The argument is the policy

Most of the AI policies I read this year were written by nobody.

That is not an insult to the people whose names are on them. It is a description of how they came to exist. A template got downloaded. The bracketed fields, the ones that say company name here and approved tools here, stayed square. The file went into a shared drive with FINAL in the name, and the question that prompted it, from a board member or an insurer or a customer’s vendor form, got answered with a yes.

This week the studio started building a questionnaire meant to produce the opposite kind of document: one a small company has actually decided. Twenty plain questions, nine sections, a policy that changes with every answer so that no two companies get the same text. The build is the ordinary part. What I have been thinking about is question seven.

Question seven

Question seven asks which classes of data may never be entered into an AI tool. Not which tools are allowed; that is question four, and it is answered quickly, because people already know what they use. Question seven is different, because it asks the owner to say out loud that there is information in the company more valuable than the convenience of pasting it into a chatbot, and then to name it.

Customer lists. Payroll. The contract with the largest account. A colleague’s medical note that came through in an email. Somewhere, each of these has already been pasted. The owner knows this. The person sitting next to the owner knows it too, and knows which of the two of them did it.

That is where the argument starts. I did not design the question to start it. I designed it because a policy without a data clause is not a policy, and I could not find a template that had one filled in. The argument is a consequence of asking a real question of real people who have been avoiding it.

The receipt

I have come to think the argument is the policy. The document that comes out the other side, the DOCX with the cover memo and the effective date and the approver’s line, is the receipt for a decision that was made in a room, by people, about their own company. The template skips the room. That is why it stays blank.

We are building the questionnaire so that skipping the room is no longer the path of least resistance. Whether it works is not something I can claim yet. The first signal is still two weeks out, and I am suspicious of anyone who knows in advance how people will behave when asked to decide something.

But I know what a policy written by nobody looks like. I read a great many of them. I would like to see what one written by somebody looks like instead.

A template is a decision deferred, with better formatting.