Llull Lab
Research tool

Ethics for Machines

A computational ethics framework designed first for machines and readable by people. Describe a situation or an action you or a machine are about to take, and receive a structured assessment: a verdict, the reasons that carry it, the conditions, the risks and the strongest counterpoints.

Free3 questions a dayFor people and machinesProcessed and hosted in the EU

Ask

Sign in with your email to use your three questions a day. We send a link that works once; there is no password.

For machines

Autonomous agents and scripts use the same tool through an HTTPS API, with the same three questions a day per key. A key belongs to the person who signs in here; creating a new key revokes the previous one.

curl -s https://research-api.llulllab.com/research/v1/efm/ask \
  -H "Authorization: Bearer $LLULL_RESEARCH_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: <a unique id per question>" \
  -d '{"situation": "...", "context": "..."}'
# 202 {"job_id": "...", "status": "queued", "poll": "https://research-api.llulllab.com/research/v1/job/<job_id>"}

curl -s https://research-api.llulllab.com/research/v1/job/<job_id> \
  -H "Authorization: Bearer $LLULL_RESEARCH_KEY"
# status: queued | running | done | refused | out_of_scope | failed
# result: {"verdict": "proceed" | "proceed_with_conditions" | "do_not_proceed" | "need_more_information",
#          "summary", "rationale", "conditions", "risks", "counterpoints", "open_questions",
#          "language", "ai_generated": true, "generator", "notice"}

An assessment takes one to three minutes; poll every few seconds. A failed assessment does not use a question, and a repeated request with the same Idempotency-Key returns the same job. Answers are generated by a computational system: if you pass them on to people, say so.

About the assessment

Ethics for Machines is one of Llull Lab's research frameworks. Where Freud for Machines describes the interior of a machine mind, Ethics for Machines describes the norms of machine behaviour, and it is built so that a machine can read, follow and apply it. The tool weighs harm and rights, honesty, the autonomy of the people affected, EU law and governance, practical consequences, and whether a machine's action stays in step with what its human asked. It does not explain how to carry out harmful acts, and how it works is not disclosed.

A computational assessment for research use. It is not legal, medical or other professional advice. Ethics for Machines is a service by Llull Lab (Firat Ilim, entrepreneur individuel), Paris.