PhilSci - Topics and Stances
What this is
The big questions of the course, one or two per lecture: what the question is, who answers it how, and what the fight between them is about. Detail on each person is in PhilSci - Philosophers; full model answers are in the lecture notes’ exam questions.
What makes something science?
Weeks 1 and 2. Mock Q1, Q2, Q3.
Astrology, psychoanalysis and physics can all point at evidence, so evidence alone cannot be the mark. This is the problem of demarcation.
- Verificationism (Carnap, Vienna Circle): meaningful means analytic or verifiable. Non-science that pretends to state facts is meaningless. The line runs between sense and nonsense.
- Falsificationism (Popper): scientific means falsifiable, able to clash with possible observation. The line runs between science and non-science, and non-science can still be meaningful. Risky predictions count; confirmations are cheap.
- Research programmes (Lakatos): judge a programme over time. Progressive (predicts novel facts that are found) is science working; degenerating (only accommodates) is pseudoscience creeping in.
- Puzzle-solving tradition (Kuhn): a science has normal science, a community working on puzzles. Astrology’s failures never became puzzles.
The fight: verification cannot handle universal laws (no finite evidence proves “all”), which falsification can. Falsification ignores that scientists rightly keep refuted theories (Lakatos’s thick skins, theories born refuted) and that a refutation never says which premise failed (Duhem-Quine). Lakatos’s answer only works in hindsight.
Saving a theory from a bad result (mock Q3): Popper allows blaming the measurement only with independent reasons, else it is ad hoc. Lakatos treats patching the protective belt as normal, and asks whether the patch predicts anything new. Both accept the invented planet (Neptune) and reject Marxism’s theory of imperialism.
What is a scientific theory?
Week 1.
- Syntactic view (Received View, logical empiricists): a set of sentences in a formal language: axioms, derived sentences, correspondence rules linking theoretical terms to observation. Kinetic theory derives .
- Semantic view: a theory is a class of models. Motivated by theories never axiomatised, one theory having several formulations, and scientists working with models.
- Verdict in the course: the difference may not survive (Halvorson, Lutz), or the two are complementary (Frigg: semantic still needs a language).
How does science change, and is it rational?
Week 2, with Lakatos from week 1. Mock Q4.
- Logical empiricists and Popper: a normative picture: knowledge accumulates, theories rise and fall by evidence and logic.
- Kuhn: a descriptive picture: normal science, anomaly, crisis, revolution. Revolutions are non-cumulative and change meanings and standards (incommensurability). Observation is theory-laden (the Gestalt switch), so paradigm debates are circular and settled by persuasion.
- Lakatos’s charge: that makes theory change irrational, “mob psychology”. His own account keeps revolutions rational: scientists leave degenerating programmes for progressive ones.
- Kuhn’s reply: choice is guided by shared values (accuracy, consistency, scope, simplicity, fruitfulness). They are imprecise and conflict, so choice is not an algorithm, and rational disagreement is possible and useful.
The fight: whether “no neutral algorithm” means “no rationality”. Kuhn says rationality without an algorithm is still rationality. Later, De Haro (week 6) defuses “different worlds” by saying the intension of a term changes while its extension stays.
Can science be value-free?
Week 2. Mock Q5.
- Value-free ideal (Weber, logical positivists): values may pick the research agenda and steer applications, never the research itself (the triptych).
- Weak value-free ideal: epistemic values (accuracy, simplicity, fruitfulness) are allowed inside research, non-epistemic ones are not. Kuhn’s five values fit here.
- Douglas: under uncertainty, non-epistemic values set how much evidence is enough, because errors have costs (inductive risk). Values set the bar without becoming evidence.
- Longino: values enter the background assumptions that make data count as evidence. Objectivity comes from the community’s transformative criticism (four conditions; replication, peer review), so knowledge has a social character.
The fight: whether the middle panel can be kept clean. Douglas and Longino both say non-epistemic values can legitimately be inside (internality positions), at different places.
Under-determination
Week 3. Mock Q6, second half of Q3.
The question: when a prediction fails, which belief is wrong, and does the evidence ever settle which theory to believe?
- Duhem: a hypothesis is tested only with auxiliaries, so a failed prediction refutes the conjunction. No crucial experiment. Good sense picks the culprit, which is not a formal rule.
- Quine: the bundle is everything we believe, logic included (web of belief). Any statement can be held true come what may. Later: empirically equivalent but theoretically different theories, though real cases are an open question.
- Laudan: logically possible is not rational; Quine argues in an evaluative vacuum. Only the egalitarian thesis (every theory equally supported) would make choice arbitrary, and non-uniqueness is harmless.
- Poincaré: geometry is a convention. The sphere world fits Euclidean geometry plus shrinking bodies, or non-Euclidean geometry, and no experiment decides.
- Kinds: semantic (what the theories even say) versus epistemic (what evidence settles); transient (later evidence may decide, common) versus permanent (no possible evidence decides, Quine’s thesis).
The fight: whether “you can always keep your theory” means “theory choice is arbitrary”. Duhem and Laudan say no. For realism, permanent under-determination would leave you not knowing which theory to be a realist about, but in week 6 it was judged a stalemate with little force: equivalent rivals are easy as toy models and rare in real physics.
What is a scientific explanation?
Week 4. Mock Q7.
- Covering law (Hempel): deduce the phenomenon from laws plus conditions; it was to be expected. Explanation and prediction are symmetric.
- Against Hempel: flagpole (asymmetry), hexed salt (irrelevance), barometer (common cause), the window (no law needed), syphilis (low probability). The shared lesson: causation is missing.
- Causal (Salmon): explanation shows the causes. Fixes the flagpole; inherits Hume’s problem and struggles with non-causal physics.
- Unification (Friedman, Kitcher): explanation fits facts into wide patterns, cutting brute facts. Handles non-causal cases; has to rebuild the flagpole asymmetry from unification.
- Contextualism (Godfrey-Smith): no single explanatory relation; standards partly depend on field and period. Explanation is internal to science.
- Pragmatics (van Fraassen): a three-term relation of theory, fact and context. Explanation is an application of science and not an epistemic aim, which matters for realism (IBE).
The fight: what explanation adds beyond deduction (causes or patterns), and whether it is part of science’s aim at all.
What is scientific understanding?
Week 4.
- Eliminativism (Hempel): understanding is psychological (“pragmatic” meaning subjective), so epistemically irrelevant.
- Reductivism (Khalifa, Lipton, Trout): understanding is just having explanations or knowledge (Lipton: knowledge of causes); the feeling of understanding is bias (Trout).
- Skill view (De Regt and Dieks): understanding is the skill to use a theory to build models that explain. It needs an intelligible theory: CUP (understood iff explained by an intelligible, empirically adequate, consistent theory) and CIT (intelligible if you can see qualitative consequences without calculating). Standards of intelligibility are set by communities and change over time (Huygens on Newton; Schrödinger against Heisenberg).
The fight: whether “pragmatic” means “subjective”. De Regt and Dieks separate subjective from “useable for certain aims”, and say only the first is epistemically irrelevant.
Scientific realism and anti-realism
Week 5. Mock Q8.
Should you believe what our best theories say about unobservables like electrons, or only what they say about what we can observe?
- Scientific realism: well-confirmed theories are (approximately) true, unobservables included. Three parts: semantic (read literally), epistemic (we are justified), metaphysical (the world is as the theory says).
- Mach and the early positivists: claims about unobservables have no truth value (Mach) or no meaning (positivists); theories are tools. The positivists needed a sharp line between observational and theoretical, and Grover Maxwell showed it cannot be drawn.
- Constructive empiricism (van Fraassen): science aims at empirical adequacy; accepting a theory means believing only that. Agnostic about unobservables.
For realism:
- No-miracles argument (Putnam): success at novel predictions (Higgs, top quark) would be a miracle if theories were false.
- Inference to the best explanation: believe the best explanation; the NMA is one instance.
Van Fraassen’s replies: “observable” is vague but usable; a rival hypothesis (we believe the best explanation is empirically adequate); explanation is not an aim of science; theories succeed because only successful ones survive.
Musgrave on explanation (mock Q8): van Fraassen confuses the demand for explanation with the demand for ultimate explanation. Non-ultimate explanations relocate and enhance puzzlement and are still real explanations, so science aims to explain without an unlimited demand. The attack hits essentialism and misses realism.
Against realism, the pessimistic meta-induction (Laudan): successful theories with non-referring central terms (aether, phlogiston) were not approximately true, so success does not indicate truth. Saatsi: read it as a prediction about current theories, or, harder to resist, as showing success is no reliable sign of truth.
Answering the pessimistic meta-induction
Week 6. No mock question, but De Haro sets the exam and this is his research.
- Only mature theories with novel predictions count (Psillos, Boyd): shrinks Laudan’s list. Stanford: arbitrary and post hoc; caloric and phlogiston were well confirmed.
- Reference case by case (Kitcher): “aether” sometimes referred to the field. Stanford: the theories were still radically wrong, and realism needs approximate truth.
- Divide et impera (Psillos): believe the causal core that produced the success; idle parts like the material aether drop out. Stanford: a Pyrrhic victory, because the split is drawn with hindsight and cannot tell us which parts of today’s theories to believe.
- Structural realism (Worrall): believe the retained mathematical structure. Objection: structure may not fix an ontology.
- Extensional realism (De Haro): separate intension from extension. Be a realist about extensions within the domain of application. Successive theories are extensionally equivalent there (predictive, material, conceptual correspondence), so the aether refers: to the electromagnetic field plus a fixed frame. No theory parts are sorted, so no Pyrrhic victory; approximate truth is extensional truth on a domain.
Are we in the same position as past scientists? Laudan and Hesse: yes, no privilege. Psillos: no, methods improved and many posits survived. Stanford: yes, hindsight will judge us. De Haro: past scientists were justified only within their domains, and there their beliefs mostly survived.
What is a law of nature?
Week 7. No mock question, but syllabus topic 6.
The puzzle: “all gold spheres are under a mile across” is a true universal generalisation and no law. Laws seem to make things true: they back counterfactuals (the match lights, the copper coin stays copper), fix what is nomologically possible and necessary, and decide whether a theory is deterministic.
- Non-Humeans: there is irreducible necessity in the world.
- Primitivism: Maudlin’s laws produce later states from earlier ones (FLOTEs); objections are time-symmetric laws and constraint laws like the Einstein equations. Chen’s minimal primitivism: primitive laws govern by constraining possibilities, any form allowed.
- Universals (Dretske, Tooley, Armstrong): laws are necessitation relations between universals. Says what laws are; constraint laws do not fit, and the necessitation is put in by hand.
- Dispositions: laws derive from dispositions like fragility. Says what laws are; conservation laws fit badly.
- Humeans, the Best Systems Account (Lewis): laws are the axioms of the best balance of strength and simplicity summarising the mosaic of local facts. The world determines the laws. No mystery, close to scientific practice; trouble with entanglement and empty spacetime, and the patterns stay unexplained.
The fight: do laws govern the world, or summarise it? Non-Humean views can explain the patterns but posit something mysterious; Humean views avoid the mystery and leave the patterns brute.