Designing AI products requires a deeper understanding of what people want. And what they voice might be different from what they want.

Humans love control and predictability while craving surprise at the same time. You can think of control and surprise as two internal dials. Their configuration might be different for different people.

Using a restaurant analogy (I love food analogies):

• Control dial: going to a familiar restaurant and ordering the favorite dish you’ve had ten times before
• Surprise dial: A fancy 10-course tasting menu at a new place

Because they are dials, we can have both in different configurations. The 10-course meal could be in your favorite cuisine — some control and some surprise. Some people might enjoy comfort, while others love to push boundaries. It’s why we often react wildly differently to the same thing.

AI models do what we tell them to do, and they will become better at that as they get safer and better aligned. But most of it won’t be very impressive. It will give what we requested — a comfort meal. We don’t need 10-course tasting menus all the time. We all have the ‘control’ dial inside us. But the perception of quality is often linked to how surprising or new something is. Where’s the magic in it if you know exactly what you’ll get? The AI examples that people love the most are the ones where it goes off the rails and becomes “unhinged.” It’s the ‘surprise’ dial being satisfied.

I now design AI experiences regularly and often need to understand those dials in people. What do people expect from a given experience? The design process now requires understanding where their control/surprise dials are. Create something predictable, and they’ll never use it again. Design a complete surprise, and you’ll have a PR nightmare. 

The control dial is visible through language. “I expect this,” and all its variations you usually hear during user testing. The surprise dial is visible through emotion and often shows up as a sense of excitement and exhilaration that is hard for most people to describe.

Designing great AI products in the future will likely require catering to both dials. Most seem to be focusing only on the control one.