“If you talk about ‘iterated multiple convolution kernels,’ which is a snappy phrase I remember from your last paper, it is impossible. I want you do to something for me, and put it as high on your priority list as anything in your models. I want you to find a way to explain the models in a way that someone with no special training will understand.”

“How can I do that?”

“Your problem. Use analogies, use pictures, use metaphors, I don’t mind if you have to try poetry and dancing. But we really need this — or all your work will be ignored, just as surely as if it came from outside the organization.”

Alex stared at her. He was feeling like a fool. She was right, and so obviously right that he should have thought of it for himself. “I’ll do my best. But how will I know when I have what you want?”

“We use the Napoleonic principle.” At Alex’s raised eye-brows,-she went on, “You’ll brief Macanelly, from Pedersen’s group. Do you know him?”

“No. But I’ve heard about him.”

“Heard what about him?”

“That nobody likes to work with him. That he’s conceited, and also that he’s close to being a moron.”

“That’s what I’ve heard, too. He’ll be perfect. Napoleon used to have a special officer, a very dim one, who read all outgoing dispatches. Unless a dispatch was clear even to that man, it didn’t go out. Loring Macanelly will be our dispatch reader. When we have an explanation of what you’re doing that he can understand and repeat back to me, we’ll be happy. Won’t we? You don’t look happy.”

“Kate, I want to work on theory, and I want to develop analytic models. I consider what we are doing supremely important. But I hate this other sort of stuff, simplifying work to the point where it’s more misleading than informative, and then feeding it to half-wits.”

“You know what they say: God must be specially fond of half-wits, because he made so many of them. Will you do it?”