The Physical World Gap
We talk a great deal about how quickly artificial intelligence is getting better.
Models grow more capable. Software appears in minutes. Knowledge can be copied at almost no marginal cost. An improved AI system can, in principle, reach millions of people overnight.
Out of that trajectory grows a compelling idea:
If artificial intelligence can eventually perform almost any cognitive task, and robots do the same for physical work, the cost of goods and services could fall dramatically. Some versions of this future go further still: work becomes optional, material abundance becomes the default, and money gradually loses much of its meaning.
As a technological argument, this is surprisingly coherent.
As a social argument, it is far more complicated.
Because between what technology can do in principle and what actually happens in the world, there is an enormous gap.
I call it: The Physical World Gap.
The Physical World Gap
The faster AI advances, the more decisive that distance may become.
Software scales. Reality has to be built.
The digital world has one extraordinary property:
It can be copied.
An algorithm written today can be deployed a million times tomorrow.
Software can be distributed in seconds.
A single model can serve millions of people at once.
One additional digital copy consumes almost no additional physical resources.
The real world works on entirely different terms.
A second house needs more material.
A second robot needs more motors, chips, sensors and energy.
A new data centre needs land, transformers, transmission lines and cooling.
A new factory has to be built.
Copper has to be mined.
Machines have to be manufactured.
Products have to be shipped.
And eventually they have to be repaired, recycled or disposed of.
AI can improve all of these processes.
It can optimise them.
It can automate large parts of them.
But it cannot simply compile matter.
Intelligence scales digitally. Matter does not.
That is the first dimension of the Physical World Gap.
It is far from the only one.
1. Resources: Intelligence Cannot Compile Atoms
Suppose we solve a large part of the intelligence problem tomorrow.
An AI designs a substantially better battery.
The design is finished within hours.
Then physical reality begins.
The battery requires raw materials.
Those materials have to be found.
Deposits have to be developed.
Mines have to be permitted and built.
Material has to be extracted, processed and refined.
It has to be transported.
Factories have to build production capacity.
And the finished battery still has to reach a customer.
A design can be produced in hours.
A mine cannot.
Even if robots eventually perform every single step, physical reality does not disappear.
The mine still has a maximum extraction rate.
The port still has a maximum throughput.
A ship still needs time to cross an ocean.
A factory still has a capacity ceiling.
A deposit still holds a finite quantity of material.
Robots do not eliminate these systems.
They automate them.
2. Energy: Intelligence Runs on Power
Almost every vision of a highly automated economy eventually converges on the same input:
Energy.
AI needs data centres.
Data centres need electricity.
Robots need electricity.
Factories need electricity.
Autonomous vehicles need energy.
Mining needs energy.
Materials processing needs energy.
Recycling needs energy.
If we genuinely intend to deploy billions of intelligent machines, we are simultaneously committing to an enormous expansion of energy infrastructure.
And again, two clock speeds collide.
AI can become substantially more capable within months.
A power plant cannot.
A grid cannot.
A transmission line cannot.
A transformer cannot.
Digital capability can compound, while energy infrastructure still has to be planned, permitted, financed, manufactured and built.
A future economy of abundance would therefore not only be an AI project.
It would likely be one of the largest energy and infrastructure programmes in human history.
3. Infrastructure: A Robot Needs a World It Can Work In
An intelligent robot on its own does not transform an economy.
It needs an environment.
Roads.
Power grids.
Connectivity.
Data centres.
Ports.
Warehouses.
Factories.
Water.
Charging.
Maintenance.
An autonomous system can be technically brilliant and still fail in an environment that was never designed for autonomous systems.
Having a robot complete a task inside a controlled factory bay is one thing.
Integrating millions of machines into a messy physical world that grew over decades is something else entirely.
So the decisive question is not only: how good will our machines become?
It is also: how quickly can we change the environment they operate in?
4. Supply Chains: Robots Do Not Remove Them
Picture a fully autonomous copper mine.
Robots drill.
Autonomous machines haul the ore.
AI runs the processing.
Autonomous trains move the material to port.
Robots load the ships.
Autonomous vessels carry the copper.
At the destination it is unloaded automatically.
Autonomous trucks bring it to the factory.
Robots process it further.
Human labour inside that chain could eventually approach zero.
But the supply chain still exists.
The mine can fail.
A port can be blocked.
A rail line can be damaged.
A grid can be overloaded.
A country can restrict exports.
A war can sever a route.
A material can become scarce.
Automation removes human labour from a supply chain.
It does not remove the supply chain.
5. Regulation: Capability Is Not Permission
Here a second form of physical reality becomes visible.
Technology does not exist outside our institutions.
Autonomous driving is the clearest example.
The vision has been clear for years.
Sensors perceive the environment.
AI interprets the situation.
Computers control the vehicle.
If the technology works well enough, a large share of road traffic could be automated.
And yet the decisive question is not only: can the car drive itself?
It is also:
Who is liable in a crash?
What error rate do we accept?
Who certifies the system?
What data may it collect?
Which decisions may a machine make?
How do insurers price the risk?
What happens when the software fails?
Which authority carries responsibility?
Technical capability therefore does not automatically imply social permission.
The more AI leaves the digital world and interacts directly with people, the more that distinction matters.
6. Trust: People Have to Hand Over Control
A technology can work objectively and still not be accepted subjectively.
An autonomous vehicle can drive more safely than a human on average.
Parents may still hesitate to let their child ride in one alone.
An AI system can make better decisions than an employee.
A board may still want a human being who carries the responsibility.
An algorithm can outperform a physician on a diagnosis.
A patient may still want to speak to a doctor.
People do not optimise their behaviour on statistical efficiency alone.
We respond to trust.
Control.
Experience.
Fear.
Accountability.
Habit.
A technology can be technically ready long before a society is ready to trust it.
More compute does not solve that problem.
7. Habits: People Do Not Update Like Software
There is no social software update.
We build routines.
Professions become part of our identity.
Companies build processes.
Organisations build hierarchies.
Agencies build procedures.
Societies build norms.
And once those structures exist, interests form around preserving them.
A company could in theory become dramatically more efficient with AI and still barely change its organisation for years.
Because an organisation is not only a set of processes.
It is people.
Careers.
Budgets.
Departments.
Responsibilities.
Status.
Power.
An org chart is therefore not simply a picture of how work is distributed.
It is also a picture of how power is distributed.
Technological change almost always means putting at least part of that distribution in question.
8. Incentives: Every Efficiency Gain Also Creates Losers
This is the point most technological forecasts underweight.
An innovation does not only create winners.
If an AI automates ten processes, ten efficiency gains appear.
But ten existing structures may lose relevance at the same time.
Occupations disappear.
Departments shrink.
Companies lose business models.
Assets lose value.
Associations lose members.
Managers lose scope.
People lose status.
Whoever loses has a rational incentive to slow the change down.
That is not necessarily technological backwardness.
It is incentive structure.
So it is not enough for a new technology to be objectively better.
A society has to give enough people a reason to support its adoption.
The speed of technological progress and the speed of social adoption are two different variables.
9. Law and Institutions: Code Scales Faster Than Institutions
Software can be rolled out globally in weeks.
Institutions cannot.
Laws have to be debated.
Rules have to be drafted.
Agencies have to settle jurisdiction.
Courts have to judge novel situations.
Liability questions have to be answered.
Insurers need new models.
International standards have to be negotiated.
And different states will arrive at different answers.
That produces a fundamental asymmetry:
Technology can scale globally.
Institutions remain largely local or national.
An AI system could in principle work tomorrow in Munich, San Francisco, Shanghai, Lagos and Mumbai at once.
But it meets entirely different legal systems, infrastructures, political systems, cultures and expectations in each.
Anyone forecasting a global technological future is therefore forecasting more than technology.
They are implicitly forecasting the institutional development of nearly every society on earth.
That is a different order of difficulty.
10. Geopolitics: The World Is Not a Closed System
There is no such thing as the world economy acting as a single agent with a single objective.
There are states.
Companies.
Governments.
Interest groups.
Owners.
And they pursue different goals.
States control resources.
They protect industries.
They tax trade.
They restrict exports.
They impose sanctions.
They compete over technology.
They control infrastructure.
They secure strategic inputs.
Even if a single country could build an almost fully automated economy, it still exists inside that system.
It may need raw materials from one country.
Chips from another.
Energy from a third.
Shipping lanes through a fourth.
AI does not automatically dissolve those dependencies.
It may deepen some of them.
The more compute, energy, semiconductors and specific materials matter, the greater the strategic value of controlling them.
Technological abundance can therefore, paradoxically, produce new geopolitical competition.
11. Scale: A Prototype Is Not a Civilisation
Building one machine that works is one thing.
Building a million is another.
Integrating a billion into the world economy is a different problem altogether.
A humanoid robot can be impressive.
But a global transformation does not need one.
It may need billions.
Which in turn requires factories, materials, chips, motors, batteries, energy, transport capacity, maintenance, spare parts, recycling, capital, regulatory approval and social acceptance.
Ironically, we first need vast amounts of the existing economy in order to build the automated one.
A prototype proves capability. Scale changes society.
Between those two points lies the Physical World Gap.
The Real Constraint Is Time
Almost all of these problems share one decisive property:
They are solvable.
We can produce more energy.
We can automate mining.
We can build robots.
We can extend infrastructure.
We can improve supply chains.
We can change laws.
We can design new insurance models.
People can learn to trust new technologies.
Organisations can change.
Societies can develop new norms.
But these processes run at radically different speeds.
AI can change within months.
A mine cannot.
A grid cannot.
A city cannot.
A legal system cannot.
An agency cannot.
A corporate culture cannot.
And billions of people certainly cannot.
Which is why the decisive variable may eventually stop being technological capability.
It becomes: synchronisation.
The Physical World Gap Framework
An AI-driven transformation can be read in four layers.
1. Capability
What can the technology do?
AI produces intelligence, knowledge, decisions, designs and software.
2. Translation
How does that intelligence reach the physical world?
Robotics, autonomous machines and intelligent infrastructure translate digital intelligence into physical action.
3. Adoption
Is society allowed and willing to deploy that capability?
Law, liability, trust, habits, interests and political decisions determine actual deployment.
4. Scale
Can the system become large enough to matter at the level of an economy?
Energy, materials, factories, infrastructure, supply chains and geopolitics determine scale.
Only when all four layers work does a technological capability become social reality.
Most forecasts extrapolate the first curve.
But societies move at the speed of the slowest one.
How Do We Close the Physical World Gap?
The answer cannot be to slow technological progress down.
We have to accelerate the other side.
Robotics Has to Become the Bridge
Robots are the most obvious connection between digital intelligence and physical reality.
But their greatest long-term value may not lie in replacing individual human tasks.
It lies in changing the physical world itself faster.
Robots build factories.
Robots build energy infrastructure.
Robots operate mines.
Robots maintain grids.
Robots move goods.
Robots recycle materials.
The more we automate the production of physical infrastructure itself, the faster the real world can respond to digital progress.
Infrastructure Has to Become Machine-Ready
We should not only build machines that somehow cope with a world designed for humans.
Over time we should also design the physical world for machines.
Autonomous ports.
Robot-ready factories.
Machine-readable warehouses.
Standardised interfaces.
Automated energy infrastructure.
Autonomous logistics systems.
The large productivity step may not come from the perfect humanoid robot.
It may come from combining capable machines with an environment built for them.
Regulation Has to Learn Faster
Regulation does not need to disappear.
The more powerful technology becomes, the more safety and accountability matter.
But regulation has to become better at learning.
Regulatory sandboxes.
Limited approvals.
Real-world test zones.
Measurable safety standards.
Data-based thresholds.
Stepwise expansion of autonomy.
Instead of choosing only between permitted and forbidden, institutions can become iterative themselves.
Trust Has to Be Engineered
Trust is not a soft factor.
It is infrastructure.
People accept new systems faster when risks are legible.
When responsibilities are clear.
When safety data is transparent.
When failures are analysed openly.
When control is handed over in stages.
And when people understand what the system is doing.
So technology does not only have to work.
It has to be built to be trusted.
Incentives Have to Be Aligned
People do not always resist change because they fail to understand it.
Sometimes they understand it very well.
And can see that they lose.
If AI produces enormous productivity while the gains stay extremely concentrated, resistance becomes rational.
A successful AI transition therefore needs more than intelligent machines.
It needs intelligent incentive structures.
The question is not only: how do we create more wealth?
It is also: how do we make sure enough people have a reason to let this transition happen?
The Constraint Does Not Disappear. It Moves.
Technological progress does not simply abolish scarcity.
It relocates the bottleneck.
When intelligence gets cheap, energy matters more.
When energy gets cheap, materials matter more.
When materials are available, production capacity matters more.
When production is automated, infrastructure matters more.
When infrastructure exists, regulation becomes the constraint.
When regulation follows, trust becomes the constraint.
When people accept the technology, interests and power structures become visible.
And when that barrier falls too, the next one appears.
Progress does not remove constraints. It moves them.
Perhaps that is the better definition of progress.
Not a world without limits.
But a civilisation that can move its limits faster and faster.
Reality Itself Could Become the Constraint
Today we concentrate on making AI more intelligent.
But we may reach a remarkable point:
Our machines would be ready for a future our world is not ready for.
Not for lack of intelligence.
But because of energy.
Materials.
Infrastructure.
Supply chains.
Law.
Institutions.
Interests.
Trust.
Habits.
Geopolitics.
And time.
At that point AI would no longer be the decisive constraint.
Reality itself could become the constraint.
That is precisely the Physical World Gap.
And the faster artificial intelligence advances, the more a different question matters than: what can AI do?
How Fast Can Reality Follow?
Because in the end it is not only the speed of our intelligence that determines how quickly the world changes.
It is our ability to translate intelligence into reality.
If you want to see where the ownership side of this question leads, it continues in Beyond Money: Who Owns Abundance?.
The Supertrends Behind the Physical World Gap
If the Physical World Gap really becomes one of the decisive constraints of the coming decades, another perspective follows from it.
The largest opportunities of the AI era may eventually not sit only where better intelligence is developed.
They may appear wherever someone makes the physical world faster.
Because every constraint creates a market.
If compute compounds while energy infrastructure does not, the value of new generation, grids and storage rises.
If AI can design new products in hours while factories need years to scale, the value of flexible, highly automated production rises.
If robots become more capable while our infrastructure was built for humans, a market appears for robot-ready buildings, factories, ports and logistics systems.
If autonomous systems work technically while regulation cannot keep pace, the value of new certification, insurance and governance systems rises.
If materials become the constraint, automated mining, materials science, substitution and recycling gain weight.
If billions of machines come into existence, maintenance, spare parts, machine-to-machine communication and autonomous repair become infrastructure of their own.
And if technology outruns social trust, trust itself becomes an economic asset.
Some of the largest supertrends of the coming decades could therefore emerge directly from the Physical World Gap.
The Automation of Automation
Perhaps the largest trend of all:
Machines that do not only produce goods, but create the production capacity for more machines.
Robots build factories.
Factories produce robots.
Robots build energy infrastructure.
Energy enables additional production.
Production creates additional machines.
At that point the physical economy itself begins to scale faster.
That may be the moment the Physical World Gap genuinely starts to close.
The biggest winners of the next technological era may not only be those who build the most intelligent AI, but those who solve one of its hardest problems: translating the speed of the digital world into the physical one.