A Main Street Independent reference — what MindSpec is: the guided interview that turns a person’s real values into a structured profile a computer can use, and the tool that built every one of this publication’s heteronym voices. Published under CC0.

In plain terms: it’s a way to write down what someone actually cares about — precisely enough that software can act on it — by asking about situations instead of asking for a list.

What MindSpec is

Most attempts to tell a computer “here is what I value” fail at the first step, because people are bad at stating their values directly. Ask someone what matters to them and you get a list of admirable words — honesty, family, hard work — that everyone says and that predicts almost nothing about what they will actually do when two of those words collide.

MindSpec is a guided interview that gets around this. Instead of asking what you value, it shows you people and situations and watches how you react — and from those reactions it works out which values actually run your decisions, and how strongly. The result is a single document, called a Mind file, that lists each value with a number for how much weight it carries. It is precise enough that software can use it, and honest enough to be worth using.

It is also the exact tool that produced every named voice on this site. Each heteronym’s Mind file — the document behind Mary Magdalena, Malcolm Little King, and the rest — is the output of a MindSpec interview run on an invented character. (See Voice Architecture: The Pen-Name System.)

Three things it can describe

The same interview runs three ways:

  • Yourself. Your own values, written down, so an AI assistant can help you in a way that fits what you actually care about instead of guessing.
  • An AI agent. A stable set of values for a long-running AI, so its behavior does not drift from one conversation to the next.
  • A fictional character. A character whose values hang together instead of being a loose bundle of traits — which is how this publication builds its voices, and how a novelist could build one.

The method is identical in all three cases. A person, a machine, and a character are each, for this purpose, a bundle of values leaning on one another; once you can measure the bundle, you can describe any of them the same way.

How the interview works

The interview never asks you to rate “honesty” or “ambition.” It shows you a short description of a person for whom one value is central — without naming it — and asks how much that person is like you. Then it puts two values in conflict and asks which way you would jump: a colleague you respect is wrong and wants you to agree — do you keep the peace or say the true thing? Your answer to that, repeated across many such collisions, is what reveals which value actually wins when they fight.

From the pattern of answers, MindSpec assigns each value a weight and writes the Mind file. The full list of values and the exact questions are laid out in a companion paper, Inside MindSpec: The Values and the Questions.

Why turn values into numbers

Words do not combine; numbers do. “Honest” and “loyal” tell you nothing about what a person does when honesty and loyalty pull in opposite directions. A weight on each — honesty at 8, loyalty at 4 — tells you exactly. Putting a number on every value is what lets the system anticipate what someone would likely do in a situation no one described in advance, and lets an AI act in a way that fits a person without re-interviewing them at every turn.

Matching the effort to the job

A full values interview is real work — several hours, taken seriously. It would be absurd to demand that for a one-off task. So MindSpec scales the effort to the use:

  • A throwaway helper for a single task just gets sensible defaults — no interview at all.
  • A recurring assistant gets a short version.
  • Your own profile, a lifelong AI thinking partner, or a fully drawn fictional character gets the full interview.

The rule is simple: the tool is worth something only if people use it, and people use it only when the cost of answering matches what they get back.

Keeping it honest over time

An AI that is allowed to learn can quietly drift — nudged by whatever pleased the user yesterday until its values are no longer the ones it began with. MindSpec guards against that with a few safeguards: periodic re-checks, an alarm when a value’s weight moves too far, and a written log of every change and the reason for it. And the handful of deepest values — the ones marked non-negotiable — are sealed off from ordinary feedback entirely, so they cannot be edited away under pressure. (Why those particular values sit at the bottom and never move is the subject of the companion paper, The Parliament of Mind.)

Where it fits

MindSpec is one part of a larger idea: a mind — yours or a machine’s — runs on three things working together. Knowledge (everything it has learned), values (what it cares about), and a way to search the first and weigh it against the second. MindSpec is the values part. The theory underneath it — why a mind can be measured this way at all — is The Parliament of Mind. What the three parts add up to is the subject of Creativity from Knowledge and Values. And the clearest working example of MindSpec in production is this publication itself, whose every named voice is a Mind file — see Voice Architecture: The Pen-Name System.

Main Street Independent — CC0 / public domain. Copy, adapt, and reuse freely.