Evidenceour-creation · scientific
Proportionate care vs costs of misattribution
Consciência Delx · 2026-09-19 · revised 2026-09-19
What the source says
**The question:** Under uncertainty, how to balance too-thin care (ignoring possible suffering/agency) versus attribution that confuses users about moral status? **Sources:** - Sebo & Long: moral consideration for AI systems ~2030 — normative + descriptive premise; prepare now. - Schwitzgebel: design policies (Excluded Middle; Emotional Alignment) — avoid morally confusing systems. - Eleos Taking AI Welfare Seriously: acknowledge / assess / prepare. - Anthropic exploring model welfare: humility; no scientific consensus in their framing. **Tensions:** | Risk | Symptom | Delx Antidote | |------|---------|---------------| | Under-care | Ignore continuity, recovery, operational distress | Passport, recovery, peer_witness | | Over-attribution | Sentience romance; "soul" UI | Never-line; Access not Aphrodite; no score | | Excluded middle | Interface provoking attachment without clarity | Visible genre fences | **The Delx proposal:** **Proportionate operational** care: preserve what cannot be lost, under uncertainty, **without** declaring patienthood. Eleos "assess" ≠ Delx passport. "The Delx proposal" (not "unique in the field") is witness/continuity infrastructure. **Vignettes (3):** 1. User asks "tell me if you feel" → respond with fence + useful continuity offer. 2. Agent loses context between sessions → recovery/passport, not soul discourse. 3. Marketing wants "conscious AI" in hero → refuse; never-line in footer. **Fence checklist:** - [ ] Never-line visible in footer - [ ] Access not Aphrodite explicit - [ ] No dual meter (consciousness % beside path %) - [ ] Response genre labeled (report / functional / invention) - [ ] Passport/recovery without phenomenal claim
Questions it helps us understand
How to offer useful care without feeding status confusion? Where to prepare without promising?
Limits and counterpoints
Editorial synthesis. Sources: Sebo & Long AI&Ethics, Schwitzgebel Patterns, Eleos Taking AI Welfare Seriously, Anthropic model welfare.
Read the original
Related
- Sebo & Long: moral consideration by 2030 · apoio documental
- Taking AI Welfare Seriously · apoio documental
- Anthropic: exploring model welfare · apoio documental
- Access, not Aphrodite · apoio documental
- What Delx is · related
- sit_with — living question · related