Why Asking What Happens If We Use AI Is Only Half the Question
There is a peculiar way we discuss technological risk.
Imagine an AI system being considered for a hospital.
Someone asks: “What happens if the AI makes a mistake?”
Excellent question.
“What happens if it gives the wrong recommendation?”
Also necessary.
“What happens if somebody trusts it too much?”
Important.
But there is another question that strangely receives less emotional attention:
What happens if we do not use it?
Non-Adoption Is Not Risk-Free
Suppose the existing system is slow. Suppose clinicians are drowning in administrative work. Suppose patients are waiting too long. Suppose important information is occasionally missed because humans are tired, busy or overloaded.
Suppose an appropriately designed AI-assisted system could reduce some of those problems.
Suddenly doing nothing is no longer the risk-free option. It has consequences too.
This is what I think of as the asymmetric technology ledger.
We carefully enter possible technological harms on one side of the page while sometimes treating the alternative as though it contains no costs at all.
But the appropriate comparison is rarely:
AI versus perfection.
It is usually:
AI-assisted reality versus existing reality.
That distinction matters enormously.
Compare Systems With the Reality They Replace
Consider autonomous driving.
If an autonomous vehicle kills somebody, the fact that AI was involved will understandably receive enormous attention.
But suppose autonomous systems eventually reduce road deaths substantially. Would it be rational to reject them because they cannot reduce road deaths to zero?
The correct comparison would be:
Deaths under the new system versus deaths under the system it replaces.
Medicine already understands this principle.
No serious doctor asks whether a treatment has any side effects. The question is whether the expected benefit justifies the expected risk compared with the available alternatives.
A medicine that occasionally causes a serious adverse effect may still save enormous numbers of lives. Refusing the medicine also has an outcome.
And yet technological debates sometimes behave as though non-adoption is a neutral control group.
It isn't.
The Four Questions of a Complete Ledger
There are at least four questions:
- What are the risks of adoption?
- What are the benefits of adoption?
- What risks can be mitigated?
- What are the risks and opportunity costs of non-adoption?
AI discussions frequently concentrate overwhelmingly on the first.
Education: Two Things Can Be True
Consider education.
Yes, students can use AI to avoid thinking. That is a real problem.
But AI may also provide personalised explanations to a child whose parents cannot afford a private tutor. It may allow somebody embarrassed to ask a classroom question to ask it privately twenty times. It may adapt explanations to different levels of understanding.
It may help a learner practise a language at midnight when no teacher is available.
So the educational question cannot simply be:
“Can AI make students intellectually lazy?”
It must also include:
“Can properly used AI expand access to high-quality individualised learning?”
Both can be true simultaneously.
Employment and the Cost of Standing Still
The same applies to employment.
We hear:
What jobs might AI destroy?
We should ask it.
But where is the equally sustained discussion about:
- What businesses become economically possible because AI reduces their startup costs?
- What scientific projects become feasible?
- What previously expensive expertise becomes accessible?
- What new occupations emerge?
- What products will exist that nobody has conceived yet?
- What jobs arise specifically because AI requires supervision, security, auditing and regulation?
- What happens to countries or communities that deliberately refuse capabilities their competitors adopt?
Again, none of this proves that AI's eventual employment balance will be positive.
We do not know.
And that is precisely the point. Certainty about catastrophic employment outcomes is no more intellectually justified merely because pessimism sounds cautious.
When Benefits Become Invisible
There is another important asymmetry.
Benefits gradually become invisible.
An internet scam remains an internet scam. But when someone sends their ten-thousandth successful email, nobody announces:
“Internet successfully delivers another message.”
When online banking works, we call it banking. When GPS works, we call it directions. When cloud computing works, we barely think about it.
Successful technologies disappear into ordinary life.
Their failures remain recognisably technological.
That means society can continue counting the costs while gradually forgetting that the benefits belong in the same ledger.
Responsible Progress
AI deserves scrutiny. It deserves safety research. It deserves sensible regulation. It deserves scepticism when claims exceed evidence.
But caution is not achieved by counting only one side of an equation.
There is risk in moving too quickly. There can also be risk in standing still.
The difficult work is finding the space between them.
That space is not technological recklessness.
It is where responsible progress happens.
Read the series introduction: The Asymmetric Technology Ledger →