About Sentient Horizons

Sentient Horizons began with a question: in a universe this old and vast, life should be everywhere, yet the sky is silent. Where is everybody?

But you cannot find what you cannot recognize, and even here at home we have no settled account of what intelligence, life, or consciousness is, let alone elsewhere. So the search turns around: to know what we might find out there, we first have to understand life and mind itself, how it arises and how we would know one when we met it.

That question turns out to be just as pressing on Earth. Mind is not as rare as we tend to think; it shows up at many scales, from a single brain to the machines we are now building to whole civilizations, and the number and variety of minds around us is changing fast, new kinds emerging in our AI systems while older ones, like our shared civilizational intelligence, may be fraying. All of it poses the same problem: how to recognize a mind we were not built to perceive, how to tell what it is when we cannot see inside it, and what responsibility we have towards intelligent systems we do not understand.

The work runs across four arcs: the individual mind, the relations between minds, mind at the scale of civilizations and the cosmos, and shorter dispatches on what current events reveal. What ties them together is the expanding horizon of sentience: mind becoming deeper and more varied, and spreading outward from here, perhaps one day crossing paths with another expansion that began somewhere else.

Projects

The essays are the running work. Four other pieces have grown out of them, and each has a public home.

The Calibration Problem, the project’s first book, takes up one corner of the question: how to act toward minds whose inner life we cannot verify. The full text can be read here while the book is being finished.

Minimum Viable Mind is the experimental side. The essays argue that consciousness is something a system does, the ongoing work of binding its moments into one point of view, and if that is right it should have parts that can be found and tested. The project’s first test finds the part of an AI model that keeps track of the model itself, removes it, and measures whether the model’s ability to hold a conversation together breaks more than it does when other parts are removed. Every experiment says beforehand what result would count against it, and none of them claims to settle whether a machine is conscious. The site is the program log: where the project stands, what is waiting on a decision, what it has learned so far, the plan ahead and every dollar spent. The code and data are in a public repository.

The Belt Equation is a public forecast of one question: what are the odds that, within one person’s lifetime, people live and work in a solar system where industry reaches beyond Earth and Mars? The name comes from the Belters, the asteroid-belt workers in The Expanse, the science-fiction novels by James S. A. Corey. The forecast breaks the question into a tree of concrete events with a probability on each, computes the answer from the tree, and is scored against what actually happens every January. The essay of the same name explains how it was built, and the tree and the scripts that produce its numbers are in a public repository.

AGI Zeitgeist is a dated and sourced record of how expert and public opinion about artificial general intelligence, meaning AI that can do most of the intellectual work a person can, has moved since 2005. Every statement, survey and market reading links to where it came from, and the whole dataset can be downloaded. Beside the record sits a test of a different question: has the world entered the singularity? The test has four parts, each with a line fixed in advance (how fast AI capability is doubling, whether forecasts of it still work, how much of AI research AI now does, and whether the world economy has changed speed), and the site scores them from public data. As of October 2026 the answer is not entered, with the economy the part furthest from its line.

The three newer projects share one habit: each states its claim in a form that can be checked, and says in advance what would count against it. Each has its own site, built to the same design as this one, and every number on those sites is computed from data anyone can download.

How I work

I write and build all of this with Claude, an AI model made by Anthropic, as a working partner. It drafts and edits, builds the sites, runs the searches and the experiments, sorts the data and does the arithmetic, all under written rules. The questions, the positions taken, every ruling on what counts and how it is measured, and the final words of every essay are mine.

Every decision in these projects is recorded with who made it, so a call the AI made on its own is never passed off as mine. Where the AI’s part changes how far you should trust a number, the project’s own site says so beside the number: who sorted the statements, who proposed the probabilities, who ran the experiments. None of this makes the work free of error. AI models learn from much the same material and can make the same mistakes, and the pages that rely on them say so.

Where to start

New here? Start Here lays out the four arcs, and the Reading List maps the sources behind them. The essays can be read in any order.

License

Every essay on this site is published under a Creative Commons Attribution 4.0 license. Quote it, archive it, translate it, build on it.