Human in the Loop

Misunderstood

A research-driven project on human–AI collaboration, built across four seasonal editions: a technical textbook, a reading workbook, a pamphlet, an interactive reader, and a deployable platform.

Lazaros Toumanidis — laztoum.com · what-if.io · limen-os.io

Winter Main textbook 18 chapters, technical appendices, teaching package, and canonical PDF. The full treatment. Summer Web workbook A reading-first edition with reflection questions, scenarios, and word puzzles. The conceptual material from the main textbook, without the apparatus. Autumn Pamphlet Short, self-contained, designed to read in one sitting. A good place to start. Reader Interactive reader All four editions with text-to-speech, ambient audio, chapter navigation, and twelve embedded games. Spring Platform A deployable treasure-hunt platform: Rust API, Python AI sidecar, React dashboard, Flutter mobile, OIDC auth. Companion Supporting materials Shorter documents, instructor notes, and study aids across the seasonal editions. Limen OS Desktop environment A desktop environment shell with all four editions embedded, nine interactive games, and an AI assistant. Demonstrates the book's HITL principles as a designed interface.

The journey

Autumn Start here — one sitting
Summer Reflect & apply
Winter Full technical depth
Reader All editions + games + TTS
Spring Deploy the platform
Companion Study aids & notes
Limen OS Live in the shell

The central argument is that human judgment in AI-assisted decisions is not a temporary limitation of current systems — it is grounded in what decisions are and who is responsible for them. The five-dimension framework developed here (uncertainty detection, intervention design, timing, stakes calibration, and feedback integration) provides a practical scaffold for evaluating any HITL system.

The Winter edition is the primary technical reference. Autumn is the shortest entry point. Summer is the workbook. The Reader and Spring editions are interactive surfaces built on the same material.

The content across all editions was produced with substantial AI assistance and fully reviewed by the author. Translations have been validated for English and Greek only; all other language editions are machine-generated and unreviewed.