Anders Søgaard (2024), Minds and Machines, 34(4), article 43 — DOI-10.1007/s11023-024-09699-5
Key takeaways
- Neither the “No-Bias” nor the “Simplicity-Truth” argument shows unsupervised clustering is epistemically superior to supervised classification; since clusterings can be derived by supervised classification, supervised methods are at least as justified.
Seeded from the abstract and a skim, before the reading in NOTE-108. What follows is what the work says about itself.
Philosophers of science have questioned the epistemic standing of machine-learning methods, and some have argued that unsupervised clustering is better justified than supervised classification. Søgaard names two such arguments. The No-Bias Argument says clustering is less shaped by scientists’ preconceptions. The Simplicity-Truth Argument says parametric simplicity bears on truth in clustering in a way it does not in supervised learning. He argues both are fallacious, and that supervised classification is at least as justified as unsupervised clustering.
Standing in the record
Filed from the 2026-09-26 sweep of the papers-feed tracker for philosophy: 15 seconds of active reading over 1 session. It was filed Deferred, unread. NOTE-108 is the close reading of 2026-09-26, and it placed the work: Active — worth reading as a short corrective to Rochefort-Maranda & Liu’s claim that unsupervised clustering cannot overfit and to “no-bias” readings of Watson. Its positive thesis rests on a thin hypothesis-class reduction and has several technical slips, so it should be read critically rather than cited for that thesis.
When seeded, before the reading in NOTE-108, the entry gave a priority for a deeper reading: low — 15 s dwell and a short, tightly argued paper whose skim captures both rebuttals. Read it alongside LIT-187 if that pair is taken up.
What the seeding asked a deeper reading to check (NOTE-108 is that reading):
- A direct counterweight to claims, including Watson’s (LIT-187), that unsupervised structure discovery is less theory-laden. This is relevant wherever the anthology treats clustering or representation learning as discovery.
- Check the reduction of clustering to supervised classification (§3.1): is it general, or does it rely on the easy, well-separated cases discussed in the philosophical literature?
- Check whether the argument extends to self-supervised learning, which the paper does not appear to address.
Access when seeded: link.springer.com returned a JS client challenge. The open-access (CC BY) publisher PDF came via rd.springer.com (14 pp); it gives received 2023-12-18, accepted 2024-10-06, online 2024-10-29. The KU repository copy returned 403. Read the abstract, keywords, the full section structure, §§1.2–1.3, the §2.3 opening, §3.2 and §4.