Why Are We Building Robots in Our Own Image?

For most of the history of robotics, the human body looked like a terrible engineering choice.

Two legs are harder to balance than four wheels. Hands are extraordinarily complicated compared with mechanical grippers. Human joints create unnecessary complexity. If the job is to weld a car, move a pallet, cut steel, harvest wheat, or fly through the air, there are far better shapes than ours.

That is why the machines that transformed industry rarely looked anything like people. They became forklifts, tractors, robotic arms, CNC machines, aircraft, conveyors, drones, and countless other specialized devices.

The machine was designed around the task.

Now something strange is happening.

Some of the most ambitious robotics programs in the world are working in the opposite direction. Instead of designing the ideal machine for one task, they are trying to build machines with two legs, two arms, hands, roughly human proportions, cameras where eyes might be, and enough intelligence to move through the world much as we do.

The obvious question is: Why?

The answer may be that the human body is not the best shape for the job.

It may be the best shape for our world.

We Already Built the World for Humans

Walk around your house and look at it as an engineer.

Door handles are positioned for human hands. Stairs assume human legs. Countertops assume human height. Drawers, cabinets, appliances, light switches, steering wheels, pedals, tools, chairs, ladders, sidewalks, vehicles, warehouses, and factories were all designed around approximately the same physical specification.

Us.

We have spent thousands of years shaping the physical world around the capabilities and limitations of the human body.

Traditional automation had to work around this problem. If a machine did not fit the environment, we redesigned the environment. We installed conveyors, cages, rails, special fixtures, dedicated work cells, and highly controlled production lines.

That works beautifully when the task is repetitive enough to justify the investment.

But millions of jobs cannot easily be automated that way because they happen in messy environments created for people.

A humanoid robot offers a different solution.

Instead of rebuilding the environment for the machine, build a machine that already fits the environment.

That is an enormously important shift.

A wheeled robot sees a staircase as an obstacle requiring another engineering solution. A humanoid sees stairs.

A specialized machine may require a custom end effector. A humanoid with sufficiently capable hands may pick up the screwdriver already sitting on the bench.

A robot designed specifically to walk a dog might be better at walking dogs than a humanoid robot. But when the walk is finished, that machine still walks dogs.

A humanoid might come home, unload the dishwasher, carry a basket of laundry upstairs, take out the trash, move a box in the garage, and perhaps eventually operate the same automobile that you do.

That is the real attraction of the human form.

It is not necessarily optimal.

It is general purpose.

The humanoid robot may prove to be a mediocre solution to a thousand individual tasks while becoming an extraordinary solution to the problem of performing a thousand different tasks.

The Human Body Has Become an Interface

Software engineers understand interfaces.

An interface does not have to be the internally perfect implementation. Its value comes from compatibility.

In that sense, the human body may already be the most widely implemented physical interface on Earth.

Civilization supports it everywhere.

Doors support it. Tools support it. Cars support it. Buildings support it. Factories support it. Kitchens support it. Construction equipment supports it.

A humanoid robot potentially gains access to all of that without requiring civilization to be redesigned first.

We have accidentally created the world's largest robot interface.

And that leads to a more uncomfortable observation.

If the human form is useful because it allows machines to occupy the spaces humans currently occupy, then humanoid robotics is not merely another form of automation.

It is particularly well suited to replacing human labor without replacing the surrounding infrastructure.

For decades, automation often meant redesigning the workplace.

Humanoid automation may eventually mean replacing the worker.

Artificial Intelligence Changed the Equation

There is a reason this idea has become more compelling now.

Mechanical engineers have been able to build impressive robots for decades. The more difficult problem has been getting them to function outside highly controlled environments.

The real world refuses to cooperate with traditional automation.

The box is six inches farther left than expected.

Someone leaves a cart in the aisle.

A door is half closed.

The tool is missing.

A package is upside down.

A child runs across the sidewalk.

Humans deal with these exceptions so naturally that we barely recognize how sophisticated the process is. We perceive, reason, adjust, improvise, and continue.

Traditional robots are extraordinarily good when the world behaves exactly as expected.

Humans are extraordinarily good when it does not.

Modern artificial intelligence is beginning to attack that gap.

If perception, language, reasoning, memory, and learned physical behavior can be connected to capable robotic bodies, the humanoid becomes far more valuable. The machine no longer needs every movement programmed in advance. It can increasingly interpret what is happening around it and determine what to do next.

And that is where the story stops being simply about robotics.

AI is attacking cognitive work at the same time robotics is attacking physical work.

We should at least ask where those two trajectories eventually meet.

What Happens to the Work We Do Not Want?

One of the easiest arguments for robotics is that machines will perform jobs people do not want to do.

That sounds wonderful.

Let robots clean sewers, work in dangerous factories, unload trucks, harvest crops, move heavy objects, clean floors, stock shelves, and perform repetitive work.

Let machines do the dangerous jobs.

Let machines do the boring jobs.

Let machines walk the dog.

But there is a flaw hidden inside the phrase jobs people don't want to do.

People may not enjoy a particular job, but they often want the income that comes from doing it.

If someone dislikes driving a truck but earns a living driving that truck, eliminating the job has not necessarily liberated that person.

We have eliminated the market value of one of the things that person knows how to do.

For generations, technological progress followed an implicit bargain. Machines eliminated some forms of labor, productivity increased, new industries appeared, and workers moved into new jobs.

That may continue.

But increasingly capable general-purpose AI and robotics force us to consider another possibility.

What happens when the machine that eliminates one task can also learn the next one?

The old assumption was that humans would simply move up the ladder.

We should at least consider what happens if the ladder itself starts moving.

And What About the Work We Do Want?

There is an even stranger aspect to this revolution.

Automation was supposed to free humans from drudgery.

Yet some of the first areas transformed by generative AI have been writing, illustration, music, software development, design, photography, and other activities people often choose to do because they find them intellectually or creatively rewarding.

Meanwhile engineers are working hard to automate cooking, driving, construction, warehouse work, delivery, manufacturing, and household chores.

This creates a peculiar possibility.

Machines could eventually become very good at both the things we hate doing and the things we love doing.

So what exactly are we saving for ourselves?

That question deserves more attention than it receives.

Why Are We Teaching Robots to Fight?

There is another side to creating machines that share our physical capabilities.

The same generality that makes a humanoid useful in a warehouse also makes it potentially useful in environments occupied by soldiers.

A specialized weapon does not need a human form. Tanks, missiles, drones, and aircraft demonstrate that clearly.

But a humanoid has another advantage.

It can potentially go where humans go.

Through doors.

Up stairs.

Into buildings.

Across terrain created for people.

It can potentially carry equipment created for human bodies and manipulate tools created for human hands.

Much of what looks like teaching robots to fight is also legitimate research into balance, recovery, rapid movement, collision handling, perception, and responding to unpredictable forces. Those capabilities could make robots extraordinarily useful in disaster response, rescue work, construction, and dangerous industrial environments.

But capabilities are not moral.

The ability to recognize a moving person, pursue a target, maintain balance while being struck, manipulate objects, and make autonomous decisions can be useful in both rescue and combat.

Eventually, sufficiently powerful autonomous machines will possess more than enough physical capability to kill a human being.

We should not have to imagine rebellious science-fiction robots to find that concerning.

The much nearer question is what happens when humans intentionally give autonomous machines objectives that can result in physical harm.

Asimov's Warning Was Really About Specifications

Isaac Asimov famously imagined his Three Laws of Robotics as fundamental restrictions on robotic behavior, beginning with the principle that a robot should not harm a human being.

What is often forgotten is that his stories repeatedly explored how difficult such rules become in practice.

What constitutes harm?

What happens when preventing one person's harm requires restraining another?

What happens when instructions conflict?

What happens when a machine interprets an objective differently than its designer expected?

Consider a simple command:

Protect this person.

That sounds reasonable until another person becomes a threat.

Or:

Do not allow anyone into this building.

How much force may the robot use?

Or:

Stop that person.

What does stop mean?

Humans interpret such instructions inside vast systems of law, morality, experience, proportionality, cultural expectations, and common sense.

The instruction contains only a few words.

The actual specification is enormous.

Asimov may therefore have been writing about something software engineers understand very well.

Requirements.

His robot stories were, in many ways, stories about incomplete specifications interacting with powerful systems.

That problem becomes much more consequential when software leaves the screen and acquires arms and legs.

What Must Never Be Allowed to Change?

Software engineers increasingly talk about AI as a way to generate implementation.

That shifts human responsibility upstream.

The important work becomes defining the objective, decomposing the problem, identifying invariants, determining acceptable behavior, and explicitly stating what must never happen.

With ordinary software, an incomplete specification might produce an incorrect report or corrupt a database.

With autonomous physical machines, specifications can eventually govern behavior in the real world.

That makes certain questions extraordinarily important:

What may the machine do?

What may it never do?

When must it stop?

When must a human take control?

Who is responsible when its interpretation differs from ours?

How do we guarantee that safety constraints cannot simply be overridden by another objective?

Those questions cannot be treated as afterthoughts.

They may ultimately be more important than the code itself.

If Machines Do the Work, What Are Humans For?

Eventually all these questions converge on a larger one.

Suppose the optimistic vision succeeds.

Machines perform dangerous work.

They perform repetitive work.

They manufacture our goods.

They transport them.

They maintain our infrastructure.

They clean our homes.

Artificial intelligence performs increasing amounts of analysis, administration, programming, accounting, writing, and design.

Productivity becomes extraordinary.

Human civilization could become wealthier than anything previous generations imagined.

Then what?

The easy answer is that humans would finally be free.

Free to create.

Free to raise families.

Free to study.

Free to travel.

Free to make art, conduct science, build strange inventions, explore the world, and spend time with one another.

That is an appealing future.

But our society currently connects the distribution of resources very closely to labor. Most people obtain housing, food, healthcare, security, and independence because somebody values their work enough to pay them.

If human labor becomes dramatically less valuable, greater productive abundance does not automatically answer the question of who receives that abundance.

And there is another problem that economics alone cannot solve.

People need to feel useful.

Work is not merely a mechanism for obtaining money. It is often where people experience competence, responsibility, accomplishment, structure, friendship, recognition, and the feeling that someone depends upon them.

We spend enormous amounts of time complaining about work.

But being unnecessary may prove considerably worse.

A civilization can theoretically become incredibly wealthy while leaving millions of its people wondering why they are here.

Perhaps That Is the Question We Should Be Asking

The humanoid robot is fascinating because it reveals something about ourselves.

We did not necessarily discover that the human body is the ideal machine.

We spent thousands of years building a civilization in which the human body is the ideal participant.

Now we are building machines capable of participating in it.

We gave artificial intelligence our language.

We gave it access to much of our accumulated knowledge.

We are teaching machines to reason about our world.

We are giving them our shape.

We are teaching them to use our tools.

We are teaching them to move through our homes and workplaces.

We are increasing their autonomy.

And yes, we are even teaching some of them behaviors that can be applied to violence.

None of this means catastrophe is inevitable. Robotics could eliminate enormous amounts of dangerous and degrading labor. AI and automation could create extraordinary prosperity. These technologies may become some of the greatest tools humanity has ever created.

But their development raises a question much larger than whether a robot can successfully walk the dog.

What happens when being human is no longer economically necessary for most of the things civilization needs done?

Perhaps the most important challenge of the age of artificial intelligence will not be teaching machines how to become more like us.

It will be deciding what remains uniquely worth doing because we are human.