PhilSci: Tutorials

Overview

Luka Vinck runs the tutorials and posts his slides after each one. They are mostly discussion prompts rather than content, which makes them more useful than they look: they are the TA telling you which questions he thinks are worth an hour of a room’s time, and they are shaped almost exactly like exam questions.

This note collects every tutorial question, with an answer, because you are not in the room to hear one. It also holds the authoritative course plan, which comes from tutorial 1 and is more complete than anything else on Canvas.

Week 3’s tutorial on 17 September was the one with graded attendance (bring a project idea). The project proposal is due Friday 25 September, 20:00, and both the week 3 and week 4 slides end on that reminder.

Course plan

From tutorial 1, slide 5. This is the most complete schedule published anywhere in the course.

Week	Topic	Project milestone
1	Logical Empiricism
2	Kuhn on Scientific Practice
3	The Duhem-Quine Thesis	Writing workshop. Come prepared with a project idea
4	Explanation and Understanding	Project proposal due Friday 25 September
5	Scientific Realism 1	Feedback
6	Scientific Realism 2
7	Laws of Nature	Presentations, plus first draft and slides due Sunday 18 October
8	Exam
		Final project due 31 October

Two things this adds that were not in the vault before: week 7 is Laws of Nature, and the presentations happen in that same week 7 session rather than on a separate date.

The project

• A philosophical research paper on one of the course topics.
• Start from an existing discussion in the contemporary philosophical literature. Not from scratch.
• Groups of 1 to 3 people. Roughly 4000 to 7000 words per person, all inclusive, so a group is more coordination for the same amount of your own writing.

Luka’s suggested way in: read the Stanford Encyclopedia of Philosophy entry for a course topic, look at the example papers on Canvas, then look at recent papers in Synthese, Studies in the History and Philosophy of Science, the European Journal for Philosophy of Science, the British Journal for the Philosophy of Science, or Erkenntnis.

Tutorial 1, 3 September: What is Science, and Logical Empiricism

Framing

Philosophy of science asks: what is science, what is good science, why and how does science work, and what does science teach us about nature. It sits across epistemology (what can we know), metaphysics and ontology (what is there), and, Luka adds, politics (what counts as truth).

The questions, answered

What were the central ideas of early logical positivism? Which ideas from the philosophy of language inspired it? How did these underpin its philosophy of science?

Meaningful factual claims are those that can in principle be verified by observation, and logic and mathematics are analytic, true by form and empty of factual content. The inspiration is Frege’s first-order logic, Russell and Whitehead’s logicism, and Wittgenstein on the relation between language and the world. Together these underpin a philosophy of science whose job is to clarify the language of science rather than compete with it: philosophy analyses, science discovers. Full detail in PhilSci-L01 - Introduction and Logical Empiricism.

What is science according to the logical positivists?

The body of statements that are either empirically verifiable or analytically true, expressed in a language whose theoretical terms are connected to observation. Anything else is not false but meaningless.

What was Carnap’s critique of metaphysics? Which central idea does he use? Do you agree?

Carnap argues metaphysical statements are pseudo-statements: they look like assertions and have no method of verification, so they say nothing. The idea he uses is the verifiability criterion of meaning. The standard objection, worth having ready, is that the criterion is self-undermining, since it is neither analytic nor empirically verifiable itself.

How was the early programme challenged, internally and externally?

Internally by the difficulty of stating the verification criterion precisely, the problem of theoretical terms, and Quine’s attack on the analytic-synthetic distinction (PhilSci-L03 - Under-determination). Externally by Popper substituting falsifiability for verifiability, and later by Kuhn’s historical turn (PhilSci-L02 - Kuhn on Scientific Practice).

How did later logical empiricism differ from early logical positivism?

It softened. Verification gave way to weaker confirmation, the demand that theoretical terms reduce to observation was relaxed, and the syntactic conception of theories gave way to the semantic one, on which a theory is a class of models rather than an axiomatised language.

Popper: what makes a theory scientific? How does falsifiability relate to verificationism? Which is more accurate?

Falsifiability, and the mark of pseudo-science is that it explains everything and forbids nothing. The relation to verificationism is the point worth being precise about, because it is a common exam trap: they are different criteria doing different jobs. Verificationism is a theory of meaning, so unverifiable statements are meaningless. Falsifiability is a criterion of demarcation, and Popper is explicit that unfalsifiable statements can be perfectly meaningful, just not scientific. Metaphysics, for Popper, is meaningful and sometimes fruitful. Full treatment in PhilSci-L01b - Popper and Lakatos.

Tutorial 2, 10 September: Science as a Socio-Historical Practice

Covers Kuhn on revolutions, Lakatos on demarcation, and Longino on values and objectivity.

Kuhn

The cycle on Luka’s slide: normal science within a paradigm, stubborn anomalies, crisis or extraordinary science, revolution and a new paradigm.

Why does Kuhn argue that “cumulative acquisition of novelty is not only rare in fact but improbable in principle” (p. 89)? Do you agree?

Because normal science is puzzle-solving inside a paradigm, and a paradigm determines what counts as a legitimate problem and a legitimate solution. Genuine novelty is by definition what the paradigm does not anticipate, so it can only appear as an anomaly, which is a failure rather than a contribution. Novelty therefore arrives by rupture, not accumulation.

Why are arguments in paradigm debates necessarily circular?

Each side argues from within its own paradigm, and paradigms supply the standards by which arguments are judged. So each can only defend its position using criteria the other does not accept. Kuhn’s comparison on the slide: “Like the choice between competing political institutions, that between competing paradigms proves to be a choice between incompatible modes of community life” (p. 88).

Why are subsequent paradigms incommensurable, and do you agree?

Because they differ in problems, standards and the meanings of their terms, so there is no neutral vocabulary in which to state a fair comparison. The standard pushback: incommensurability is often overstated, since scientists in practice do communicate across paradigm boundaries and do cite empirical successes both sides recognise.

What is the necessity of scientific revolutions?

They are necessary because normal science cannot correct its own foundations. Anomalies accumulate until the paradigm’s puzzle-solving capacity is exhausted, and only a change of paradigm can restore it. Revolution is the mechanism by which science changes its mind at all.

Lakatos

The demarcation problem is not solved by Popper or Kuhn.

Luka’s slide states Lakatos’s charge directly. Popper is naive, because scientists do not in fact abandon a theory the moment it meets a contrary fact. Kuhn makes science irrational, because on his account change of paradigm is a change in commitment with no rational criterion, and so no demarcation is possible at all. Lakatos’s answer is to shift the unit of appraisal from the theory to the research programme.

What makes a research programme scientific? How does Lakatos nuance Popper and modify Kuhn?

A programme is scientific when it is progressive, meaning it predicts novel facts, some of which are corroborated. It is degenerating when it only accommodates facts already known. He nuances Popper by appraising a series of theories over time rather than a single theory against a single test, which makes room for the tenacity scientists actually show. He modifies Kuhn by keeping the historical unit while restoring a rational criterion for choosing between programmes. Detail in PhilSci-L01b - Popper and Lakatos.

Do you agree that scientific value is an “objective” property of a theory?

The case for: progressiveness is assessed against the public record of predictions and corroborations, not anyone’s opinion. The case against: judging a programme progressive requires deciding how long to wait before calling it degenerating, and Lakatos gives no principled cut-off, so the verdict is only available with hindsight.

Longino

Her position on the slide: the objective character of science is a consequence of its being a social enterprise. Objectivity is a property of a community’s practice, dependent on the depth and scope of transformative interrogation, and it does not differ in kind for non-empirical sciences, contra the logical empiricists.

What produces scientific knowledge, and how can it be objective? What are the necessary conditions?

Not the individual, but the community subjecting individual work to criticism. Knowledge is what survives transformative interrogation. See PhilSci-L02 - Kuhn on Scientific Practice for the four conditions in full.

The distinction the lecturer went out of his way to make

De Haro posted an announcement on 8 September specifically to correct a question asked in class. Longino’s contextual empiricism is a theory of scientific objectivity and epistemic methodology. It is not an ethical theory, and it does not answer whether a piece of research is permissible.

Ethical permissibility is the business of consequentialism, deontology, virtue ethics, care ethics and theories of justice, which ask about harms and benefits, rights and consent, and fair distribution of risk. Longino is at most indirectly relevant, in that she encourages scrutiny of the assumptions built into research and of whose perspectives are included in criticism.

A lecturer writing an unprompted announcement to correct a confusion is telling you where he thinks the confusion lies. Do not conflate these two in the exam.

Tutorial 3, 17 September: Underdetermination and Holism, plus the writing workshop

The session with graded attendance. First half a writing workshop, then a break, then the text discussion on Quine and Laudan. The last slide is the reminder: project proposal due Friday 25 September.

Writing workshop

Luka opened by asking which of the course authors so far (Carnap, Godfrey-Smith, Popper, Lakatos, Longino, Kuhn, Quine, Laudan) had a style you particularly liked or disliked. The point being that you have now read eight philosophers and have opinions about how philosophy should be written.

His two references: Erik Curiel (2025), “Notes on Learning Philosophy”, and Harvard’s “A Brief Guide to Writing a Philosophy Paper”.

The advice, slide by slide. Analytic philosophy: argumentation is everything. Your paper should have a clear thesis, be well structured, explain concepts and arguments appropriately, and use clear and concise language.

• Thesis. The start of everything. It should not be too “big”. It should be relevant and well embedded in the literature: take an existing discussion as your starting point.
• Structure. Start by making an argumentation scheme: that scheme is the structure of your paper. Clear sections and paragraphs. Each paragraph contains one point.
• What goes where.
  • Introduction: state your position, outline the route to the conclusion, demarcate what you are and are not doing.
  • Beginning of a section: explain its role in the overall structure.
  • Beginning of an argument: state the underlying assumptions and define the technical terms.
  • End of an argument: consider possible limitations and objections.
  • Conclusion: summarise the argument, indicate consequences and future directions.
• Explanation has two aims: make sure your reader understands, and show that you understand. Help the reader often. Explain for the target audience.
• Language. Clear and concise. Be precise: take a moment to think about what you actually want to say. Avoid overstatements (“completely”, “extremely”, “absolutely”).

The hands-on exercise, in groups. If you have a topic but no thesis: pick a small number of recent papers on the topic, summarise each one’s main argument in your own words, work out how they relate, and see whether you can take a position with respect to them. If you already have a thesis and literature: draft the argumentation structure that will become the outline of the paper, with the relevant literature attached to each argument.

That second exercise is also, not by accident, most of what a project proposal is.

The questions, answered

What is the problem of underdetermination according to Quine, in short? And according to Laudan?

For Quine, our theory of the world is underdetermined by all possible observations: there can be theory formulations that are empirically equivalent (they imply the same observation sentences) but theoretically inequivalent, so no possible evidence decides between them. For Laudan, the problem is a family of theses that get run together, and he separates them. The non-uniqueness thesis says that for any theory there is at least one rival as well supported by the evidence. The egalitarian thesis says every theory is as well supported as any rival. Only the egalitarian thesis would make theory choice arbitrary, and Laudan argues nothing in Quine establishes it. Detail in PhilSci-L03 - Under-determination, sections 4 and 5.

Why does Quine say that underdetermination is a thesis about the world?

Because it is not a claim about our current evidence running out, which more evidence could fix (that is transient underdetermination). It is a claim about the relation between any theory of the world and all the observations the world could ever supply: the totality of possible observations is itself compatible with irreconcilable systems. So it says something about what the world, as experienced, leaves open, rather than about the limits of a particular experiment. needs-review: this answer is reconstructed from the lecture, not from Quine’s text, which is not in the vault. Check the wording against On Empirically Equivalent Systems of the World before relying on it.

What is a “theory” according to Quine? Reconstruct his principle of individuation for theories.

Quine separates a theory formulation (a sentence, in practice the conjunction of a theory’s axioms) from the theory it expresses. Two formulations count as the same theory when they are empirically equivalent and one can be converted into the other by reconstrual of predicates: systematically swapping predicates for other predicates or open sentences, as when two physical theories differ only by exchanging the words “molecule” and “electron”. Empirically equivalent formulations that no reconstrual can bring into agreement are different theories, and those are the cases underdetermination needs. The same idea sits behind the “common core” quotation in PhilSci-L03 - Under-determination: if a theory is nothing more than the conjunction of its observation conditionals, rivals share it rather than compete with it.

What is the logical relation between theory and empirical data?

Implication runs one way. The theory, together with standing conditions, implies observation conditionals (“if this, then that”). The data never imply the theory. And because of holism the implication comes from the theory as a whole, not from individual hypotheses: single sentences of theory usually have no observation consequences of their own.

Bonus: why does Quine say that there is no “extra-theoretic” truth?

Truth is immanent: we can only assess truth from within some theory, our own best system of the world, since there is no standpoint outside all theories from which to compare a theory with reality directly. So when faced with an empirically equivalent rival, “which one is really true?” has no answer from outside, and we call our own theory true while it remains our theory. needs-review: reconstructed, same caveat as above.

Why does Quine argue that holism (the Duhem-Quine thesis) lends credence to underdetermination? And why does Laudan argue that it doesn’t?

Quine: if hypotheses face experience only as a whole, then a recalcitrant observation can be accommodated by revising any of many sentences in the web. Since there is always more than one way to repair the system, it seems there must always be more than one system that fits all the data. Holism makes rival fits look everywhere available.

Laudan’s critique, reconstructed:

1. Holism shows that any hypothesis can be retained in the face of any evidence, by adjusting something else. That is a claim about logical possibility.
2. Underdetermination, in the form that threatens methodology, is a claim about rational warrant: that the evidence supports rivals equally (the egalitarian thesis).
3. To get from (1) to (2) you would need to show that the ampliative rules of theory choice are too weak to prefer one repair over another. Quine never examines those rules: he argues in an “evaluative vacuum”.
4. So holism supports at most non-uniqueness (some rival exists), and that is harmless, because choosing among rivals is exactly what methodology is for. It does nothing for the egalitarian thesis.

This is the Laudan passage the L03 note puts side by side with Quine, and it is the standard reply to anyone who uses underdetermination to argue that theory choice is arbitrary.

Tutorial 4, 24 September: Explanation and Understanding in Science

Two texts, split by a break: Godfrey-Smith on explanation, then De Regt and Dieks on understanding. Lecture material is in PhilSci-L04 - Scientific Explanation and Understanding; the answers below are the tutorial questions in the order Luka asked them. Last slide: next week is scientific realism, with Matteo, and the project proposal is due Friday 25 September.

Godfrey-Smith: Explanation

What is the importance of the notion of explanation for philosophy of science? And why is it problematic?

Important because science seems to tell us not just what happens but why, and because explanation is tied to evidence: scientists prefer theories that explain puzzling phenomena well, and a lot of scientific inference is explanatory inference, from data to a hypothesis about the hidden structure that would explain it. So an account of explanation feeds straight into an account of evidence. Problematic because empiricists have long distrusted it: explanation sounds like a claim to deep metaphysical insight into hidden causes, which is exactly what they wanted science to do without. The positivists only accepted it by making it low-key.

What is the logical empiricists’ theory of explanation (the “covering law” or D-N model)? What is its main challenge?

To explain something is to derive it in a logical argument whose premises include at least one law of nature, used essentially, plus particular conditions. Explaining is showing that the phenomenon was to be expected, so explanation and prediction have the same structure. The main challenge is the asymmetry problem: the flagpole’s height explains the shadow’s length, and the same laws let you deduce the height from the shadow, which explains nothing. Godfrey-Smith calls it close to a knockdown argument. Full treatment in PhilSci-L04 - Scientific Explanation and Understanding, sections 2 and 3.

What are the two main alternative ideas on explanation, and their challenges?

Causation: to explain is to describe what caused the phenomenon (Salmon). It fixes the flagpole at once, since sunlight and pole cause the shadow. Challenge: it needs an account of causation, which empiricists since Hume regard as suspect, and it seems too narrow for explanations that are hard to state causally. Unification: to explain is to fit phenomena into a broader pattern, reducing the number of facts we must accept as brute (Friedman, Kitcher). Challenge: it has to recover the flagpole asymmetry without appealing to causes, which Kitcher tries to do by arguing causal talk is a loose summary of deeper unification asymmetries.

Which notion of explanation do you find most convincing, or do you agree with Godfrey-Smith’s pluralist or contextualist view?

The case for Godfrey-Smith: causation and unification each handle cases the other struggles with, and their own champions ended up conceding ground to each other (Salmon accepted unification, Kitcher accepted causation). He goes one step further than pluralism: there is no fixed list of explanatory relations at all, because each field sets its own standards, as Kuhn argued with Newton’s gravity. It is not “anything goes”, since a standard can embed a factual error. The case against: if “explanation” means different things in different fields, it is not obvious what makes them all explanation, and contextualism risks describing practice without saying what makes an explanation good. De Regt and Dieks’s answer, after the break, is that understanding is the common aim that unifies the plural forms of explanation.

Is explanation crucial to the inner workings of science, or something external to it (cf. van Fraassen’s pragmatic account)?

Van Fraassen: external. Explanation is a three-term relation between theory, fact and context, an application of science to answer our questions, and it adds nothing to what we believe about the world. Godfrey-Smith: internal. Assessing what explains what is an important part of scientific reasoning; fields just differ in the standards they use. Both say explanation is context-dependent, which is why the question is worth asking: the agreement on context hides a disagreement on whether explanation is epistemic. The stakes show next week: if explanation is internal and epistemic, inference to the best explanation can support belief in unobservables, which is what van Fraassen’s constructive empiricism must deny.

De Regt and Dieks: Scientific Understanding

The framing slide: the goal is a general characteristic of scientific understanding that can account for the variation in actual historical practice. Understanding plays a unifying role in a pluralist view of explanation, and it is pragmatic, meaning context-dependent.

What is the relation between explanation and understanding?

Understanding a phenomenon requires an explanation of it, but is more than having one: it also requires the ability to construct and use explanations, which is a skill. Explanation is the product; understanding is the achievement of a scientist who can produce it. That is why understanding can unify plural kinds of explanation: causal, unificatory and law-based explanations are all ways a skilled scientist can come to understand a phenomenon.

“Contra Hempel, van Fraassen, and Trout, we hold that the pragmatic nature of understanding is not inconsistent with it being epistemically relevant” (p. 141). What is the epistemic relevance of understanding, and do you agree? What are the conditions for scientific understanding of a theory?

Hempel treats “pragmatic” as “subjective” and so as epistemically irrelevant; van Fraassen makes explanation a non-epistemic application of science; Trout reduces the sense of understanding to cognitive bias. De Regt and Dieks separate two senses of “pragmatic”: subjective, and useable for certain aims. In the second sense, understanding is epistemically relevant: it is an aim of science in its own right (a macro-level aim), and it is needed to produce the explanations and models through which science gets knowledge at all. The conditions for understanding a theory are its intelligibility: a theory is understood, in the relevant sense, by scientists who can use it, as measured by the CIT below. Whether you agree turns on whether you accept that “able to use” is an epistemic achievement rather than a practical one.

CUP: a phenomenon is understood scientifically iff there is an explanation of it based on an intelligible theory that conforms to empirical adequacy and internal consistency. CIT: a theory T is intelligible for a scientist in a context if they can recognise qualitatively characteristic consequences of T without performing exact calculations. Does this reduce understanding to a subjective affair?

No. The criterion is relative to scientists in a context, but the context is a scientific community with shared, trained standards of intelligibility, the meso-level in De Regt and Dieks’s three-level picture. Whether a physicist can recognise qualitative consequences without calculating is a public, testable fact about her skills, not a feeling. What it does make understanding is contextual: relative to a community and its level of progress. The pressure point is cases like Schrödinger against Heisenberg in 1926, where two groups in the same discipline disagreed about what was intelligible. There, contextual starts to look a lot like subjective, and the defence is that the disagreement was settled by the community over time, in the way Pauli predicted. Full version in PhilSci-L04 - Scientific Explanation and Understanding, section 7.

What kinds of conceptual toolkit help scientists with qualitative reasoning? At what level do these operate?

Visualisation and diagrams (Feynman diagrams, orbital pictures), causal-mechanical stories, analogies, thought experiments, toy models, idealised models and simulations. In an MSc AI setting: loss landscapes, attention-map visualisations and scaling-law plots are the same kind of thing, tools that let you anticipate what a model will do without running it. They are taught and shared within a discipline, so they operate at the meso-level of the scientific community, and individual scientists at the micro-level have them to different degrees.

De Regt and Dieks argue that understanding is an “additional epistemic aim of science” (p. 153). Is their account convincing?

For: it explains the historical variation (Newton, Huygens, Maxwell; Schrödinger versus Heisenberg) that a single fixed standard of explanation cannot, and it answers Trout, since a skill that produces empirically adequate models is not a feeling. Against: it is not obvious that a skill is an epistemic aim rather than a means to the real epistemic aims of truth or empirical adequacy. If understanding just is the ability to build good models, a reductivist like Khalifa can say that this is knowledge plus competence, and nothing needs adding to the list of aims.

Key Takeaways

Exam Focus

The tutorial questions are shaped like exam questions because the TA is drawing on the same material. Two of these map directly onto the mock exam:

• Q4 asks why Kuhn compares revolutions to Gestalt switches. The tutorial’s circularity and incommensurability questions are the same territory.
• Q5 asks what Longino means by the social character of objective knowledge, how something becomes scientific knowledge, and for examples of social mechanisms. The tutorial 2 Longino questions are that question in three parts.

The Popper-versus-verificationism question from tutorial 1 is the one most likely to catch you out, because the intuitive answer (“they are two versions of the same idea”) is wrong. Verificationism is about meaning, falsifiability is about demarcation, and Popper thinks unfalsifiable claims can be perfectly meaningful.

Tutorial 3 on 17 September (past)

Graded attendance, worth 10% of the paper grade and 3% of the course grade. Bring a project idea. Not a finished proposal, an idea you can defend for two minutes. The two that fit your existing work: Duhem-Quine underdetermination applied to LLM benchmark evaluation, or explanation and understanding applied to explainable AI.

The second of those is now covered: PhilSci-L04 - Scientific Explanation and Understanding has De Regt and Dieks’s criterion for intelligibility, which is the natural hook for an explainable-AI paper (“is a model intelligible to the people who use it, by the CIT?”).

The tutorial 3 and 4 questions most likely to recur

• Laudan against Quine (tutorial 3): holism shows a hypothesis can be saved, not that rivals are equally well supported. Logical possibility is not rational warrant.
• Van Fraassen against Godfrey-Smith (tutorial 4): both say explanation is contextual, but for van Fraassen it is external to science and non-epistemic, and for Godfrey-Smith it is internal to scientific reasoning. That difference is what mock question 8 turns on.

Links

• Course: Course overview
• Lectures: PhilSci-L01 - Introduction and Logical Empiricism · PhilSci-L01b - Popper and Lakatos · PhilSci-L02 - Kuhn on Scientific Practice · PhilSci-L03 - Under-determination · PhilSci-L04 - Scientific Explanation and Understanding
• Source: the four tutorial decks on Canvas, all now in Assets/: Tutorial slides week 1 - What is Science and Logical Empiricism.pptm, Tutorial slides week 2 - Scientific Practice.pptm, Tutorial slides week 3 - Underdetermination and Holism + writing tutorial.pptm and Tutorial slides week 4 - Explanation and Understanding.pptm (weeks 3 and 4 uploaded 2026-09-24), plus De Haro’s announcement of 2026-09-08. Tutorial 2 was re-checked against its deck on 2026-09-24 and matches.