I keep having the same conversation.
I'll be talking to someone about AI, and they'll wave it off. "That'll never work. I tried it…"
AI messed up my trip planning. It'll never replace a travel agent.
AI put six fingers on the hand in my photo. It'll never replace a photo editor.
AI botched my translation. It'll never replace a professional translator.
AI cited case law that doesn't exist. It'll never replace an attorney.
AI got a number wrong. It'll never replace a financial advisor.
AI got my taxes wrong. It'll never replace a tax advisor.
AI makes up facts about history. It'll never replace a teacher.
Right now you can tell AI wrote this. It'll never replace a writer.
Every one of those is correct, if you draw AI's future as a straight line. But watch what you're doing. You're taking today's flaws and concluding AI will never get good enough to matter. That snapshot is going to expire. The list you handed me isn't a verdict. It's version one.
This happened with the iPhone
When the iPhone came out in 2007, plenty of people looked at version one and called it a toy, nothing the existing phone makers needed to worry about. Their reasons:
No physical keyboard, so it was useless for real email. Five hundred dollars when a perfectly good phone cost ninety-nine. A 2-megapixel camera that couldn't even shoot video. No third-party apps. No copy and paste. No picture messages. 2G, so painfully slow internet. Locked to one carrier. A battery you couldn't replace.
And I can't argue with them. The first iPhone was pretty bad. If you assumed it would stay that way for the next twenty years, sure, it should have been a small blip in the history of the mobile phone.
The brain glitch that causes this
There's a name for this, and it's one of my favorite terms in psychology: exponential growth bias. Humans are terrible at intuiting things that compound. Our brains draw straight lines. When we see a technology improve, we instinctively extend that improvement out in a flat, gentle slope, a little better each year. But a lot of technology doesn't move in a straight line. It compounds. It moves in step changes. And when something is curving upward and your brain is drawing it as a straight line, you don't just underestimate the future a little. You miss it entirely.
So you've got one glitch telling you "this is how it'll always be," and another telling you "and it'll only get a tiny bit better from here." Stack them, and you've got a person confidently, calmly walking into a wall.
How fast is AI actually improving? There's a number for this.
A research group called METR measured it. The length of task an AI can do on its own has been doubling about every seven months, and it's held that pace for six years. If it holds for six more, that's another thousandfold improvement from here. There's just no way a human can picture a thousand-times jump in six years. That's the whole problem.
The most expensive snapshot in tech history
In 2007, Steve Ballmer was the CEO of Microsoft, one of the most powerful companies on earth. When the iPhone came out, he laughed at it. In an on-stage interview he said, and I'm quoting him directly: "There's no chance that the iPhone is going to get any significant market share. No chance."
His reasoning? At the time, the iPhone kind of sucked, for all the reasons above.
Now sit with that for a second. His entire case against the iPhone was built on a snapshot in time. He looked at version one and drew a flat line into the future.
Eventually, future models got cheaper and more powerful. The keyboard objection evaporated the moment people got used to glass. And by the time he admitted the iPhone had "clear market momentum," it was years too late. The iPhone alone would eventually generate more revenue than all of Microsoft.
It's kind of ironic, because I'm reading Hard Drive: Bill Gates and the Making of the Microsoft Empire right now. The first computer Bill Gates worked on could hold about 250 characters of text. Yes, that's correct. The entire machine stored 250 characters. And he was able to see into the future and imagine the utility of a computer that could store infinitely more, and the change that would set off in the world. He didn't draw a straight line out into the future. He understood technology's unique ability to compound. Now a USB thumb drive in your junk drawer holds 256 gigabytes. That's a billion times more storage than Bill Gates's first computer, sitting in your pocket.
And it's not just computers. Sequencing the first human genome cost about $2.7 billion in 2003 and took over a decade. By 2022 it was under a thousand dollars, a millionfold drop, and today it runs in under an hour. Moore's Law did the same to the chip, transistors doubling every two years until the phone in your hand, pure science fiction in 1975, became background noise. Solar power per watt has fallen more than 99% and is still dropping. Different industries, same shape.
Here's the part that matters. Every one of these looked like a toy or a boondoggle at the start. 250 bytes. A multibillion-dollar science project. A chip with a handful of transistors. The skeptics in each era weren't wrong about the snapshot. They asked "is this good enough today" when the only question that ever mattered was "where is this going."
So when someone shows me a six-fingered hand and tells me AI is a dead end, I know exactly which mistake they're making.
The fix
You can't fully out-think a cognitive bias. It's wired in. But you can install a habit that catches it.
Whenever you catch yourself deciding a technology "won't work," stop and ask one question:
Am I judging this by what it is today, or by where it's actually heading?
If your entire case is built on the current price, the current quality, the current limitations, congratulations. You're holding a snapshot. You're doing the Ballmer. That's not analysis. That's describing the present and calling it a prediction.
And Ballmer wasn't alone. In 2007, Nokia and BlackBerry owned the phone business. They saw the same flawed version one everyone else did, decided it was a toy, and went back to building keyboards. Within a few years both were essentially gone from the business they had dominated. They didn't lose because the iPhone was good in 2007. They lost because they bet it would stay that way.
That's the part that should keep you up at night, because it's happening again. Only this time it's aimed at jobs instead of handsets.
Look back at the top of this piece. Travel agents. Photo editors. Translators. Attorneys. Financial advisors. Tax preparers. Teachers. Writers. Every one of those industries is doing right now what Nokia did in 2007. Staring at a flawed version one, pointing at the six fingers and the made-up case law, and deciding they're safe.
Maybe your industry really is the exception. But "it can't do my job today" is not the evidence you think it is. It's the spec sheet for version one. And version one is the worst this technology will ever be again.
So stop asking which jobs are safe. The real question is how long before we're all asking what AI can't do better than we can.