If a Human Wants to Do a Job, a Machine Shouldn’t Take It

In Part 1, I argued that the world we want is one where we can do anything we want, and that it takes two things: freedom and safety. Both, I think, can come from the same source: machine labor on demand.

This piece is about what that means, and about the one rule that keeps it from going wrong: if a human wants to do a thing, a human gets to do it.

Agency first

Let’s start with defining the value that everything else in this series is built on.

We must prioritize human agency as the paramount value we build towards.

In the world that we want, you must be maximally free to do what you choose to do. Not what is economically useful for you to do. Not what society asks you to do. What YOU CHOOSE to do.

But to do this, your basic needs must be met.

Normally, our needs are met by trading money to pay for them, that we earn from a job, working for a corporation.

Oftentimes, the work we can get to cover our needs is not the work we choose.

It is either work that is commercially useful in the eyes of a corporation willing to hire us, or work that is necessary to do to keep our civilization running.

Wait a minute, you might wonder: How does society not collapse around us if everyone only ever does what they choose to do and not what society needs them to do?

Well, we need to have systems in place that can do the things people don’t want to do, available on demand.

Machine labor on demand

The world we want is one where we can do this with intelligence and labor. Intelligence, in the form of machine intelligence. Labor, in the form of machine labor.

Right now, a significant load of digital intelligence can be called on demand, thanks to cloud-based AI models… if you’re willing to pay the price.

The same is not true of labor today, but eventually, it will be.

With the rise of humanoid robots, and the combination of these humanoid robots with machine intelligence, we’ll soon have intelligent bodies (robots) we can call on demand.

These could be robots you rent, or robots you own, or if we produce enough of them, surplus robots that can be deployed anywhere and everywhere at any given time for any given need.

If these intelligent bodies are general purpose enough, they will be able to fill the demand for all kinds of work.

Wherever we have a labor shortage, we’ll be able to deploy these machines to plug these gaps in demand.

Put intelligence and intelligent bodies together, and you get what I’m calling machine labor on demand.

Its inputs are physical parts (the motors, actuators and bodies), energy, and machine intelligence. Its output is labor: any kind of work that needs doing, from carrying a crate to reading a health scan.

And everyone should be able to call on it. That’s what I mean by universal basic machine labor: machine labor on demand, for all.

Everyone gets a base: a personal machine of their own, with personal machine intelligence and personal machine labor linked together.

Above that, anyone can request more labor for their quest, their guild or their community (I’ll explain those in Part 4), and requests are met by demand.

Essential infrastructure need is the first tier of that demand, so the essentials always come first.

The two markets we live in

Here’s why this matters: there are two markets we all live in, that drive what we do.

The first is the market for labor. On one side is the work that needs doing. On the other are the people willing to do it.

When there aren’t enough people willing to do a job, one of two things happens: wages rise until someone takes it, or the job doesn’t get done.

The second is the market for goods and services. On one side is everything we want and need. On the other is everything we can actually produce.

Therefore, what we can produce is limited by the labor and materials we have available to work with.

Labor and materials are two kinds of resources. When this series talks about sharing resources, here’s the full list I mean:

Materials: the raw stuff, like ore, timber, water and seed.

Energy: the power that runs everything else.

Land and space: somewhere to build, grow or store things.

Labor: work done by people, or by machines on demand.

Built things: tools, machines, buildings and devices. They’re made from the other resources, then become resources themselves.

Information: data, designs, code and know-how.

Time and attention: the hours and focus people choose to give.

Reputation: your credibility with your peers, and your track record of getting things done.

Money: a stand-in for the value of everything above. It’s useful for trading, and its value comes from the resources it stands in for.

Machine labor on demand changes the supply side of both markets at once.

Wherever there’s a gap between the work that needs doing and the people willing to do it, machines fill it. And every gap filled means more goods and services get made.

Build for peak demand

Think of machine labor on demand working like the power grid: it has to be built big enough to meet peak demand on the most demanding day of the year.

Build machine labor on demand to that standard, and on every other day, everything below the peak is surplus.

Surplus that can support universal basic goods and services, for example.

And it isn’t only a surplus of goods. It’s a surplus of labor too: machine hours nobody is using, ready to point at whatever we decide is worth doing.

And we're going to need it.

Something has to give

Once machines can do most of the work, companies have less and less reason to pay humans to do it.

But your rent doesn't go down with your wages. And neither does your grocery bill.

So when what you can earn keeps falling…

And what you need to spend doesn't…

Something has to give.

Multiply that across millions of people…

And you get a serious upheaval of the social order as we know it today.

So what do we do about this?

Build toward abundance

We build toward abundance: enough universal basic goods and services that everyone’s needs are met without needing a job to earn them.

As a thought experiment: what if we didn’t actually need “money” to trade with one another?

What if instead, we could produce enough of what we need that anyone could walk into a store and take what they need, and we’d still have enough to go around?

And not just “enough.” More than enough. Like everything we have today, and then some.

To enable this, we need to pool the excess machine intelligence and machine labor we possess into producing enough of what we want and need so that we can all have some.

Why machines shouldn’t do everything

Now, that might sound like a world where machines do everything for us.

It shouldn’t be.

A world where machines do everything might seem comfortable, at first, but it would make us spectators in our own lives. Living ghosts drifting through a world inherited by zombie workers that we no longer belong in.

More importantly, it would lead to what venture capitalist Sam Lessin calls a meaning crisis, which I write more about in Part 8.

So what do we actually want, instead of machines doing everything for us?

The freedom to do whatever we want to do.

There’s a brilliant novel by Ursula K. Le Guin called The Dispossessed. In this book, the people of an anarchist colony on a moon called Anarres speak a language that uses the same word for work and play.

Work there is simply what you do because it matters to you.

That’s what maximum agency looks like.

And it’s why machine labor on demand has to work differently from most visions of the future.

Humans first, machines fill the gap

Most post-AGI scenarios imagine robots and AI agents filling the roles humans fill today, inside companies, 24/7. Human labor replacements.

Machine labor on demand should be the opposite: less a replacement for people, and more a resource we ratchet up or down based on what we want to do ourselves, versus what we just need done.

Go back to that labor market. In the world we want, humans are always first in line, and machines fill whatever gap is left.

More service is needed? Flex machine labor up.

Do humans in that community have that job covered today? Ratchet it down.

Think of human roles in this context as anything from something as impactful as working on the machine-labor system itself, to cooking a meal at home for yourself and others.

A simple example: do you need a chair?

Want to build the chair yourself? Do it.

Don’t want to? A machine can.

Now play that out across everything that needs to get done, anywhere in the world, at any given time. All the work that no one wants to do at any given time across all of society gets handled the same way.

That’s the principle the rest of this piece is built on: if a human wants to do it, a human gets to do it.

Because we humans will ALWAYS want things to do. It might not be the same thing every day, again and again. It might be many different things, on many different days. But we’ll always want to be doing something.

So shouldn’t we build a world built around that obvious, deeply human need? A world where on days we want to work and do something meaningful, we can. And on days we don’t feel like it, we don’t have to?

We can build that very world, so long as everything else is covered by machine labor on demand.

Human-centered design

We already have a model for how to build systems like this. It’s called human-centered design.

The cognitive scientist Don Norman made the case for it in 1988, in a book first called The Psychology of Everyday Things and later retitled The Design of Everyday Things.

His argument was simple: when people struggle to use something, the fault usually lies with the design, not the person.

You know the door you pull when you’re supposed to push? There’s a name for it now: a Norman door.

Norman gave designers a vocabulary for this. An affordance, a term the psychologist James J. Gibson coined in 1966, is what an object lets you do: a chair affords sitting, a handle affords pulling.

A signifier, which Norman added in the book’s 2013 edition, is the cue that tells you what to do, like the flat plate on a door that says push.

And usability is how easily people can actually get done what they came to do.

The international standard for it, ISO 9241-210, describes human-centered design as making systems “usable and useful by focusing on the users.”

Now put the technical definition of affordances and usability aside, and just think of the actual words being used: human. CENTERED. design.

Meaning to design with humans at the center of it all.

That’s what I’m advocating for. Not human-centered design at the cost of machine labor on demand:

Machine labor on demand designed around human will in the center.

So let’s bring those three terms back, this time for machine labor on demand itself.

The affordance is agency. A chair affords sitting. Machine labor on demand affords choosing: you work on something when you want to, and machines do it when you don’t.

The signifier is the system that coordinates demand. Like the plate on a door, it shows what needs doing, where, and right now, so necessary work gets covered and handed off the moment it’s needed.

Think of a decentralized power grid, routing resources to wherever the load is, but for everything we do, everywhere around the world.

Usability means access for everyone. It’s affordable, because it’s built to the scale of surplus, and it’s easy to use. Calling on it should feel as natural as any tool or service in your life today, and more so.

The economics of machine labor

Machine labor on demand answers two problems at once: labor shortages, and an economy that can’t produce enough of what people need.

Wherever there’s work that needs doing and not enough people willing to do it, machine labor on demand fills the gap. I’ve already made that case, so you get the picture.

Here’s what it does to the economy. Every unmet need that a machine can now fill becomes a transaction that wasn’t happening before. More gets made, more gets bought, and money changes hands more often.

That only counts as a good thing if people still have money to spend, which they won’t. But that’s another problem we’ll get to.

There’s a second effect. When labor stops being scarce, services get cheaper to provide (goods come next, in Part 3). As many have written, this would be a massively deflationary event.

Why? Because competition pushes prices toward the cost of producing one more unit.

As machines do more of the producing, that cost falls toward the price of the energy and materials involved. Make those abundant too, and the cost of almost everything keeps falling.

Follow that all the way down, and you get a market so efficient that the price of anything machines can make heads toward zero.

When there’s enough of something for everyone, nobody can charge extra because it’s scarce.

Eliminate scarcity, eliminate price.

Whatever does manage to stay scarce keeps a price... until we make enough of it to be free, too.

Now the catch. If machines can do the work humans do, companies stop needing to pay humans to do it. And people who aren’t earning wages can’t afford whatever isn’t provided for free.

So prices fall, but so does the amount of what people can afford to buy, both at the same time. So we might as well make everything free, right?

So instead of today’s world, where corporations compete to make their stock prices as valuable (high) as possible...

Should we actually just build a world where corporations compete to make the production of their goods and services so efficient (low) that they become free?

Maybe. But first, let’s analyze how the people building this AGI-driven future plan to handle it.

What the AI industry plans instead

The AI industry spends most of its energy on explaining how we get to machine labor on demand. It rarely says what we should do with it once we have it.

Its favorite story is recursive self-improvement: AI that helps build the next AI, which builds a better one still, until you reach what the mathematician I.J. Good called an “intelligence explosion” back in 1965.

The most detailed, recent, and credible plan for managing that loop is AI 2040: Plan A, from the AI Futures Project, which is also the team behind AI 2027. It’s a recommendation, not a prediction.

In the scenario they imagine, the US and China should strike a deal that caps AI around top-human-expert level, pause there while safety research catches up to the frontier of model capabilities, and only resume the climb toward superintelligence around 2040.

Even that “slow” path is wild. Robots multiply, growth explodes, and the bottleneck shifts from ideas to atoms: someone still has to mine the ore, build the motors, and connect the power. By 2040, the scenario has the share of American adults with a job falling to about 12%.

Notice how Plan A takes care of people: it hands them money. Governments auction permits to build scarce compute and robots, and most of the proceeds flow to every American adult as a dividend that climbs toward $1 million per person a year by 2035.

That’s essentially Elon’s prediction of universal high income. But let’s get serious for a second: if everyone is getting $1 million per year by 2035, isn’t that just pretend money?

What stops prices from rising to meet it, until we end up in a Weimar Republic kinda situation? I’ll come back to this in a moment.

It’s not a bad idea. In fact, I think it makes a lot of sense. But it still doesn’t prepare us for a future of post-scarcity abundance.

What the authors worry about is the right thing: control of AI concentrating in too few human hands.

But a dividend is still a check. And a check only works if there’s enough of the things it buys.

The problem isn’t wages. A dividend can replace those. The problem is the things themselves.

Hand everyone a million dollars, and the price of anything still scarce rises to meet it. Even in the world they predict, land is still land.

More money can’t get us there. Only more of the things can.

So here’s my point: whatever we build should serve humanity. That means machine labor on demand pointed at the things people actually need, and owned in a way that keeps it working for everyone.

Two problems stand in the way. Some things are scarce no matter how much labor we throw at them. And someone still owns the machines.

That’s where we go next, in Part 3: Point the Machines at What’s Scarce.