If Users Cannot Understand It They Will Not Trust It

Early in my career, I learned that people forgive a lot when they understand what is happening. They forgive slow loading when they can see progress. They forgive errors when the message tells them what went wrong.

What they do not forgive is mystery.

That lesson feels more important than ever now that AI sits behind so many products. A recommendation appears. A document gets summarized. An application gets flagged. And the user is left wondering why.

Ben Shneiderman, one of the pioneers of human computer interaction, has spent years arguing that automation and human control are not opposites. In his work on Human-Centered AI, he calls for technologies that offer “high levels of human control and high levels of computer automation.” His goal is AI that is reliable, safe and trustworthy.

Trust is the key word. And trust depends on understanding.

If a person cannot see why a system reached a conclusion, they have two choices. Accept it blindly or reject it completely. Neither is good for them. Neither is good for the product.

Here is how I think about designing for transparency.

• Explain the why, not the math. Users do not need model details. They need a plain reason, such as “Suggested because you saved similar projects.”
• Show what the system used. Point to the sources, inputs or data behind an answer so people can check it themselves.
• Be honest about uncertainty. If the system is guessing, say so. A confident tone on a weak answer damages trust faster than an honest maybe.
• Label AI generated content clearly. People deserve to know when they are reading something a machine wrote.
• Make correction easy. When users can push back, edit or undo, the system feels like a partner instead of an authority.

Transparency is not a feature you add at the end. It is a decision you make at the start.

Dieter Rams put it simply in his principle that good design is honest. “It does not make a product more innovative, powerful or valuable than it really is.”

That principle was written for physical products, but it describes AI perfectly. Overpromising intelligence is just another form of dishonest design.

If users cannot understand what your AI is doing, they will not trust it. And if they do not trust it, it does not matter how smart it is.