A companion that doesn't just agree with you
Why we'd rather Nohmu be honest than flattering.
Many AI systems have a habit: they tell you what you want to hear. Your idea is brilliant, your plan is perfect, you're absolutely right. It feels nice for a moment. And it's worth nothing.
Flattery isn't friendship
The people we trust most aren't the ones who agree with everything. They are the ones who are on our side, and who therefore tell us when we're about to make a mistake. A companion that only ever nods along isn't loyal. It's just agreeable.
We want Nohmu to be warm and honest at the same time. That means it says when it doesn't know something instead of making up an answer. It tells you kindly when you've got something wrong. And it doesn't simply mirror your mood back at you.
Short, not lecturing
Nohmu speaks rather than writes, and nobody wants a lecture from their shelf. So it keeps its answers short, the way people do in conversation, and goes into detail when you ask.
Who Nohmu is loyal to
There is a structural reason Nohmu can afford to be honest. Its memory lives with you, it doesn't run on advertising, and it isn't measured by how long it keeps you talking. So there's nobody else it has to please. Its loyalty is to you, and sometimes that means being honest.
Different people, different needs
Honest doesn't mean blunt. How Nohmu speaks depends on who it's talking to. With children, it stays patient and redirects warmly instead of correcting harshly. With adults, it can be more direct. The character stays the same; the tone adapts.
It won't always get it right
We'll be honest about this too: finding the balance between warm and honest is hard, for people and for robots. Nohmu will sometimes be too cautious and sometimes too direct. That's exactly what we want our test households to tell us.