5 Reasons There Won't Be an AI Jobs Apocalypse

Cope & Glory

I genuinely love this campaign; funny and unignorable.

• ATTENTION: On Monday June 16th I’m returning to the Kiln Theatre in West London to talk JOHN & PAUL. Book your tickets now!

• I’m also talking J&P at York Festival of Ideas this evening.

Catch-up service:
Why Is TikTok Rocket Fuel For Populists?
We’re All Viennese Now
Why Am I At My Lowest In The Middle of the Night?
Seven Features Of Post-Literate Politics
Nine Principles For Success In the Age of AI

Last week the CEO of Open AI’s rival, Anthropic, sounded an alarm about employment in the age of AI. Dario Amodei said that politicians and CEOs are in denial about what’s just around the corner, either out of ignorance or fear. Stop sugar-coating it, he said: AI is going to put huge numbers of people out of a job. With intelligence ‘on tap’, the sectors hit hardest will be white collar ones: law, finance, technology, consulting. Amodei thinks half of all entry-level jobs in these industries could be eliminated within five years.

An AI researcher at his company recently went even further: “Even if AI progress completely stalls today and we don’t reach AGI… the current systems are already capable of automating all white-collar jobs within the next five years.” Indeed we may already be feeling the first tremors of the earthquake. In the US, unemployment among college graduates has increased sharply and particularly in technical fields like computer science where AI is most applicable.

My default position on this, for now, is that if the job you were aspiring to has been taken by an AI it was probably a really boring job; one that was more about following routines and procedures than exercising judgement, imagination, or social intelligence. You haven’t lost a job so much as gained a thousand days that don’t feel like living death.

But maybe the machines are about to eat the jobs we actually like doing, too (or at least the jobs which act as pathways to fulfilling careers). Maybe we’re about to enter a future in which, as Amodei puts it, "Cancer is cured, the economy grows at 10% a year, the budget is balanced — and 20% of people don't have jobs."

Let’s leave aside the fact that this would be all upside, since about a third of working-age adults in the US and UK don’t have jobs to begin with. Let’s also refrain from dwelling on whether Amodei’s scenario makes any economic sense at all (in a consumer economy, you can’t have fast economic growth and mass unemployment; someone’s got to buy all those refrigerators). Let’s presume the AIs keep getting better. Let’s also presume that even when AIs don’t directly replace existing jobs, they may make it harder to do them profitably, by slicing away parts of the job.¹

Given all that, are all white collar jobs doomed? Should we just accept it’s over? Maybe. It’s the end of a busy week and I feel like a lie down anyway. But maybe not.

I’m not sure Amodei and co. are wrong, but I do think they are madly over-confident in their prognoses. Unlike some, I’m not instinctively hostile to “tech bros” and I don’t think they say this stuff just to hype their product; I believe they believe it.² But they are experts one thing: the abstract world of AI systems. They’re not experts in economics (plainly), or sociology or psychology. They know little about most of the industries they say will be transformed. They’re not the deepest thinkers about history or human nature. In short, when they make these sweeping predictions, they are way out of their lane. Yet their pronouncements are received with the utmost seriousness by the media, even though they might as well be pop stars opining on politics.

I won’t claim to know what’s going to happen, but here are five reasons to doubt these apocalyptic predictions, drawing on a few things I’ve noticed about how the real world works.

The first is free to read, the others are behind the paywall, along with a big, beautiful Rattle Bag of links to what I’ve been reading and enjoying. The Ruffian relies entirely on paid subscriptions.

1. Curse of Expertise. I worked in advertising for a long time before deciding to move into the famously lucrative freelance writing industry. In recent years I’ve occasionally consulted on various ad-related projects. If I don’t know what to say in these meetings, I’ll venture something that to me sounds rather banal, just to remind people (and me) that I’m in the room. More often than you’d think, these statements of the obvious are greeted as remarkable insights. I don’t think this is just people being polite; it’s also that all those years of struggling to do the job well compacted into a sediment of domain-specific priors which now forms the basis on which I think about new problems. This subconscious common sense makes me smarter than I feel.
   
   We don’t know everything we know. Humans have many skills they’re not aware of - for instance, you can probably tell hot water from cold just by hearing the tap run - and this is also true of your job. If you’ve been asked to train someone, you can impart your explicit knowledge about how to do what you do. That knowledge can be articulated, written on PowerPoint charts, codified into processes, entered into a training dataset. But it’s much harder to transfer your implicit knowledge, those hidden layers of know-how and intuition acquired from experience.
   
   It’s why experts make bad teachers. Implicit knowledge is tricky to put into words or to make abstract in any form. You often don’t even realise you even have it until you’re in situations that require it. So - presuming it’s valuable - I don’t know where the AIs will pick it up from. Junior employees can pick up some of it from observation of more experienced peers, but it is only truly gained from acting in and on the world; from hanging out, blundering around, making errors. Humans continually learn on the job, and the smart ones acquire the subtler aspects of their work quickly. AIs are not learners in the same way - not yet anyway.
   
   Crucially, I think it’s implicit knowledge that helps people define the problem in front of them rather than just accept it; to ask, for instance, whether the boss, or the client, is asking the wrong question.³ That’s a very valuable ability. Will AIs be able to replicate it? Perhaps, one day. But this is the difference between judgement and intelligence.
   

2. Hidden Complexity. In 2017, Elon Musk went all out (he always goes all out) to automate his Tesla factory in California, in order to achieve his wildly ambitious productivity goals. He introduced an intricate network of conveyor belts and robots for tasks like welding and assembly and quality control. And the factory’s productivity dropped. The robots weren’t as fast or reliable as humans at what were apparently simple tasks. So Musk had to rip out many of the robots - he cut a big hole in the side of the factory to get the defunct machines out quickly - and replace them with people. He later admitted, “Humans are underrated”.
   
   In 2018 I read a comment on Marginal Revolution (it was highlighted by Tyler Cowen) that I still think about a lot. It’s one of the most powerful insights into AI and automation I’ve ever read.