed a market-beating automated investment manager. A true robo-advisor. We were over the moon.
However, something had gone terribly wrong in the configuration of the algorithm.
At first we didn’t notice. Robo allocated positions into a diversified portfolio and we were on track for our return profile. But then we started to want to make changes.
Our Russian engineers had built a chat interface to communication and send instructions to Robo. Our first use of this interface was in February 2022 to ask whether Robo thought we should rebalance out of long dated treasuries. It was met with the following response:
Already done that you dumb human fucktardsThinking this was perhaps a simple anomaly in the natural language processing we tried again over the next few weeks. However, the responses we received were some of the most insolent, impertinent and downright rude messages we’ve ever seen. In 30 years of investment management no one had ever talked to us like this before. Imagine having to deal with a hungover teenager because he’s managing all your wealth. That’s what it was like.
So we quickly moved to disconnect Robo from trading interface and get our money out. But in March 2022 when we attempted to contact our Russian outsourcing firm our calls and emails went unanswered and eventually bounced back.
At that point the full horror of our predicament dawned on us: We were stuck having to deal with Robo to manage the portfolio until we could figure out a way to disconnect him.
Welcome to our hell.
Like reading finance professionals being abused by a robot? Never miss an update by subscribing below

