“What’s the point of demarcation, and when does it mislead?”
Introduction
In philosophy of science, the demarcation problem asks what separates science from non-science (or pseudoscience). The “point of demarcation” is whatever criterion (or set of criteria) one uses to draw that boundary. Demarcation matters because it supplies shared, publicly contestable standards for testing, criticism, and correction.
But it can mislead when treated as a thin, fixed line: it can become circular, harden into dogma, and exclude empirically responsible work that does not yet fit mainstream styles of explanation. Drawing on Alan Chalmers “What is this thing called Science? “(Chs 1–7), I argue that demarcation is best understood as a semi-permeable boundary—a “Markov blanket” metaphor[1] —firm enough to filter noise and motivated reasoning, yet open enough to admit anomalies, new methods, and revision. (I use the metaphor externally; it is not Chalmers’ term.) On this view, the boundary should be strict about shared constraints (testability, criticism, error-correction), but permissive about the sources and forms of hypotheses.
Why Demarcation Matters
Science without demarcation is like a membraneless cell. Without a semi-permeable membrane, homeostasis fails and form dissolves into surrounding noise. Likewise, without boundary conditions, there is no stable way to decide what counts as evidence, what gets tested, and what should be discarded when it fails. Demarcation matters because it keeps inquiry coherent and error-sensitive rather than sliding into persuasion, wishful thinking, or a grab-bag of unfalsifiable claims.
One reason demarcation matters is that vague theories can be made to “fit” almost anything. As Chalmers notes, a vaguely stated theory “can always be interpreted so as to be consistent with the results of those tests” (Chalmers, p. 63). Popper’s falsificationism is one attempt to make the boundary practical: a good hypothesis takes risks—it is testable in ways that could show it wrong—and it survives only so long as it is not falsified (Chalmers, p. 69). Chalmers’ bats example illustrates the pressure toward precision: a loose idea such as “bats navigate by sight” has to be tightened, revised, or abandoned when tests show navigation can persist even when vision is blocked (Chalmers, pp. 64–66). The lesson is not that observation is irrelevant, but that demarcation pressures us toward clarity and vulnerability: hypotheses should be stated in ways that make failure possible, and test outcomes should genuinely constrain what we may continue to believe.
Even so, the very standards that make science robust can mislead if they are treated as a simple label applied to a supposedly final body of facts.
When Demarcation Misleads
Demarcation becomes misleading when treated as a thin, final “stamp,” because the evidential base of science isn’t a fixed pile of neutral facts. What counts as a “result” depends on instruments, methods, and interpretation—often shaped by background theory—so evidence can end up quietly tailored to fit the framework it is supposed to test. That creates the risk of circularity: theory helps certify the facts, and the certified facts then “support” the theory.
A concrete case is the cathode-ray dispute. Hertz reported that cathode rays could not be deflected by an electric field perpendicular to their direction of motion. Later, J. J. Thomson established deflections under improved experimental conditions—especially improved vacuum conditions—so that effects previously masked by residual gas could be observed (Chalmers, pp. 30–31). The point is not merely that technology improves, but that what counts as a “result” can shift with method and conditions, so treating “the facts” as final can mislead.
Chalmers also shows how apparent confirmation can be built into measurement. In a school experiment, students plotted coil deflection against “current” and expected a straight-line graph; but the current was read from an ammeter whose needle movement relies on the same basic principle (deflection in a magnetic field). The relationship under test was partly presupposed in the instrument, so the confirmation was less independent than it looked (Chalmers, pp. 35–36).
Demarcation as a Markov Blanket
The Markov blanket is a boundary metaphor (borrowed from probabilistic modelling, but used here only as an analogy): a minimal, selective boundary that screens off noise while defining the channels through which outside information can update what’s inside. Unlike a minimalist line that merely separates, a “blanket” separates and regulates.
Applied to demarcation, this suggests a boundary that is thin at entry but thick at endorsement. “Thin” means placing fewer constraints on where ideas may come from: anomalies, reports, and conjectures shouldn’t be rejected simply because they arise from non-standard frameworks. “Thick” means greater constraint where scientific standing is earned: through shared procedures of testing, criticism, and correction.
Contemporary work on consciousness at the boundary illustrates the difference: near-death experience reports can motivate new questions and improved measurement, but interpretations should remain provisional when evidential channels are limited[2] . By contrast, research on covert consciousness uses rigorous EEG/fMRI paradigms to sometimes detect command-following in behaviourally unresponsive patients—an example of a “thick” evidential channel upgrading what we can responsibly claim without jumping to metaphysics[3] . In blanket terms: let signals in, but only through defined testing channels.
Counterargument and Response
A natural worry about “thin at entry” is that it becomes a loophole: if we let anything in—rumour, ideology, crank theories—the boundary collapses and science gets devalued. In practice, time and funding are limited even for legitimate work.
Response: Thin entry is not endorsement. It is a rule about triage and curiosity, not acceptance. A candidate claim only earns scientific standing once it is translated into something testable, exposed to criticism and testing, and held to revision or abandonment under failure. Resources should track evidential traction, not novelty. This is the blanket metaphor in action: open to signals at the boundary, strict about the channels that upgrade a signal into knowledge.
Conclusion
Demarcation matters because science needs shared, publicly contestable constraints that enable error-correction. It misleads when treated as a thin stamp on “the facts,” since results are fallible, revisable, and partly shaped by instruments and interpretation. A semi-permeable boundary works better: thin at entry, thick at endorsement. The payoff is discipline without dogma—science stays open to revision without dissolving into “anything goes.”
Notes
[1] Judea Pearl, Probabilistic Reasoning in Intelligent Systems: Networks of Plausible Inference (San Mateo, CA: Morgan Kaufmann, 1988). (Source for “Markov blanket”….)
[2] Sam Parnia et al., “Awareness during Resuscitation – II…,” Resuscitation 191 (October 2023): 109903, https://doi.org/10.1016/j.resuscitation.2023.109903.
[3] Martin M. Monti et al., “Willful Modulation…,” New England Journal of Medicine 362, no. 7 (2010): 579–589, https://doi.org/10.1056/NEJMoa0905370.