Fairmind
Fairmind helps people have informed and fair-minded opinions on polarizing topics. The project has created guides for topics like:
The guides use Fairmind Format, a way to explain complex, contested topics fairly, accurately, and concisely. The goal is to make it easier for people to be fair-minded. The easier it is, the more people will do it. The more people who do it, the better we—as individuals and collectively—will be able to think and talk about hard issues that matter.
This is especially important for young people who’ve known nothing but our hyper-polarized era. As they become ready to reason through difficult questions for themselves, they deserve examples of how to approach contested issues without prejudice—like a judge, not a partisan. Fairmind aims to help make that kind of reasoning feel normal and expected.
I have been thinking about the need for Fairmind since the late 1990s. I registered the domain fairmind.com in 2001 and added fairmind.org later.
I was busy with a career and raising a family back then, but the idea was always with me. I’d collect articles and read books related to the challenge. I’d capture notes and tell myself, someday I’ll take on this project.
In late 2022, I reached a point in my life when I could do that. Over the next few years, with the help of hundreds of volunteers who were willing test subjects, I experimented with many different designs for what later became Fairmind Format.
This period coincided with the rise of large-language-model (LLM) AI services like ChatGPT. These were a godsend for me in being able to quickly create and evolve prototypes for people to test. In some cases, the prototypes themselves used a conversational AI interface to walk people through a topic. Although the early years of LLM AI had many shortcomings, there was obvious and compelling value. It was fun and exciting to explore how to use the technology as it rapidly developed.
As of late 2025, if I gave a state-of-the-art AI model a carefully crafted prompt and highly curated content, it could do a mostly solid job as a conversational partner in helping a user understand a polarizing issue. But even the best models weren’t fully reliable or consistent. Much of the public discourse around polarizing topics was distorted in ways that didn’t just average out across an AI’s training; there were real biases, often subtle. Or, when attempting to be balanced, the AI could dilute valid arguments alongside the invalid. Moreover, these effects could change not just across model releases but sometimes with the same model.
To get the AI to do the right thing, I found myself having to give it more and more curated knowledge for a guide’s topic: important facts properly sourced, balanced arguments that could withstand scrutiny, key questions driving people’s positions, example viewpoints, as well as explanations of why certain widely used “facts” or arguments were problematic. The prompt also needed a lot of process instructions to ensure the conversation covered the topic systematically yet concisely. Without this extra content and process knowledge, the AI would just wing it based on its native training—which sometimes worked, until it didn’t.
I ended up concluding that Fairmind’s best contribution would be to focus not on the AI-driven experience but rather on the highly curated knowledge necessary for people to truly understand a polarizing issue. That could happen either by reading a guide themselves or by having their AI assistant consume a machine-optimized version and then converse about it. The key insight was, the AI technology would continue to evolve rapidly, and there were many different ways to design conversational experiences. In those respects, the target was moving and multivariate. But the need for clean, curated knowledge was a relatively singular and stable constant.
That’s when I locked into what now is Fairmind Format. It’s a document structure and set of authoring principles designed to represent polarizing issues fairly, accurately, and concisely. With the initial five guides, I exercised the format on what are widely considered the hardest issues out there. Academics and other experts reviewed every guide’s content. In many cases, they helped make improvements, but always they confirmed that the format and purpose were sound.
So as of mid-2026, I am halfway through creating the initial set of ten guides, each reviewed by experts and other volunteers. Already, the existing guides have drawn thousands of readers through search engines and AI chats. When I’ve finished the first ten guides, I will start working to encourage adoption by educators at the AP high school and college undergraduate level. The guides are for people beyond their school years too, of course.
If you have questions or ideas for the Fairmind project, feel free to contact me through the fairmind.org site.