Source
Schwitzgebel & Sinnott-Armstrong (2026), “Sacrificing Humans for Insects and AI”, LIT-166 (Ethics 136(3), 670–696), read in full in NOTE-116 from the authors’ draft of 3 September 2025. The version of record was not reached, so page numbers are the draft’s.
What was actually shown
A conceptual argument in §2.1, aimed at Birch’s The Edge of Sentience.
- The thesis. “By itself, an appeal to precaution does not specify the type of error most to be avoided” (p. 8). “Err on the side of caution” can mean protecting a creature that may be sentient, or protecting human liberty and practice.
- Two precautions in conflict. Mill’s harm principle, read as a precaution in favour of liberty, puts the burden of proof on whoever would restrict liberty. Birch’s framework puts it on whoever risks harming a sentience candidate. The reviewers “see no reason always to prioritize” one over the other (p. 11).
- Why the choice is not free. Birch’s “risks are not equal” argument (Birch p. 17) assumes that a false negative costs more than a false positive. Assessing that “turns on the very science at issue”, since the cost of wrongly denying sentience depends on how much, and what kind of, sentience there is.
- Worked cases. Fishing, bathrooms and an ant-colony trolley (§2.1).
What could have come out otherwise: a reading of precaution on which one error is worse whatever the facts about sentience turn out to be. The reviewers find none in Birch’s text, and give a general reason there is none.
What this does not say
- That precaution toward animals or AI is wrong. The argument is that a choice of error preference must be made and defended, not that the harm-precaution is the wrong choice. The reviewers take no side.
- That Birch’s books say what the review says they say. The record holds the review, not the books. Its page references to Birch are the reviewers’ (NOTE-116, limitations).
- That the review’s other charges stand as firmly. The charges that sentience candidature is policed without principle, that citizens’ panels cannot be both inclusive and deliberative, and that Sebo’s live views imply two ants over a stranger are rated moderate in the reading. They are not part of this account.
- That there is no principled way to choose. NOTE-116 leaves this open: is there a non-question-begging way to choose between harm-precaution and liberty-precaution when they conflict?
Connections
- The conditional the same review defends, that common grounds of moral standing put human priority under threat, is filed separately as THEORY-554.
- Terrill & Veit (THEORY-449) take a non-negligible chance of plant quasi-sentience to give a pro tanto obligation, under “some version of a confidence threshold, expected value principle, and/or the precautionary principle”. That step needs the error preference this account says precaution does not supply.
- Birch’s AI-specific programme (LIT-111) and the indicator method it rests on (LIT-056) are where an AI precaution would be stated. Vieira (LIT-126) rejects precaution toward AI outright, which is one side of the choice this account says is not made by precaution alone.