Paul Frijters paints an expressionist picture of artificial intelligence’s impact on work
Artificial intelligence (AI) is impressive. That much is becoming difficult to dispute. When I asked ChatGPT, “In what way is Bitcoin not money?”, it produced a better answer than most of the 40 finance students I had previously asked in an exam. When I suggested that Hansel and Gretel might have been lying about the witch trying to eat them and asked why they might have killed her, ChatGPT immediately grasped the point: perhaps they were poor and wanted her wealth.
Facial-recognition systems can now pick me out as I cross European borders. These technologies are not simply hype. But, as with every major technological change, AI has brought both hallelujahs and predictions of apocalypse. Neither is particularly useful. The more interesting question is: what is it actually likely to do to work?
I see four developments already emerging:
- the disappearing middle;
- knowledge retreats behind walls;
- an atomised workplace; and
- an industrial revolution in bullshit.
The disappearing middle
AI is making experts more productive while potentially making it harder for younger workers ever to become experts themselves.
This is something I hear repeatedly from experienced people using AI in their work. Production specialists can use it to redesign factory processes faster, reducing the need for assistants. Experienced software developers increasingly use AI to write code, relying on their own expertise to spot where it has gone wrong. Engineers can generate designs rapidly and then use their judgement to correct them.
The same applies to lawyers, academics and other knowledge workers. The better you already understand your subject, the more useful AI becomes.
But there is a catch.
Experts are able to spot AI’s mistakes partly because they have spent years making similar mistakes themselves. Expertise is built through struggle: trying things, getting them wrong and slowly acquiring an intuitive understanding of why.
AI offers younger workers a shortcut around much of that struggle. The danger is that the shortcut also removes the process through which expertise is created.
If organisations stop employing and training juniors, where will the next generation of experts come from?
In the short term, this could mean fewer entry-level opportunities. Why employ several junior staff to undertake research, coding or routine analysis when one experienced worker equipped with AI can do much of the same work?
The longer-term problem is harder. If organisations stop employing and training juniors, where will the next generation of experts come from?
We may eventually invent new forms of lifelong learning capable of developing that depth of understanding. But they do not yet exist at anything like the scale required. The result could be a hollowing-out of the middle of the knowledge economy.
Academic careers, for example, may become more elitist. Only those with a realistic prospect of reaching the research frontier will have much incentive to endure the long process of developing deep expertise.
At the same time, many jobs requiring less than frontier knowledge could need much less formal training. An engineer repairing a machine may increasingly be told by an AI what has failed and what to do about it.
Something similar could happen to many personal services. A gardener might use AI to diagnose a garden’s problems and devise a planting plan, while concentrating their own skills on understanding what the client wants and carrying out the work.
Doctors, lawyers and other professionals may likewise spend less time researching and analysing and more time dealing directly with people.
AI would then produce an apparently paradoxical labour market: a small group of highly knowledgeable specialists at the top and a much larger group implementing AI-generated recommendations beneath them, with fewer people occupying the professional middle.
Knowledge retreats behind walls
The internet on which today’s AI systems depend was built partly from knowledge that people and organisations made publicly available.
Now that companies can monetise that accumulated public knowledge through AI, others have an obvious incentive not to make the same mistake.
Businesses, universities and professional organisations may increasingly keep valuable new knowledge inside closed systems, available only to their own employees and their own AIs. Confidentiality becomes commercially more important precisely because information has become easier to exploit as this paper explains.
The supposedly frictionless knowledge economy could therefore produce more secrecy, not less. AI will also increasingly connect to physical machinery. Today’s language model that writes an email can become tomorrow’s system controlling cars, drones, household equipment or industrial machinery.
Systems capable of acting on instructions can be given malicious instructions too. That is likely to encourage another technological arms race.
Much of this will be wonderfully convenient. Tell the kitchen to prepare three pizzas with baked beans and serve them on the good china, and perhaps one day it will. But remote action also makes destructive behaviour cheaper. Systems capable of acting on instructions can be given malicious instructions too.
That is likely to encourage another technological arms race: more surveillance, more security and greater physical separation between those able to protect themselves and those who cannot.
The rich have always purchased insulation from social disorder. AI may increase both their reasons and their ability to do so.
The atomised workplace
AI also offers something human societies have traditionally obtained from one another: attention.
Young people already use chatbots for advice and companionship. AI companions could become commonplace in nursing homes and other settings where genuine human interaction is expensive.
Workplaces may follow the same logic.
Teachers, nurses and care workers could increasingly find themselves instructed, monitored and assessed by AI systems. Employers will have a powerful incentive to replace expensive conversations between people with cheaper interactions between a worker and a machine.
Some people will inevitably choose predictable artificial relationships over the difficult business of dealing with other humans.
That may improve efficiency as measured by management. It may simultaneously make work more lonely.
The consequences stretch beyond employment. Artificial friends and romantic partners are already technically plausible. As they improve, some people will inevitably choose predictable artificial relationships over the difficult business of dealing with other humans.
Communities and families will have to compete with machines specifically engineered to provide us with the interactions we want.
An industrial revolution in bullshit
AI may have found its perfect market in what David Graeber called “bullshit jobs”. A large part of organisational life consists of producing strategies, visions, reports and presentations whose relationship with what an organisation actually does can be remarkably slight.
Organisations still need to justify hierarchies, protect powerful groups from scrutiny and maintain comforting stories about what they are doing. AI merely makes producing the necessary words dramatically cheaper.
Large language models are extraordinarily good at this. They can generate plausible corporate prose almost instantly: strategies full of priorities, frameworks, pathways, commitments and stakeholder engagement. Producing such material, which once consumed armies of managers and consultants, is becoming almost costless.
That does not mean bullshit disappears. Its social function remains.
Organisations still need to justify hierarchies, protect powerful groups from scrutiny and maintain comforting stories about what they are doing. AI merely makes producing the necessary words dramatically cheaper.
Indeed, the quantity of such material could explode. There may soon be far more reports, strategies and consultation documents than any human being could conceivably read, let alone check. One paper documenting this nicely calls them “redundant publications.”
AI could automate bullshit without eliminating the bullshit job. The human job would shift from writing the bullshit to convincing other humans to read the AI-created stuff.
What happens next?
None of these developments requires science fiction. Elements of all four are already visible: the hollowing-out of junior knowledge work, the enclosure of information, AI companionship and the industrial production of corporate prose.
What is much harder to predict is how people will respond.
Technologies do not simply arrive and determine society. Humans build institutions around them, resist them, regulate them and invent new ways of living with them.
The more interesting consequences of AI may therefore not be technological at all.
If AI makes human interaction scarcer, communities that deliberately protect it may become more valuable. If knowledge becomes enclosed, institutions committed to sharing it may acquire new importance. If conventional career ladders collapse, new systems of apprenticeship and learning will have to replace them.
Perhaps even new religions and forms of community will emerge: institutions designed to preserve what people value in older human societies while making selective use of the extraordinary technologies now appearing around us.
The future of work will not simply depend on what AI can do. It will depend on what humans can do collectively that even AI cannot yet dream of.
This article was edited with the assistance of ChatGPT.
