We Count the Jobs Technology May Remove. What About the Jobs Created Around Making It Safe?
Aircraft can crash.
That statement sounds like an argument about danger.
It is also, strangely enough, part of an argument about employment.
Because aircraft could crash, aviation required safety systems.
Those systems required people.
- Pilots needed training.
- Aircraft required maintenance engineers.
- Airports needed air-traffic controllers.
- Accidents required investigators.
- Flight required meteorology.
- Airlines required safety personnel.
- Governments required aviation regulators.
- Manufacturers required reliability engineers.
Aircraft needed inspection regimes, certification systems, maintenance schedules and increasingly sophisticated safety technology.
Some of these roles would not exist, or would exist very differently, without the dangers created by aviation itself.
Technology created a capability. The capability created risks. The risks created problems. And solving those problems created work.
The Simple Employment Equation
This matters when we discuss artificial intelligence.
Much of the employment conversation follows a remarkably simple equation:
AI can perform Task X.
Therefore:
Jobs involving Task X disappear.
Then:
Unemployment increases by approximately that number.
Reality may prove considerably more complicated.
Problems Create Demand
Suppose organisations begin deploying AI widely.
Immediately new questions appear.
- Is the model reliable?
- Is somebody auditing it?
- Has it been tested for bias?
- Who is responsible when it fails?
- Can employees safely enter confidential information?
- Can the system be manipulated?
- How do we detect AI-generated fraud?
- How do we establish whether an image is authentic?
- How do regulators inspect these systems?
- How do businesses prove compliance?
- How do hospitals validate clinical AI?
- Who trains workers to use it appropriately?
- Who investigates incidents?
- Who builds the infrastructure underneath it?
- Who secures autonomous machines?
Those questions are problems.
Problems create demand.
The Professions Around AI Safety
We can already see the beginnings of professions around AI governance, model evaluation, AI security, data governance, AI assurance and responsible deployment.
Other specialisms may become ordinary professions that currently barely exist:
- Synthetic-media forensics
- AI incident investigation
- Algorithmic auditing
- Model-behaviour evaluation
- AI compliance
- Autonomous-system safety
- AI provenance and authentication
- Human-AI workflow design
- Robotics safety
- AI transition management
Perhaps some of these names will sound ridiculous twenty years from now.
So would “social-media manager” have sounded to many businesses in 1985.
Counting the Whole Ecosystem
This does not prove AI will create more jobs than it removes. Nobody can responsibly promise that.
AI may automate unusually broad categories of cognitive work. Some communities and occupations may experience painful disruption.
But there is an equally serious analytical error in the opposite direction:
Counting every conceivable job destroyed while assigning zero jobs to the economic ecosystem created around the new technology.
We did not forecast aviation employment by counting horse-carriage workers and stopping there.
We did not understand the automobile economy by counting only the activities cars replaced.
We could not understand the internet economy by looking only at travel agents, newspaper classifieds and physical retailers.
First-Order and Second-Order Effects
Technology produces first-order effects.
Then human beings respond.
Then those responses produce second-order effects.
Then entirely new markets develop around those effects.
AI will probably be no different in that respect, even if its eventual magnitude is different.
The fascinating possibility is that some future workers may owe their careers specifically to the fact that AI is dangerous.
They will be employed to make it safer.
Just as people today earn livelihoods making aviation safer, financial systems safer, medicines safer and computer networks safer.
Prepare for Both Questions
So yes, ask:
“Which jobs could AI eliminate?”
But don't stop the analysis halfway.
Also ask:
“What work becomes necessary because AI exists?”
The future labour market will be shaped by both questions.
We should prepare for both.
Read the series introduction: The Asymmetric Technology Ledger →
Read Part III: You Cannot Regulate What You Refuse to Understand →