The tools that wear best are usually the ones that are willing to say what they are doing. Nicole Junkerman has been collecting the qualities that seem to separate a piece of software people get along with from one they merely tolerate, and almost none of them are technical. They are mostly about language, and about a willingness to be asked.
Five qualities turn up again and again.
- It says what it did. Not a progress bar standing in for an explanation, but a plain sentence afterwards: this was changed, this was kept, this was skipped. A tool that reports its own work can be checked, and a tool that can be checked does not have to be believed.
- It uses ordinary words first. Specialist terms are fine once they have been introduced. The trouble starts when the first sentence someone meets assumes a vocabulary they were never given, because at that point the only honest response is to guess.
- It admits what it does not know. Confidence is cheap to produce and expensive to trust. Something that says it is unsure, and says why, is easier to work alongside than something that is never wrong out loud.
- It can be undone. A step that can be reversed invites people to try it. A step that cannot turns every decision into a small act of nerve, and nervous users learn less, because they stop exploring.
- It leaves a way through to a person. Not as an apology for a failure, but as a normal part of the design. Most questions worth asking are the ones a form was never shaped to receive.
What these have in common is modesty about the role the tool is playing. Each of them assumes that the person using it is the one who has to understand the outcome and live with it, and that the tool's job is to make that understanding easier rather than unnecessary. It is a small shift in posture and it changes almost everything downstream.
The opposite is familiar enough. Something arrives polished, performs a step nobody can inspect, and produces a result delivered with great assurance. It may well be right. The difficulty is that nothing about the way it was presented helps anyone work out whether it is, so trust has to be granted rather than earned, and trust granted that way is brittle.
She is reading about this rather than working on it, and the interest is mostly in the language. A field that can explain itself in plain sentences is usually a field that understands itself well, and the explaining tends to improve the thing being explained. That is true of writing, and it seems to be true here.
The glossary is where the plainer definitions are kept, an earlier note on reinforcement learning in plain words takes the same approach to one term, and the rest is on the blog.