When Machines Do the Work: Automation and AI, Threat or Liberation
Personal, Interpersonal & Societal
Now
In a currency-based economy, most people's income comes from a job, and that income is what stands between them and losing their home, their healthcare, and their food. So a machine that can do a job is not simply a useful tool. For the person who holds that job, it is a threat to survival. Not every new machine eliminates work, and many change jobs rather than end them. But the risk is real, and so is the dread that comes with it.
Car manufacturing shows the pattern clearly. The auto industry has long been the largest single user of industrial robots: in a Brookings analysis using US data available in 2017, it employed nearly half of all industrial robots in use, concentrated in places like Detroit and Toledo. A highly automated plant needs far fewer people on the assembly line. Economists Daron Acemoglu and Pascual Restrepo, studying US labour markets from 1990 to 2007, found that each additional robot per 1,000 workers reduced the employment-to-population ratio by about 0.2 percentage points and wages by about 0.42% (Journal of Political Economy, 2020). In the areas where robots were deployed, that works out to roughly 6.6 jobs lost per robot locally and 3.3 nationally, once gains elsewhere are counted.
Automation keeps spreading. Factories worldwide installed 542,000 industrial robots in 2024, more than double the number ten years earlier, bringing the total in operation to about 4.66 million, according to the International Federation of Robotics. The global average is now 132 robots per 10,000 manufacturing employees; South Korea has 1,220, Germany 449, and the United States 307. And artificial intelligence now extends the same pressure from the factory floor to offices, clinics, and studios. An IMF staff study estimated that almost 40% of jobs worldwide, and about 60% in advanced economies, are exposed to AI. The authors judge that roughly half of the exposed jobs could benefit through higher productivity, while the other half could see lower demand for labour, lower wages, and fewer hires.
When a plant automates or closes, the loss does not stay with individual workers. It reaches whole communities. A study published in Demography examined US manufacturing counties built around auto assembly plants and found that rising automation from 1993 to 2007 was linked to higher death rates among working-age adults, including deaths from drug overdoses and suicide; the authors estimate that automation accounts for about 12% of the increase in overdose deaths among working-age adults over that period. This is what it looks like when a livelihood, and with it a sense of purpose and standing, disappears and nothing takes its place.
Who gains from the higher productivity? The answer depends on who owns the machines. Acemoglu and Restrepo estimate that between 50% and 70% of the changes in the US wage structure over four decades came from relative wage declines among workers whose tasks were automated. Karabarbounis and Neiman found that the share of national income going to labour has fallen worldwide since the early 1980s, and attribute roughly half of that decline to cheaper capital equipment, much of it computing. These conclusions are debated. Economists at the Economic Policy Institute argue that policy choices and employer power, not automation, explain most of the growth in inequality, and research by David Autor and colleagues shows that most of today's US employment is in kinds of work that did not exist in 1940. New work does appear. But the people who lose a job are often not the people who get the new ones, and the gap in between still has to be survived on whatever savings they have.
None of this means automation is bad. Its benefits are real: machines take on heavy, hot, toxic, and repetitive work, and greater robot use has been linked to fewer workplace injuries in the US and less work-related disability in Germany. Output rises; goods become more plentiful. But under the current arrangement, those benefits flow mainly through ownership. The owner of the robot receives the gain, and the worker it replaces receives the risk. That is why a technology that could lighten everyone's load is reliably experienced as a threat by the people whose work it does.
After
When survival, livelihood, and quality of life no longer depend on holding a job, the same machines mean something entirely different. Homes, food, healthcare, education, and the materials for a good life are open to everyone, at no cost, whether or not a given task is still done by a person. A robot that takes over an assembly line no longer takes anyone's livelihood with it. It takes the work, and returns the time.
That changes what people want from automation. Today, a worker has every reason to fear a machine that could do their job. Afterward, a community has every reason to want machines to take on the repetitive, dangerous, exhausting, and unfulfilling work first, including much of the work nobody wants to do. What cannot be automated is shared in short rotations, so no one is permanently assigned to it (see Honest Questions and Answers on jobs nobody wants). Every task a machine reliably handles can reduce the net labour people must give, once maintenance, supervision, energy, materials and ecological effects are counted — automation changes the composition of work more than it simply erases it.
The gains from higher productivity no longer go to whoever owns the machine. They show up where they belong: as shorter hours for everyone. The project's projections already estimate a median work week of roughly 15–17 hours once work that exists only to move money, and production built in to wear out, is removed. Automation and AI can shorten it further. The hours returned go to what a job so often crowds out: caring for children, parents, and each other; craft and making; learning at any age; community life; art, music, science, and play.
This is why, after the transition, the benefits of automation and AI can be made to outweigh their harms — when the technology is democratically controlled, ecologically justified, independently tested, and always subordinate to human judgment. The harm described on the left of this page does not come from the machines themselves. It comes from tying livelihood to employment, so that every gain in efficiency is someone's lost income. Remove that tie, and a robot in the auto plant is simply a robot in the auto plant: something that makes cars with less human strain. No one is left without a livelihood because a task became easier.
Some real problems remain, and they deserve honest attention. Automation and AI need energy and materials. Data centres already used about 415 terawatt-hours of electricity in 2024, a figure the International Energy Agency projects could more than double by 2030, and every robot and server carries metals that have to be mined and, at the end of its life, becomes electronic waste. A post-transition society has to weigh these demands across the whole lifecycle — extraction, manufacture, operation, repair and recycling — against ecological limits, and some automation will not be worth its footprint. Nor may the burdens simply be exported: mining, e-waste and the remaining unwanted work cannot be pushed onto poorer regions so that wealthier ones enjoy the benefits. Humans also have to keep meaningful oversight. AI is a tool that works under transparent, democratic rules, with open models people can inspect and override. It is never a sovereign decision-maker (the Transition Roadmap takes the same position on AI in coordination: a forecasting and matching tool, never an allocator). And some work should stay human by choice. Caring for the sick and the dying, teaching children, and making art are not merely tasks to be completed. For many people, they are the point.
Safeguards make this real rather than hopeful. Automation is adopted only when the people doing the work, the people using the resulting service, and the affected communities determine that it produces a net human and ecological benefit. Essential systems require independent testing, open documentation, auditable operation, privacy by design, human override, manual fallback, cybersecurity protections, and safe shutdown and recovery procedures. No community becomes dependent on a single opaque model or automated system, and manual skills are kept alive so that people can still run things when machines fail. Technical expertise carries responsibility, not permanent authority: the people who design or maintain a system receive no additional governing power over those who depend on it. Questions about whether some future AI systems may warrant moral consideration are separate from the question of authority — no AI system, sentient or otherwise, receives unreviewable power over another being’s liberty, necessities, or participation in society.
Some of the most valuable automation is the kind that gives people more independence rather than less involvement: assistive technology that lets an older person stay in their own home, a wheelchair user travel without waiting for help, or someone with limited sight read anything they choose. Here machines do not replace a person’s role; they widen it.
People whose roles are automated will still need support, even when their livelihood is secure. Losing a familiar role can unsettle a person's sense of identity and purpose. So adjustment is supported: counselling for anyone who wants it, social engagement and connection, help discovering passions and aptitudes, and education and training open to all at no cost, with no pressure and no deadline. And during the transition itself, displaced workers still live under the current system, where a lost job can still mean a lost home. The transition must protect them first, not ask them to absorb the shock on behalf of everyone else. For more on how AI fits the transition, see AI and the Transition and the book chapter Chapter 20: Capitalism and Automation.