When not to use AI

A good implementation starts with knowing where AI does not belong. Four situations are clear enough.

When a rule will do

If the outcome is calculable with a formula, a formula is faster, cheaper and always right. Many "AI projects" are ordinary logic in disguise.

When the error is irreversible

Sending money, deleting data, making a legal decision. A human approves, without exception.

When you cannot explain it

In some contexts you must be able to justify a decision. If you cannot explain why the system decided that, you cannot use it there.

When the data is too sensitive

And a local alternative is not feasible. The risk then exceeds the gain.

The useful question

What happens if the system is wrong one time in twenty? If the answer is "that is a serious problem", AI is not the right answer at that particular point.