Share article

ChatGPT and the Mass Production
of Office Work

Will the democratisation of Artificial Intelligence instigate the next wave of workplace automation?

The discourse around the future of AI seems to have shifted overnight. Fears of sentient technology, largely provoked by a Google engineer frightened that the company’s Large Language Model LaMDA had become conscious, have quickly given way to discussions of AI’s capacity to replace, or complement, human labour. With large language models like OpenAI’s ChatGPT now at the disposal of everyone with an internet connection – and able to efficiently de-bug code, analyse data sets, construct human-like dialogue, or describe quantum physics in the style of Shakespeare – the potential implications of the technology are becoming apparent. Are we at a turning point where AI fundamentally changes the nature of work and, by extension, society at large, or will the technology simply become part of the “giant organ of bullshittery” that is the internet, as one columnist for The Atlantic put it?

Prediction is impossible at this early stage, but this hasn’t stopped people from trying. Some commentators are convinced that the AI revolution is here and that what we are witnessing today are the early stages of a mass production moment for white collar work. In this scenario, AIs will either completely replace office and knowledge workers or greatly augment their productivity and output by supporting workflows and automating certain time-consuming tasks. Either way, work as we know it will be no more. Other observers are more restrained in their expectations, preferring to point out how easy it is to trick the AI to come up with ludicrous answers to queries, or that the technology is nowhere near reliable enough to do things on its own that people would be willing to pay money for – not yet at least. Attempting to rein in the most untamed expectations, the CEO of OpenAI Sam Altman recently urged the public to view ChatGPT as a “preview of progress” rather than a finished tool.

Subscribe to FARSIGHT

Subscribe to FARSIGHT

Broaden your horizons with a Futures Membership. Stay updated on key trends and developments through receiving quarterly issues of FARSIGHT, live Futures Seminars with futurists, training, and discounts on our courses.


Although the true potential remains to be seen, the many promising uses for large language models have already been demonstrated, and it’s not clear where their limits are, or if and how rules for their use should be enforced. While helping college students churn out assignments, the technology may also help teachers review essays or put together drafts for school curriculums – a potential win-win for both lazy teenagers and overburdened educators. OpenAI have themselves flouted the idea of cryptographically watermarking the bot’s outputs, creating recognisable patterns of specific word choices in its text to avoid, among other things, students cheating on their homework. Yet, as other open-source language learning models begin to catch up with the training level achieved by OpenAI, such initiatives will likely be redundant. Looking further ahead, education leaders will have to re-define the value of traditional, as well as modern, pedagogy – in a similar fashion to how the advent of pocket calculators forced teachers to redefine the math curriculum. Is recall memory still valuable in an era of AI? Should schools return to hand-written assignments and exams in supervised conditions? Or should they view AI similarly to tools like Grammarly and spell-checkers, an aid for those students who struggle to eloquently express their knowledge and ideas through writing?

Education is just one area where large language models can cause disruption. Its potential is far larger when it comes to the future of work. When trying to understand the deeper effects of this potentially world-changing technology it can be helpful to look for analogues in history. If the implications of the technology are as revolutionary as some believe, then what better place to search for comparisons than in the industrial revolutions of the past? The world has gone through three of these since the 1700s (with a fourth one currently underway). Each one has been characterised by a handful of key technologies – steam engines, assembly lines, IT systems etc. – which led to the automation of human labour, a reduction in the cost of producing goods and services, the creation of economies of scale, and the emergence of new types of work and employment patterns. Through their reshuffling of economies and labour markets, past industrial revolutions significantly impacted the political, cultural, and social fabric of society in various ways. The first one set in motion the transition from agrarian to urban industrial society; the second led to the rise of mass politics and suburban life; the third drove the migration of much of the workforce from factories to offices.

Importantly, the tipping points which enabled these revolutionary spill-over effects didn’t come from the invention of new technologies so much as from their wide application. On its own, the invention of the mechanical loom or the sewing machine in the 1700s didn’t change much, but when these tools became so widely adopted that low-skill textile workers could produce clothes at a lower cost and at greater speed and precision than artisans could compete with, the knock-on effects started kicking in. The textile industry exploded in output and demand, and garment prices plummeted. Global labour markets shifted, new industrial districts arose in the cities, novel markets for textile goods were created, and new consumer cultures and fashion trends emerged. Countertrends such as the arts and crafts movement even arose in response to mass production, and ‘handmade’ became a mark of quality.

The rifts in the social fabric caused by this wave of mechanisation were deep. Famously, it led to English textile workers banding together in protest of manufacturers who they accused of using machines to subvert existing labour practices by hiring cheaper, lower-skilled workers to replace them. They formed makeshift militias and smashed textile machinery in a desperate bid to prevent their livelihoods from being destroyed.

Will mass-produced knowledge work soon flood the web like mass-produced textiles flooded the global market in the 19th century? Most likely. Will large language models make parts of the white-collar workforce as superfluous as 19th century textile artisans? Not unthinkable. Will the societal effects of this new technology be as significant as the mechanisation of labour was in the past? It’s too early to say. But we may not have to wait long to find out. Decades passed between the invention of the mechanised loom and its widescale industrial application. With ChatGPT having reached one million users within a week of it becoming available to the public, it’s safe to assume this tipping point in adoption has been or will be reached soon – assuming this and other similar tools will remain relatively open to the public in the future. In addition to the timeline from invention to mainstream adoption having become greatly compressed, the know-how required for an office worker to competently use large language models is also much lower than it was for a textile worker needing to retrain to use mechanised tools.

This is where historical comparisons to today may reach the limit of their usefulness. It would be hard to imagine a scenario in which every 19th century English textile artisan was gifted a power loom and somehow supplied with the means of selling and distributing the products of their labour on their own as well. But we can safely assume that the outcome of history would have looked much different if it had happened that way.

Beyond restructuring modern work by relegating more tasks to the machines and freeing up human capacities to do the things AIs still can’t do, will large language models lead to overall technological unemployment? If so, which professions will be most at-risk?

The most immediate answer is those professions that are characterised by repetitive, routine, and high-volume tasks, especially ones requiring little to no human interaction. That is, large language models may eventually replace types of employment that are the 21st century equivalent of assembly-line workers of the past. Whereas the third industrial revolution led to the opening of the workforce and the ‘taskification’ of unskilled labour, large language models have the potential to reverse this trend, greatly decreasing the number of humans needed in the loop. One plausible scenario would be one in which we see a widescale ‘proletarianisation’ of the white-collar workforce, with digital technologies of automation enabling the displacement of skilled white-collar workers in a comparable way to how mechanisation enabled the displacement of skilled artisans.

The same could be true for the capacity of large language models and similar AI tools to shape the creation of cultural products – fiction, art, music etc. In this context, much depends on our appetites for consuming AI-generated content once its novelty wears off. Do we want to read novels or social media posts with no human author? Will we be happy with procedurally generated playlists, or will we prefer the human touch (no matter how imperfect or slow) to the spotlessness of an AI? Will cheap, instantaneously, and ‘good enough’ content make up 99% of our needs, with the artisanal, time-consuming, and expensive products of human labour filling the remaining 1%? Will ‘humanmade’ be to cultural products what ‘handmade’ is to physical products?

Although there are currently more questions than answers, those who wish for a humanmade future could find solace in the fact that not many enjoy spending their time watching chess computers compete, even though it’s been decades since any human has been able to consistently beat them. AI exceeding human speed and skill hasn’t mattered much in this case, nor has it caused the demise of professional chess. Plenty of people still enjoy watching human grandmasters play.



Explore the world of tomorrow with handpicked articles by signing up to our monthly newsletter.


Of course, it may be that most of us simply won’t care whether a human or AI is pulling the strings, and that we will eventually arrive at the kind of future prophesized in Orwell’s 1984, where books – a commodity to be produced and consumed like any other – are written entirely by computers. If this is the trajectory large language models have set us on, writers working today had better start looking for new careers soon. And the same logic could apply to any other kind of cultural product or human-authored content you could think of.

A future with nearly total automation of white collar and creative work, if even plausible, is far down the line. A more likely outcome in the near term is one where we see more deeply integrated collaboration between human and AI rather than total displacement. The AI exceeds the human in experience (achieved through deep learning) and can work at greater speed, but the human understands the meaning of the task and the quality of solutions in a way the machine can’t unless it becomes sentient. Medical work provides examples of this dynamic at work today; a deep learning AI may surpass a single radiologist’s ability to correctly diagnose cancer from X-rays. But a human radiologist and an AI working together has been demonstrated to be a far more effective combination because their strengths complement each other’s weaknesses.

A ’softer’ version of the future sketched above sees AI doing the grunt work of any kind of white collar and creative work with humans defining the parameters and putting the finishing touches to the output. Since attention will remain a finite resource no matter how much content AIs can churn out, this would likely spell bad news for the copywriters, editors, marketeers, and social media managers of the world – or good news, if you are one of those select few who will oversee and manage the labour of computers. In such a scenario we would likely see a broad need for reskilling, assuming there are new jobs to reskill into. If not, we will be forced to consider whether our definition of ‘work’ as something that takes place from 9-5 at the behest of an employer is sufficienlty broad. Housework, child rearing, volunteering – as long as we are comfortable letting go of conventions, work in a more AI-dominated future could be anything that positively contributes to society or the lives of the people around you.

If history is a guide, sudden technological changes can be both frightening and disruptive in the short term, but there will likely always be meaningful things for humans to do. We may just have to prepare for the future of work being weirder than we have come expect.