Make Light Work

Artificial intelligence can now do a number of things I cannot. It can write software faster than me, translate languages I do not speak, read a hundred-page report without losing the will to live, and explain in detail how to repair almost anything in my house.

It cannot repair the house.

Someone still has to climb the ladder. AI has made some kinds of thinking cheaper, but physical work still costs roughly what it did before. A computer can diagnose the problem, find the part and tell you what to do. Getting somebody round on Thursday still costs what it costs.

Physical AI is the attempt to close that gap: AI that can perceive the world and act in it through a robot or another physical system. Robots themselves are not new. Factories have used them for decades. What would be new is a robot that can face a task it has not been programmed for step by step, work out what needs doing and do it.

I do not think the moment that matters will be a humanoid robot doing a backflip at a conference. It will be when you can send one to do something useful on a wet Tuesday morning and expect the job to be done. A field needs harvesting, a warehouse unloading, a power line inspecting, an older person helping with the shopping. Most physical problems still end somewhere along the line with a person doing something with their hands.

If that changes, a lot changes with it.

Intelligence Gets Hands

We have been trying to reduce the amount of human labour required to live for about as long as we have been inventing things. Animals pulled loads. Water turned mills. Steam moved ships and trains. Electricity spread power into homes and factories. Tractors allowed a small number of people to farm land that once required many more. Washing machines removed hours of domestic work without causing much nostalgia for Monday morning with a washboard.

Each wave changed what human effort was worth and where it was needed. Industrialisation made physical power cheaper. Computers made calculation cheaper. The internet made information cheap to move. Generative AI is lowering the cost of some kinds of cognitive work. Giving AI a useful body would extend that into the physical world.

Robotics has always struggled once the environment stops cooperating. In 2017 a Knightscope security robot called Steve began patrolling Washington Harbour in Washington, DC. Five days later it ended up in a fountain. The robot was not programmed to handle ground that moved; unsecured bricks had been enough to defeat it.

Factories avoid much of this by arranging the world for the robot. The same part arrives in the same place, facing the same way, thousands of times. Outside the factory, objects move, people get in the way, bags collapse when you pick them up and doors are left half open. Humans adjust to this without thinking about it. Robots have historically needed the world to be much more polite.

That is beginning to change. Vision-language-action models can take visual information and instructions and turn them into robot actions. Human demonstrations can become training data. Simulation can provide more. Skills learned in one setting may transfer into another instead of being programmed again from scratch.

None of this means we have a general-purpose robot. I do not know whether robots that can handle a wide range of ordinary tasks arrive in ten years or thirty. The progress is enough to take the possibility seriously.

If it works, the effect goes well beyond replacing one worker with one robot. Labour sits inside nearly every physical thing we consume. Food is planted, harvested, processed and delivered. Products are assembled, packed, transported, inspected and repaired. Roads are built and resurfaced. Hospitals, hotels, warehouses and homes consume people’s time simply to keep functioning.

If robots can do more of those jobs, production gets cheaper when the labour saved is worth more than the machines cost to buy, power and maintain. Lower production costs do not automatically mean lower prices. Competition matters. A company that can make something 20 per cent more cheaply does not contain a tiny mechanism forcing it to charge you 20 per cent less.

There is still plenty of room for costs to fall. Rich countries need more energy, more infrastructure, more care for ageing populations and more repair of what they already have. Housing shows both the opportunity and the limit. If machines can excavate, frame, wire, plumb and finish buildings with fewer human hours, construction should get cheaper. That does not make a flat in central London cheap. The land is still in central London. Planning rules still exist. Roads, schools and utilities still have to cope with whatever gets built.

Automation can remove one constraint and reveal another.

The timing will not be convenient either. A company can reduce its wage bill years before cheaper housing or energy makes much difference to the employee whose bargaining power has just fallen. Weak competition can delay the benefit further.

The theoretical abundance may be coming. The direct debit is coming on Friday.

The cost of work also changes form rather than disappearing. A business that once paid people to perform a task starts paying for robots, chips, electricity, software, maintenance and finance. Labour becomes a smaller part of production and capital becomes a larger one. That changes who receives the income.

Who Owns the Work?

Imagine two societies with exactly the same robots.

In the first, goods and services get cheaper, working hours fall and small companies can rent capabilities that once required large organisations. Households own enough of the productive economy that rising returns to capital are shared beyond the people who were wealthy before the robots arrived.

In the second, the technology works just as well, but most of the robots, data centres, energy systems and software belong to a small number of firms and investors. Wages become a weaker way of sharing in growth because less of the work is being done by people. Consumers can buy the cheap things the automated economy produces, but they own very little of the economy producing them.

The second society may be extremely rich on paper. I am not sure it would feel especially free.

We already spread ownership through pensions, retirement accounts and ordinary investment. The problem is that this works best for people who already own assets. Someone who is twenty-five, renting and only beginning to save has little claim on an economy in which returns to capital suddenly matter much more.

If physical AI shifts income from labour towards capital, more people need a stake in the capital. That could come through wider individual ownership, employee equity, citizen wealth funds or other arrangements. There is no settled answer. But a society in which machines do more of the work while ownership remains narrow is likely to have a distribution problem even if everything in the shops is cheap.

Ownership will not solve shortages that have other causes. If robots can build houses cheaply but planning rules stop houses being built, planning remains the problem. If electricity is cheap to generate but the grid cannot carry it, build the grid. If a few firms can keep most of the gains because competitors cannot reach the market, competition policy matters too.

Income support may be needed while all this changes. People cannot pay today’s bills with tomorrow’s productivity. But helping someone through disruption is different from building a permanent system in which most people live on transfers funded by an economy they have little ownership of. The latter may provide an income without providing much control.

That control seems a better test of whether automation is working for people. Can you leave a job you hate? Afford somewhere decent to live? Start something? Care for somebody? Take a risk? Spend several years getting good at something that nobody currently pays much for?

A country can become more productive without making any of those things easier.

What Was Work For?

Even if the economics works, employment still leaves another hole to fill.

A job gives the week a shape. It puts you around people you did not choose, which is occasionally awful and probably quite good for us. It gives you problems, deadlines and things to improve at. It gives you something to say when somebody asks what you do.

It also lets people earn respect by being useful. You can be a good electrician, nurse, teacher or engineer and be valued by people who know almost nothing else about you. In a society where status often comes from money, family or social connections, competence gives people another route.

In 1930 Keynes divided human wants into two types. Some are absolute: we want food, shelter and comfort regardless of what anyone else has. Others are relative: we want them because having more makes us feel ahead of somebody else. Keynes thought the first category could eventually be satisfied. The second could not, because the comparison never ends.

That distinction looks useful in an automated economy. Cheap goods do not abolish status. Land remains limited. Attention remains limited. There will still be a nicest neighbourhood, a most selective school and a table everyone wants at the restaurant. If everyone becomes richer while the number of houses overlooking Central Park stays the same, people can simply spend more money competing for the same houses.

So a richer society can still have envy, hierarchy and political fights. Some scarcity comes from the physical world. Some comes from wanting what other people cannot also have.

I am less worried about people having nothing to do. We already spend huge amounts of time doing things nobody requires of us. People run marathons, learn instruments, coach teams, build companies, volunteer and become alarmingly serious about recreational cycling. Some of these things are enjoyable. Some seem designed mainly to make Sunday morning more difficult. Either way, people do not need an employer before they can find reasons to exert themselves.

The harder part is replacing what employment currently supplies besides money: structure, company and a sense of contribution. A society in which paid work takes up less of adulthood would need more places for those things to happen. Families, clubs, sport, education, local institutions and voluntary organisations would matter more. We would also need to get better at respecting forms of contribution that do not come with salaries attached.

We already understand this in parts of life. A parent caring for a child is doing something useful whether or not anyone pays for the hour. A friend who turns up when things are going badly matters in a way that never appears in national accounts. Someone can spend years mastering something with almost no commercial value and still have spent the years well.

The same is true of work we choose to keep doing after machines could do it for us. A factory can make a table more cheaply, yet a table somebody made for you may matter more. Recorded music is abundant, yet people still crowd into rooms to hear it performed. A robot may eventually cook dinner perfectly. That does not make somebody cooking dinner for you pointless.

Who Gets the Time Back?

For most people who have ever lived, necessity decided a large part of the day. Food had to be produced. Things had to be carried. Buildings had to be built. Clothes had to be washed. Money had to be earned because almost everything we wanted required somebody else’s labour somewhere along the chain. Technology has been reducing that burden for centuries, although never evenly and never without disruption.

Physical AI could push much further. It could make the physical world easier to build, move, maintain and repair with fewer human hours. That would be a huge gain. But the gain is not just cheaper goods or higher output. It is the time no longer required to produce them.

I am more interested in who gets that time back.

If the machines are owned broadly, if lower costs reach people, and if we stop treating paid employment as the only serious use of an adult life, the result could be more time for family, care, learning, building things, taking risks, creating, competing, or doing nothing at all.

That is the version of abundance worth aiming for. Not a society where nobody works, and not one where machines do everything while people wait for an allowance. A society where less of life is dictated by what has to be done simply to keep things running, and more of it can be spent on what we decide is worth doing.

The machines may make the work lighter.

We should make sure the time they save becomes ours.