AI: Time for a Butlerian Awakening

I see a lot of my colleagues defending human translation on LinkedIn at the moment, which they are absolutely right to do. Large Language Models are useful tools, but for now, the translations they produce are still fairly mediocre and they require the input of an experienced professional to make them sound idiomatic. Buyers of translation services need to be aware of their limitations.

However, though understandable — laudable even — I feel that defending the translation industry in this way is somewhat missing the bigger picture. Even if human translators have both intrinsic and economic value, this value cannot offset the structural incentives now shaping the market as a whole. AI is not as incompetent as some of its detractors like to claim, and the translation market will continue to shrink anyway, even if humans do remain better than AI.

AI agents are now capable of working autonomously on certain tasks for days at a time, so a single highly competent translator/proofreader with an AI agent will surely soon be able to do what previously required a team of project managers, translators and proofreaders. And make no mistake, this is certainly also true for most other knowledge workers, and anyone else whose job is mainly carried out using a PC.

According to Mustafa Suleyman, CEO of Microsoft AI, the majority of white-collar jobs will be at risk within 18 months. If we assume his beliefs about the capacity of agentic AI to replace human workers are more or less correct, then it is no longer enough to simply assert the superiority of human translators, we need some sort of concerted response from the political class and civil society alike. You don’t need to be an economist to understand the potentially devastating ramifications of job destruction on this scale… and yet I can’t shake the feeling that we are just sleepwalking into this new reality.

Depending on how you calculate it, the white-collar jobs Suleyman is referring to would concern around 50% of the workforce (and this is without mentioning the ongoing automation of many jobs in industry and services). Even if it is safe to assume there will not be mass layoffs overnight (especially in France where there is some measure of labour market protection), I do think it’s worth thinking seriously about what would happen to society if a significant proportion of the population was made redundant in short order.

Unemployment is a crucial economic metric because it has a scissor effect on public finances: it simultaneously reduces tax revenue while increasing government spending in the form of unemployment benefits and housing support. In most Western countries, social security is largely funded by payroll-based contributions, so mass unemployment also drastically reduces the pension system’s inflows. In order to keep paying pensions, housing and unemployment benefits, governments would have to borrow heavily, which would in turn raise the cost of servicing their debt, leading to a sovereign debt spiral.

Moreover, the population most likely to be affected by unemployment is the young, and blocking the pathway to jobs for young people is akin to cutting the branch our economies are sitting on. If not enough people have the incentive or opportunity to spend the thousands of hours of practice required to achieve subject-matter expertise, who will be competent to sign-off AI agents’ work in 20 years’ time?

As we’ve already seen in parts of the United States and Europe, when jobs disappear, the damage isn’t only economic, it can be seen in health outcomes, social cohesion, and political radicalisation. So, even if you are sceptical of the more extreme claims about job replacement, it’s worth being cautious about technological choices that systematically close off entry-level routes for the next generation.

This is dark, I know, but it’s not just about protecting jobs, it’s also about becoming cognizant of our relationship to technology. There is now more and more research on cognitive debt, which may still seem like a very arcane concept, but it is one that everyone is familiar with. Who is over reliant on their GPS? Who still memorises phone numbers? That is cognitive debt: as soon as you outsource cognitive tasks to a new technology, you lose competence in that domain.

To be clear, I’m not calling for a new Luddite revolt or a Dune-style Butlerian Jihad against AI; first and foremost, we need a cultural awakening: people should be aware of the coming wave and its implications. Economic incentives matter, but it is individual managers and business owners who decide whether productivity gains are reinvested into quality and growth with due consideration for employees and suppliers… or simply used as excuse to reduce headcount. There should be transparency about which companies are replacing humans with AI, and why, and the most egregious cases should have a reputational cost. Large companies already have to publish a lot of workforce data under the EU’s corporate sustainability reporting rules, so it should be realistic to tweak the existing metrics to make AI-driven downsizing visible in this reporting.

It seems fanciful at this point in history to imagine that our political elites have the measure of this problem, but in so far as they understand it, governments also have a duty to inform and prepare the population. Finally, some kind of regulation is also now surely urgent, both to slow and manage the pace of change and to make sure companies do not capture all the added value. In the same way that companies have to pay for negative externalities associated with environmental damage, the productivity gains from AI should be redirected towards offsetting negative externalities associated with the societal damage they cause. The stability of the welfare state, and society more generally, depends on it.