Will AI take everyone’s job? Only if we hire it to do so.

Recent headlines about AI have been dominated by fears of extinction risks and tech companies hacking governments. Yet, when you ask most people what they’re worried about, it is a lot more tangible: "Will AI take my job?". 

The conversation on AI and jobs usually starts from a statistic, such as “18% of jobs in the UK could be automated by 2035”, and then jumps straight into damage control mode, with proposals like a Universal Basic Income or retraining everyone towards manual jobs. That’s a big jump, one that seems to suggest we’re all spectators who should wait to see where the technology leads us. 

The impact of AI on jobs isn’t predetermined. It will be the result of millions of choices people and firms make about whether to hire AI systems for the job. Conveniently, influencing and constraining choices to create a socially positive outcome is exactly why governments exist in the first place. For all the talk about out-of-control AI agents, it is policymakers who seem to have lost their agency when it comes to AI and jobs. It’s time to reclaim it. 

Modelling is helpful, until it isn’t anymore 

Research from OpenAI's GDPval found the best models are on par with human experts on many economically valuable tasks, but how that translates to AI’s ability to do actual jobs is up for debate. Economists Soumitra Shukla and Alex Imas highlight that jobs with multiple dimensions are likely to stay, even if AI gets really good at some of the tasks. Alex Richmond and Ronnie Chatterji, meanwhile, suggest that attention, not intelligence, will start gaining economic value and be a growing source of employment. 

None of these researchers are pretending to have a crystal ball. Instead, they’re showing us where AI systems are incredibly powerful and where they have limitations, a level of complexity which too many policymakers still ignore. Looking at you, Rep. Jim Clyburn, who didn’t know what ChatGPT was until a week ago…  

Whether an AI will be able to teach students, administer medical treatments, or conduct a sales pitch is a question that will likely be settled in a lab in Silicon Valley or Shenzhen. But capability is only half the story. What matters for the future of work is where, when and how we decide to use it. 

 The path to too much AI 

Let’s return to those millions of decisions being made on AI deployment. Each will be taken by a (mostly) rational economic agent, like a firm wanting to increase productivity or an employee hoping to free up time for more enjoyable tasks. And yet, added up, these understandable decisions could end up putting millions of people out of work, making work more intense for those who stay employed, pushing wages down as expertise loses value, and concentrating profits in the hand of those who own the models and the chips. They could also lead to an all-too-familiar situation where machines replace humans in a way that degrades the service or product, like having to argue with an automatic customer service agent about an urgent problem.   

On the other hand, if AI is developed in the interests of the majority, there could be a lot to gain. Shorter working days would leave more time for other parts of our lives. We could keep the tasks we find satisfying, such as creative work or connecting with others, and automate the rest. Wider access to knowledge and learning could help communities develop in the ways that suit them. 

The real debate on AI and jobs therefore cannot start with the question of what we do once everyone has lost their job. It must start much earlier, in asking: how do we ensure that the millions of AI deployment decisions across the economy add up to the socially optimal use of AI?   

Reclaiming our agency over the AI and jobs narrative  

Ensuring every AI deployment decision adds up to an economy in which AI improves our lives, or at least improves the world of work, is a familiar coordination problem.  

Some governments might turn to central economic planning as one way to solve it. Many will impose red lines, with external commentators suggesting a “human job reserve”  that fences off some job categories from automation.   

For liberal economies like the UK, Canada and the US, one avenue could be to modernise corporate governance. It might not make for the most glamourous campaign slogan, but the idea is to change whose interests are considered when companies make decisions to add AI to their production process.  

This could include mandating that a third of corporate boards have elected worker representatives, requiring that employers consult employees before AI deployment and negotiate transition plans for anyone impacted by automation, and removing barriers to unionisation. It could also mean changing the legal duty of corporate directors, requiring they give more consideration for a company’s long-term prosperity over short term shareholder returns. 

Versions of these codetermination practices are common in many Europeans countries, giving workers structural power over decisions to deploy technology. This approach is likely to yield better outcomes from AI, for workers but also for the economy, because it includes in the decision making those impacted by the consequences. For example, a worker director can decide that if an AI tool is available to make production more efficient, everyone in that function could have reduced hours to share the productivity gains, instead of laying off part of the team. As employees know their work best, they can also raise flags against AI tools which promise to be efficient but could degrade the service, and advocate for worker-assisting tools instead. This may avoid, among other things, more companies re-hiring the workers they had initially fired when they believed AI could do the job. 

The call for economic democracy, where decisions over companies aren’t only in the hands of shareholders, is not new. However, the case for it may be more urgent today under the threat of AI automating everything to benefit a few, while the majority of us face an economy that neither needs us to run, nor runs to serve us. 

We’ll need many more institutional and policy innovations to get it right on AI, but one thing must change today: policymakers must stop waiting for a prediction to tell them exactly how large the welfare check will have to be when AI takes all the jobs and start realising AI will only take the jobs we hire it to do. Their focus must be on making sure those hiring decisions are in the interest of the majority.