The jobs don’t come back

“There’s an enormous amount of productivity that we’re getting from AI - the equivalent of thousands of hours of work,” Dropbox CEO Drew Houston talked about in an interview in early 2024. “It doesn’t mean we don’t need people, but it does change the shape of what work looks like here.”

This straightforward admission didn’t exactly surprise me. Talking to my peers at Cambridge, I realised that we had honestly just grown used to hearing leading figures in tech talk about AI in such positive ways - a strong tool, a force multiplier. The argument boils down to the fact that you can do “more with less”. However, the deeper consequences are of course still sinking in and at some point, a job that would have gone to someone, didn’t. And no one really noticed.

The underground nature of AI’s deployment across the labour market has not always been clearly acknowledged, as it is not always dramatic. There are no mass layoffs, no flashy automation robots marching into factories (cue I, Robot factory scene!) but rather just a slower, steady absorption of tasks by algos that are getting better and better. A Harvard Business School study found that BCG consultants using GPT-4 completed tasks 25% faster and with 40% higher quality than their peers. There’s a quiet efficiency revolution underway, and it’s not being tracked on most job reports.

Of course, profit-maximising firms are never hiring for the sake of it, and never did. What AI really crushes is the reasoning for employment across many sectors - if a single lawyer can review more contracts with an AI assistant compared to three, or if a small marketing team can generate campaigns without hiring as many freelancers, what would occur with the rest of the workforce?

Recently, the impact is becoming harder to ignore, even in larger, traditionally “secure” corporate environments. In a recent internal memo, Amazon CEO Andy Jassy said that many white-collar roles may be reduced because generative AI is starting to take over routine tasks. However, he did note that some new positions may emerge, but the overarching message was very clear: office work is not immune to being automated away. The same technology that once promised to “liberate the workers from drudgery” is now viewed more and more as a reason to not hire them.

Of course, some industries have already realised the effects e.g. legal services, customer support, copywriting - all fields where LLMs can replicate human effort in narrow, repetitive tasks, often times better than the average human as well. As models get better at understanding context, nuance, and tone - three very “human” features, the boundaries of what can be AI-wrapped keeps shifting. As one of my friends put beautifully: “The disruption isn’t total, but it is cumulative - it chips away over time”.

The most common retort to this argument is that new jobs will emerge, as with previous technological revolutions in the past - from the steam engine to the Internet. However, this time, it’s not particularly obvious to me where the new work will come from. This is because I believe AI will cut into cognitive work and creativity - the fact that it can replicate and scale instantly with no rest and minimal costs (compared to labour) means this would be different to past technological revolutions where jobs were reshuffled instead of being lost.

Some economists are still optimistic, and they argue that AI could boost productivity so significantly that it spurs new industrials we can’t yet imagine (Phillipe Aghion). It could be, but that kind of technological optimism feels a bit like betting the house on future mystery jobs. Right now, many of the jobs disappearing aren’t being reallocated - they just vanish. – CONTINUE FROM HERE –

Over a couple of work experiences I completed in sixth form (at JP Morgan Chase and Macquarie), I chatted to executives about what they think comes next, and they argued that many AI initiatives aren’t just about streamlining work to make it easier, but rather just [replacing human roles outright]. One VP advocated for “humane automation” - a deliberate approach to AI integration that is transparent in acknowledging the social costs of job displacement, specifically to avoid hiding it behind vague terms like “optimisation”. I was always of the belief that this kind of honesty is rare, and it still is - but maybe it is becoming necessary. These views are mirrored well in this article.

And so this forces the question: what happens to those who cannot find work as the world moves forward? Not just temporarily, but structurally - an increasingly automated economy would force work (as we define it today) to become scarcer, and this means that the old model of tying income to employment begins to break.

Some key people, like Sam Altman, have been fairly blunt about this. The OpenAI CEO has repeatedly suggested that artificial general intelligence (AGI) could create such vast economic abundance that traditional jobs become obsolete. In that world, he feels a universal basic income (UBI) or similar redistribution mechanism would be necessary. “We should aim for a world where everyone can share in the benefits,” he said in a 2023 blog post. To be honest, I find it to be a compelling vision, apart from the obvious issues with UBI (immense implementation costs and inflationary risks). This vision would also require enormous political coordination, public trust, and crucially, actual buy-in from those profiting most from the AI boom. Additionally, his view is of course biased due to his stake in the AI boom, so please do take it with a grain of salt!

However, so far, there is little sign of global protective UBI measures happening. Most governments are sadly still grappling with basic AI regulation - let alone serious social safety nets. In the US, the idea of UBI remains politically very much on the fringe. In Europe, conversations around potential “AI dividends” or robot taxes exist, but merely at the level of think tanks and opinion pages.

Meanwhile, the labour market is also shifting - a growing number of workers are in limbo isn’t quite replaced, but not really needed either. Some retrain, others drop out of the workforce altogether. And the old promise, that hard work and upskilling will always lead to stability, starts to look doubtful.

To me, none of this is inevitable. Policy can shape how technology is deployed; companies can choose always augmentation over replacement; societies can build systems that value care work, education, and other roles that machines still struggle with. However, those decisions require foresight - and right now, to put it succinctly, it feels like everyone’s racing ahead without checking the map.

There’s an uncomfortable irony in the current moment, in the way that the more productive we become, the harder it may be to justify paid employment for all. The question isn’t just what AI can do - but what we, as a society, choose to do with it. And that’s not a problem any algorithm can solve.

Click here to go back to the home page!