Leave the Ninety-Nine

Every system we build is tuned for the many. That is not a flaw in the systems. It is close to the definition of what a system is. You design for the common case, you handle it well, and you accept that some small share of cases at the edges will be handled badly or not at all. When the share is small enough, you call it a success. Ninety-nine percent is an excellent number on almost any dashboard I have ever looked at.

The machines we are now building are the most powerful expression of that logic yet. A model is trained on the average of what people have written and done, and it is very good at the average. It serves the many astonishingly well. The person who does not fit, whose question is strange, whose situation is rare, whose life does not resemble the data, is the one most likely to be served badly. That is not malice. It is arithmetic.

A question asked as if it were obvious

Jesus told a story that runs exactly the other way. In the World English Bible, it begins like this: "Which of you men, if you had one hundred sheep, and lost one of them, wouldn't leave the ninety-nine in the wilderness, and go after the one that was lost, until he found it?" (Luke 15:4)

I have read that verse many times, and what keeps catching me is not the shepherd. It is the framing of the question. Jesus does not present the shepherd's choice as heroic or radical. He asks it as if the answer were obvious, as if any sensible person in the crowd would of course do the same.

By the logic of every system I have built, it is not obvious at all. It is close to reckless. The flock is ninety-nine percent intact. Leaving it in the wilderness to search for one stray puts the whole operation at risk. A good operations manager would log the loss, note that it is within tolerance, and keep the ninety-nine moving.

That is what makes the story so unsettling. It assumes a way of valuing that most of our institutions have quietly abandoned, and it assumes it so confidently that it does not even bother to argue for it.

Counted versus known

I think the difference comes down to two ways of relating to people. A system counts them. A shepherd knows them.

Counting is not bad. Counting is how a hospital knows it needs more beds, how a church knows it needs another service, how a city knows where to put a bus line. I build software for a living, and much of it counts things, and I am glad it does. Counting serves people in aggregate, and aggregate service matters.

But counting cannot tell you that one is missing in a way that matters. To a count, one in a hundred is a rounding error. To someone who knows the hundred by name, one missing is a face that is not there. The shepherd notices the absence because the shepherd knows who should be present. You cannot leave the ninety-nine to look for the one unless you knew the one in the first place.

There is something else in the story that is easy to miss. The shepherd does not wait for the lost sheep to file a request. He does not post a notice and see who responds. He goes into the wilderness, which is to say into the place where the one is, on its terms, at some cost to himself. Most of our systems, even the good ones, are built to respond. The person who has wandered far enough is often the person least able to ask. A system that waits to be asked will never find them. A shepherd goes.

The opening chapters of Genesis say every person is made in God's image. If that is true, then no person is a rounding error, however rare their case, however small their share of the total. The image is not distributed across the flock. It is carried whole by each one.

What machines cannot do here

It is worth being precise about what our technology can and cannot contribute. Today's AI wants nothing. It does not notice that someone is missing, because noticing absence requires caring who should be present. It can be built to flag an outlier, and that can be genuinely useful. But a flag is not a search. Someone still has to decide that the one is worth leaving the ninety-nine for, and go.

That decision has always belonged to people. It belongs to the pastor who notices that a family has not been seen for a few weeks and drives over instead of updating a list. It belongs to the teacher who stays after class for the one student who did not follow. It belongs to the manager who answers the strange complaint instead of closing it as an edge case.

Who is your one

I do not read the parable as a rejection of good systems. Shepherds counted their sheep. That is how this one knew one was missing. The story is a reminder of what the counting is for. It exists to serve people who are each known, not to replace knowing them.

So the question I keep asking, of the companies I run and of myself, is simple. Who in my care is the one percent? Who has fallen into the gap that every well-designed system leaves at its edges? And is anyone going after them?

I do not always like my answers. It is easier to celebrate the ninety-nine, and they deserve to be served well. But the measure of whether we see people as image bearers or as data is not how well we serve the average. It is what we do about the one.

The answer the story expects is obvious. That is exactly why it is so hard.