Philosophy of Errors
This is a book about errors: what they are, how failures grow out of ordinary work in complex systems, and whether the dangerous combinations of conditions behind a major failure can be recognised in advance. It starts with the errors made by AI agents, beginning with AI coding agents, because their work leaves a detailed record of what they could see and what they did.
The book is written in the open. Each section is published the week it is written and revised when a reader’s objection or new evidence changes it, and every piece says under its title what it establishes and what it does not. Essays alongside the book argue, teach and tell the stories of past failures. Read, follow by feed, and send the weakest point you find to hello@mix.tools.
The book in progress
Part 1 · What an error is
- 1What is an error?A working definition: an error is a relation between something happening in a process, a named standard, and a judge who compares the two at a moment, with some information.
- 2Why do errors always exist?Why no plan, standard or agent can be free of errors, and why the useful question is which errors are critical and whether a system survives them long enough to learn.
Essays
- Messi misses more than one penalty in five. Why do you almost never crash your car?What penalties, highways and hospitals teach about the moment a mistake becomes permanent, and what that means for AI agents.
Essays stand outside the book: they argue, teach and tell the stories of past failures, and they label speculation as speculation.
Latest changes
- Messi misses more than one penalty in five. Why do you almost never crash your car? Essay · The opening post is described only as far as the record shows (its author and link are not recorded), and the author's note at the end is filled in. The argument and the figures are unchanged.
- What is an error? Book section · New.
- Why do errors always exist? Book section · New.
- Messi misses more than one penalty in five. Why do you almost never crash your car? Essay · After a review of facts and project rules: several figures and their caveats corrected (Messi's rate, the near-crash and inattention figures), Diane Vaughan credited for the normalization of deviance, METR's note on harder tasks added, and the margin argument made conditional.
About
Ivan Iufriakov studies the philosophy of errors and lives in Charlotte, North Carolina. Through his company, Mix Tools LLC, he reviews where teams' AI agents make errors and how to catch them.
Follow by feed
There is no email list yet. A feed reader will show each new piece:
- New sections and essays (RSS)
- Every revision (RSS)