Author: Byron Gilliam (The Breakdown)
Compiled by Deep潮 TechFlow
DeepInsight Summary: As AI agents begin to dominate market trading, human investors may no longer be able to understand the logic behind price signals. Princeton economist Brunnermeier has put forward a seemingly regressive but potentially lifesaving proposal: maintain a trading zone accessible only to humans, serving as an emergency backup for the entire financial system. This serves as a reminder to everyone relying on market pricing about trust and vulnerability.
Once a year, I and several colleagues set out from the investment bank where I worked that year, traveling from the gleaming office towers of Canary Wharf to an unknown wilderness in southern London.
When we got there, we only hoped not to be robbed on the way and to arrive safely at an unassuming building whose door our employee badge could open. We took an unmarked elevator up to the second floor, as we had been told that was the place.
As the creaking door swung open, it revealed a trading floor that was nearly an exact replica of the one at Canary Wharf.
But here it’s dim and windowless. The computer is at least ten years old: a bulky beige monitor with a thick chassis behind it, and the tower still has slots for a CD-ROM and floppy drive.
We are checking whether modern trading systems can run on this batch of vintage equipment.
Amazingly, they actually worked. We could execute trades and record positions just like at Canary Wharf. The speed was slow because those boxy monitors could only run one program at a time—but they did work.
Everything is normal.
This is our company’s “disaster recovery trading hall,” maintained year-round by the IT department. Cleaners come by regularly, but aside from dusting, they likely have little else to do—no one has ever used this space.
The bank must have spent a significant amount on this. But it’s an important form of insurance—not just for the bank.
In the event of an emergency—earthquake, cyberattack, or Godzilla invasion—every investment bank in London has a backup trading floor to rely on. No matter what happens, the market must keep running.
Today’s market may also need a similar form of insurance, just on a much larger scale.
Economist Markus K. Brunnermeier believes that to respond to an AI emergency, we should prepare a contingency option for the entire financial system: a market that functions at least as well as it did a decade ago.
Trade like it's 1999
Bill Gates believes policymakers should reserve certain economic activities for humans: “I believe that as AI and robotics advance,” he recently wrote, “we will deliberately keep some tasks for humans to do.”
He is referring to human-centered tasks, such as elder care, childcare, and teaching. It may also include healthcare (but only for delivering bad news).
But Brenner-Meyer believes traders should also be included on this list.
In "Artificial Intelligence and the New World of Finance," this Princeton economist warns that financial markets will become increasingly incomprehensible to humans as trading decisions are increasingly made by AI.
But AI can still understand us. “AI agents can learn how humans think and react,” he wrote, “while humans may not be able to understand or reliably predict how these agents will behave.”
This "asymmetric understanding" will become an issue.
At the most fundamental level, Brunnermeier warns that the asymmetric understanding between humans and AI could undermine the Hayekian function of markets: “Price signals are the mechanism that guides the operation of a decentralized economy,” he explains. But if we don’t understand what is driving prices, we won’t know what signals they are conveying.
Misreading the signals, we could drag the economy over the cliff.
Trust can also fail.
When AI agents trade freely in the market, collusion, manipulation, and rogue trading will become imperceptible to humans. “Under asymmetric understanding,” Brenner-Meyer writes, “the goals of AI agents cannot be fully described using human categories nor verified from the outside, making misalignment undetectable.”
If we fail to recognize misalignments in the market, we simply assume they exist. If we assume they exist, we dare not invest. If we don’t invest, we can’t build anything.
This issue may surge during a market crash.
If human investors cannot understand the reasons for the crash, they are unwilling to buffer the decline by taking the other side of the trade. They may simply follow the sell-off. “Asymmetric information can prevent some traders from participating,” Brennermeier warns, “and may cause the market to collapse entirely.”
It's pretty scary.
Fortunately, Brunnermeier also has a solution: regulators should create a market segment that allows only human trading.
I can only hope that place looks like the trading floor where I used to work: traders shouting out bids and offers, salespeople yelling orders, phones ringing nonstop, keyboards clacking, palms sweating.
That was truly a nostalgic scene.
However, there may not be much to shout about, as the conventional, AI-driven market will still handle the majority of trading volume.
Brunnermeier also acknowledges this: "In normal times, this slow market may not have much activity."
But like the expensive disaster recovery trading floor maintained by my former employer, this itself is a valuable form of insurance.
"The non-AI sector serves as a backup plan," Brennermeier said.
Then, if the market crashes without any clear reason, regulators can "temporarily open the boundary between the two sectors, allowing critical activities to migrate to the one that is still operational."
That’s the human desk. Orders are handled by people using computers from the 1990s.
Order processing speed won't be measured in nanoseconds like it is now. But at least we can understand what's happening.
“The underlying logic is to sacrifice efficiency during normal times in order to prevent chain reactions during crises,” Brennermeier explained.
He believes we should start trading this way as soon as possible, lest we forget how: “A market operated by humans keeps trading skills alive; without it, this backup plan will gradually wither as our reliance on AI grows.”
Just as GPS made us forget how to read maps, AI agents might make us forget how to trade.
I think the new market section should also have a dress code, so we don’t forget how to tie a tie or polish our shoes.
Although "human-only trading" may sound whimsical, it is not without precedent.
The same logic explains why governments around the world have consistently encouraged the use of physical cash. Although cash has become less convenient, it provides a crucial backup when cyberattacks disrupt the internet-based payment systems we heavily rely on.
Brunnermeier concluded: "Our current financial system was not designed for a world with AI agents possessing asymmetric understanding."
In that world (which now seems inevitable), a market segment where people trade as they did in 1999 would be a valuable backup option.
I still remember how to do it.
(but you can't remember it for too long.)
Byron Gilliam