The Finance Robots are Coming
Â
In 1987, Thomas Peterffy became the first person to connect a trading algorithm directly to a Nasdaq terminal, allowing his models to generate and execute orders without human intervention.Â
Â
The resulting surge in trading activity soon elicited a visit from a Nasdaq official, who deemed Peterffyâs arrangement an unfair advantage.Â
Â
To level the playing field, Nasdaq instituted a rule that all ensuing orders had to be entered into the exchangeâs trading terminals via keyboards.
Â
Peterffy, who later went on to found Interactive Brokers, complied by building a robot with nimble rubber fingers that speedily typed his algo-generated orders into the Nasdaq trading terminal.
Â
Nasdaq, appreciative of the liquidity that Peterffyâs robot provided (not to mention the trading fees), soon embraced automation. The rest is history: The vast majority of stock market trades today are either generated or executed by algorithms.
Â
The algorithmic revolution has not been without its hiccups, of course: That same year, 1987, is more famous for Black Monday, when the Dow Jones Industrial Average, for no evident reason, fell 22.6% â marking its biggest ever one-day decline.
Â
We still don't know what caused the Black Monday crash. But we're pretty sure it had something to do with computers.
Â
An AI-induced crash would be even more inscrutable.
Â
Artisanal stock markets
Â
AI bots cannot yet operate in markets without a human chaperone: They wouldnât get past compliance departmentsâ KYC checks without a driverâs license, for starters.
Â
And finding a bank to clear large trades is not easy: You will probably have to go see your prime broker in real life at least once.
Â
But once theyâre approved to trade, AI bots wonât really need humans.Â
Â
Last week, Ramp, a financial software firm, introduced a new offering, whereby its AI bot will analyze the business software products you use and then contact your software providers to negotiate lower prices on your behalf.
Â
Soon enough, it'll be AIs on both sides of every financial negotiation: Your AI trading bot will negotiate margin fees with your prime brokerâs AI, clearing fees with your clearerâs AI, exchange fees with the exchangeâs AI, âŠ.
Â
And at that point, we will have no idea whatâs going on.
Â
Iâm not an AI doomer: The most likely result is that AI, like Peterffyâs robots, will make financial markets more efficient, to everyoneâs benefit.
Â
But there's a tail risk that things go horribly wrong: Imagine, say, an AI trading strategy that learns to hack into corporate email systems and then trade on that information.
Â
Large-language model AIs are inscrutable â thereâs no telling why they do the things they do, even after the fact. So, weâd have no idea if some outperforming AI was trading on insider information.
Â
Even the suspicion that it might be could be enough to undermine faith in markets.
Â
And that faith would not be easily restored: We wonât know which AIs are the bad actors or have any real way to deter them even if we did.
Â
If we donât want to accept that tail risk, the only solution I can see is to âair gapâ financial markets: Disconnect all the computers and go back to open-outcry floor trading with paper tickets and physical delivery.
Â
Weâve got artisanal everything else these days. Why not artisanal stock markets, too?
Â
Stress-testing DeFi
Â
The AI bots will be coming for crypto markets, as well, of course.
Â
But that might be good news: Transparent, deterministic DeFi could well prove more AI-resistant than opaque, unpredictable TradFi.
Â
Cryptographic proofs-of-humanity may be more effective than human compliance departments in keeping adversarial AIs out.
Â
And to the extent they get in, they may do less damage: DeFi markets may fare better against adversarial AI because DeFi is already highly adversarial. Â
Â
The rules of DeFi are enforced by code, which is stress-tested by attackers the moment itâs released into the wild.Â
Â
TradFi, by contrast, relies on regulators promising retribution â follow the rules or else â which seems unlikely to deter robots. Will robots have any fear of either prison time or social disapprobation?
Â
There is no single point of failure in DeFi: Imagine if an AI attacker hacked the DTCC, the ultimate authority of who owns what stocks. TradFi would receive historyâs biggest ransom note and weâd have no choice but to pay it.Â
Â
Fortunately, thereâs no such weak spot in decentralized crypto markets.
Â
But there are still plenty of smaller weak spots.Â
Â
So, we need more DeFi attackers, not less â and we need them now, before the robot attackers get here.
Â
Because, as evidenced by Thomas Peterffyâs robot, there will be no keeping them out.
Â
(Assuming my artisanal markets idea doesnât catch on.)