Heather Douglas (2000), Philosophy of Science, 67(4), 559–579 — DOI-10.1086/392855
Key takeaways
- Because every internal scientific choice carries inductive risk, non-epistemic values are legitimately and necessarily required — indirectly, via weighing the consequences of error — wherever those consequences can be foreseen, including in choosing significance levels, characterizing data and interpreting results.
Seeded from the abstract and a skim, before the reading in NOTE-145. What follows is what the work says about itself.
Epistemic values are widely accepted in scientific reasoning, but non-epistemic values have mostly been confined to science’s “external” parts: which hypotheses to pursue, ethical limits on methods, and how results are used. Douglas argues that inductive risk, the risk of error, makes non-epistemic values necessary wherever the non-epistemic consequences of error ought to be considered. Using dioxin studies, she shows how such consequences can and should enter the internal stages of science: choice of methodology, characterization of data, and interpretation of results.
Standing in the record
Filed from the 2026-09-26 sweep of the papers-feed tracker for philosophy: 30 seconds of active reading over 1 session. It was filed Deferred, unread. NOTE-145 is the close reading of 2026-09-26, and it placed the work: Active — worth reading as the canonical, compact statement of the modern inductive-risk argument. It is carefully scoped, with an explicit direct/indirect distinction and exemptions for negligible uncertainty and consequence-free research, and it is well illustrated. Its normative premise, that scientists bear moral responsibility for foreseeable consequences of error, is explicitly left unargued (§2, p. 563).
When seeded, before the reading in NOTE-145, the entry gave a priority for a deeper reading: high — 30 s dwell, and one of the most-cited papers in values-in-science. The skim captures the argument, but it is load-bearing for any anthology treatment of evaluation thresholds and should be read in full.
What the seeding asked a deeper reading to check (NOTE-145 is that reading):
- The canonical source for the inductive-risk argument. It applies directly to ML evaluation choices (thresholds, significance levels, labelling ambiguous data, what counts as a capability or a harm) wherever errors have social consequences.
- Check how Douglas separates the legitimate indirect role of values from an illegitimate direct role; her later book (Science, Policy, and the Value-Free Ideal, 2009) develops this.
- Check responses (e.g. Betz; Steel) arguing that uncertainty reporting can avoid value-laden acceptance decisions.
Access when seeded: The captured URL is JSTOR stable/188707, which I did not fetch. The DOI (10.1086/392855) comes from Crossref and Semantic Scholar. The PhilSci-Archive preprint (27688) is behind a captcha. I read the JSTOR-scanned published version (22 pp) hosted as course material on a University of Toronto personal page (individual.utoronto.ca), found by web search. Read the abstract, §1, section heads, the openings of §§3–6 and §7. The December 2000 issue gives no day, so published: is the issue month. Keywords are mine.