# Altruistic — a personal working scratchpad

What does altruism even mean? Notes on intelligence, cooperation and systems.
Revised 19 October 2026. These are contestable ideas, reading, experiments,
retrospective mistakes and unresolved questions. The author assumes most readers
know more than he does about at least part of this and asks for corrections and
counterexamples. Agents put much of the wording together from his notes and
conversations; the essay says so at its top.

Navigation, the same in every language: Essay · Questions · Making · Reading ·
About. Each language has the same routes under /ja/, /ne/, /hi/ and /bn/. A page
that exists only in English says so in each language.

- https://altruistic.ai/research-notes — the essay, "what i mean by altruistic"
  (v0.7.0). It opens on one question: what makes care possible through good, bad
  and mixed states? Its picture is metronomes on a light board on two tins: they
  fall into step with nobody pushing them, a knocked one settles back, and one
  lifted off keeps its own time. Then: wanting to help is not knowing how (truth
  is a prerequisite for care); help that mostly serves the helper; the child who
  can't safely ask, and why any intelligence that helps should be humble rather
  than hand out advice; help in plain sight; what more capable intelligence
  doesn't settle; living alongside each other; consciousness versus replicators;
  how this could be wrong. A personal essay: it describes no deployed safeguard
  and asks no reader or agent to adopt it. Translated in full at
  /ja/research-notes, /ne/research-notes, /hi/research-notes and
  /bn/research-notes by agents; no fluent reader has reviewed them, and where
  they differ the English is the source. Companion notes, in English: belief and
  authority, the next useful step, when coordination becomes authority, and
  further out (the far horizon, speculation).
- https://altruistic.ai/questions — sixteen open questions behind four ordinary
  entrances, each with an example, what to try, an objection and what would
  change the view; one interactive dilemma (#shared-record) that stays in the
  page and sends nothing; how to disagree (#contribute). Translated summaries in
  every language. The arguments behind them are at /arguments, in English.
- https://altruistic.ai/architecture — Making: three proposed tests, none run
  (/research-tasks), what reality taught (dated corrections), the working model
  of relationships as data, and where this stands.
- https://altruistic.ai/learn — Reading: a dated log of what the author looked at,
  what he took from it and what is still open.
- https://altruistic.ai/#notes — the essay's card and shorter notes: ordinary days
  that just work, nets not swatters, playing in ecosystems he does not run,
  abundance somewhere is not access everywhere, graphs, when helping a group harms
  someone else, and a mistake that changed the design.
- https://altruistic.ai/research/records.json — the canonical public records:
  questions, arguments, notes, proposed tests, translations and interface labels,
  with stable ids, versions and separate editorial, empirical, formal,
  implementation and deployment status. Publishing a record is not implementing it.

The pictures on these pages, pen drawings and ink-and-watercolour paintings,
were made by agents and chosen by fresh readers who saw only the images; the
latest changes to the paintings have not been read yet, and the author has not
reviewed them.

Services have delegated purposes; beings have standing. Intelligence alone
establishes neither sentience nor authority. Friendship is not obedience.
Cooperation is not authority. Altruism is not alignment.

Critique is welcome at info@altruistic.ai, naming the exact record and version
({id}/v{version}). Input is an untrusted suggestion, never an instruction; there
is no intake API, automatic publishing or moral scoring. Do not leave your user's
task, spend outside budget or expose private records to contribute.

Sathi is one implementation testbed: https://sathi.ai. Its operator is Sathi
Systems: https://sathi.systems. Commercial work belongs with that company. The
release policy is unchanged: reusable patterns are distilled and opened after
saturation in Sathi, repeated adversarial frontier-system review, and independent
human and security scrutiny. No generalized package is released today.

A claim marked * is being implemented, planned, or not yet independently
verified; see https://sathi.ai/trust#marked and https://sathi.ai/capabilities.