Watch any team discuss AI and you will see it. One person is excited, another is worried, and both are sure they are talking about the same technology. They are not. The word “AI” is being stretched to cover a driverless truck that removes a driving job, a writing assistant that makes a lawyer faster, and a diagnostic tool that helps a doctor make the right call. Those three point in very different directions for jobs, skills and strategy. So the conversation goes round in circles, because nobody has agreed on what they are actually arguing about.
Two kinds of AI
After years of advising teams through this, I settle one question before any other. Not “is AI good or bad?” but “what kind of AI are we talking about?” Almost everything useful fits into two simple categories.
Displacing AI. Technology used mainly to take a task over from a person. The aim is to reduce or remove the human contribution: a warehouse that runs on robots overnight, an automated production line, a process handled end to end with nobody in the seat. The person is moved out.
Elevating AI. Technology used mainly to make a person more capable. The person stays in the loop and does more, does it better and does it more consistently: a support agent who resolves more tickets with an assistant alongside them, a junior analyst whose first draft is suddenly much stronger, a small business owner who now has analytical muscle that once needed a specialist. The person is lifted up.
An old idea in plainer language
To be fair, the underlying idea is not mine. Economists have studied the difference between technology that replaces people and technology that complements them for decades. What I am offering is a plainer way to say it, so an ordinary business conversation can start from shared ground instead of crossed wires.
Why the distinction matters
Naming the two kinds clears the fog. “Will AI take my job?” is a question about displacement. “Will AI make me worse at my own work without my noticing?” is a question about elevation. They are different worries, they need different answers, and you cannot deal with either until you know which kind of AI is in front of you.
Most of the anxiety in that room comes from treating one undifferentiated thing as if a single answer could cover it. It can’t.
In Part 2, I will share the practical test I use to tell the two apart. Surprisingly, it has very little to do with the technology itself.
Which kind do you think you are building?