Insights
In the end, the truth will be the only currency that matters.
A decade ago, when skinny jeans and no-show socks were still in fashion, two friends were playing chess cross-country by taking turns posting images of a graphical board in the comments section of a GitHub issue. Github comments are often a battle of wits, so this was not different from business as usual, lest you think this is a weird way to play chess.
I knew one of them was cheating. Across our desk, I’d see him make his opponents move into a chess game, and then replicate the computer’s countermove into GitHub. Back and forth, the game continued for most of the day. I don’t remember how it ended because watching other people play a game is the most boring thing in the world. This is true of all sports and board games. A few days later, we discovered that the opponent was cheating similarly. Two computers were playing chess against each other, helped by humans in the loop. When the computers made a wrong move, my human friends would faithfully copy this over, doomed to lose their piece.
A similar thing is happening with GPTs today.
Alice used to spend a lot of time on Twitter, being a thought leader on blockchain, COVID, VR, and recently AI (in that order). Briefly, she also flirted with robotic process automation, but that yielded fewer keynotes. Alice is too busy to play chess. Alice’s primary skill, though, is expressing an idea in a few words. Alice uses this skill to write prompts that Merlin, her favorite AI, uses to generate long-form essays. Some of these make it to her LinkedIn; a few are published on her SubStack. I’ve heard rumors that the best are circulated on an invite-only list. You’re most certainly not on it.
Now, Bob’s business card says Capture Manager. He hands them out a lot; you probably have one from the last conference you attended. Surprisingly, Bob’s job isn’t retrieving escaped puppies from the Florida Everglades. He’s actually in government technology contracting. He wears suits (dark blue, no tie) and drinks whiskey with other similarly dressed business professionals in mediocre hotel lobbies. This supposedly yields contracts and is critical to his line of work.
A key part of Bob’s job is to be conversant on the bleeding edge of emerging technologies. He never knows when he might run into a contracting officer or a government official quickly crossing the lobby to go up to the cheapest room the per diem could afford.
The government only has two kinds of problems. Either they must modernize a system developed over 50 years from specifications buried somewhere in Federal Register footnotes, or digitize a paper form. It is a provable theorem that all government problems can be reduced to one of these two. While the computational and sociocultural difficulty is akin to solving P vs NP, this does not prevent Bob from suggesting the latest blockchain, VR, or AI approach. Bob once rode an elevator with Grace five years ago and subsequently won an $11M contract. That was only the third of five digitization attempts, so it wasn’t successful. Bob did get a tidy bonus, though.
He hasn’t had much success recently, but is a massive fan of Alice’s work. They were at a karaoke afterparty a few months ago, and he’s even on the invite-only list. Things are moving quickly in technology these days, and Alice, assisted by Claudio, is pumping out a lot of text. Bob needs to keep up with the latest trends in technology. He’s been stressed recently; it’s a lot.
So, Bob does what he did in college to get through Basket Weaving in Medieval Literature class. He turns to the CliffsNotes of our time.
He takes Alice’s writings and pastes them into the AI subscription his company begrudgingly allowed him (after three justifications memos and six email threads). This generates summaries he can paste into the Notes app on his iPhone and have Siri read to him on the long drive between events. He’s been happy with this, and the premium subscription to Claudio, it turns out, is mostly paying off. Recently, the incumbent on a large healthcare contract handed Bob a business card and said, “Let’s talk.” Wow.
And so, in corporate America, large swathes of content are already both being produced and consumed by machines, sometimes even by the same AI. However, the humans in the process, people like Bob, are desperately relying on AI to advance their careers and land their next deal lest they be fired. Or perhaps, only slightly more critical, doctors use AI to help with patient care.
But AI makes mistakes. It does it when it takes Alice’s notes and expands them into essays. It does it when it takes that essay to create notes for Bob. It does it in the office, the clinic, the boardroom, and the bedroom. Right now, somewhere at a computer near you, AI is making a mistake.
AI has no shame; it makes mistakes with confidence. Perhaps the technology will improve with fewer errors, or maybe the mistakes will just become harder to detect.
In the near future, AI will produce and consume most content. Some of it, helpfully summarized to individual tastes, will leak out of the machine and end up in front of a human. And that person, about to make an important decision, will want to know if what they see, hear, and feel is the truth.
And so they will eschew digital forms of communication (after all, the AI is always listening), trudge up a tiny brownstone, and walk through the weathered door of their favorite hallucination verification company. Because the way to combat AI hallucinations will be for the wealthy and busy to pay others for verification. And so, human armies will be deployed to confirm that the island has sunk into the ocean, the war has ended, or the hungry have been fed. The street price of the truth will skyrocket as markets adjust.
And while the machines burn dinosaurs, mainly talking to each other, this truth will be the only thing that matters. I just hope it’s easy to find. Bob’s job is on the line.