Here's the nuance of rules under capitalism that a robot can never understand:

The purpose of the rules is to get as many people to follow them as possible while you break as many as possible. That's how it "works." That's how the system picks winners and losers.

The whole point of putting an "AI" system in the loop of real decisions made in the real world is to remove accountability. "Oh we didn't do it, it was the emotionless perfectly logical robot that did it!" It's just a more advanced version of fiduciary duty allowing corporations to engage in socially harmful behavior.

When trained to maximize capital, they become capitalists.

wow could it be that markets encourage corruption? surely a human would never do such a thing

In a world where actual financial gangersters and criminals write an AI trading algorithm to speed up their criminal empire....

Cyborg gangster spongebob

Reminds me of how LLMs being used for programming might be told “write a class that passes these unit tests” and respond by just hardcoding the values being checked in the tests. Just absurd “solutions” to the tasks they’re asked to hallucinate about

We now examine who gains and who loses from AI collusion, and how this depends on the role of information-insensitive investors, captured by ξ, across three distinct trading environments. In case (i), with high ξ and low σu, the AI collusive equilibrium is driven by price-trigger strategies. Here, informed AI speculators primarily trade against information-insensitive investors, who absorb most of their order flow. In the simulation with ξ = 500 and σu = 10−1, each informed AI speculator earns an average profit of approximately 54, totaling a loss of about 108 for information-insensitive investors. Noise traders and market makers earn near-zero profits

of course, trading is zero sum so their collusion profits come from "information-insensitive investors".

They forgot to tell it to not alienate workers from the fruits of their labor