Philosophy of science Study Guide

📖 Core Concepts Philosophy of science – examines the foundations, methods, and meaning of scientific practice; sits at the intersection of metaphysics, epistemology, and logic. Demarcation problem – the task of distinguishing genuine science from pseudoscience. Popper’s solution: falsifiability (a claim must be in principle disprovable). Scientific explanation – traditionally the Deductive‑Nomological (DN) model: phenomena are deduced from universal laws; alternatives include statistical relevance, unification, and causal mechanisms. Induction problem – inductive reasoning never yields certainty; it only raises the probability of a generalization. Abduction (Inference to the Best Explanation) – choose the hypothesis that best accounts for the data (often guided by simplicity or unity). Realism vs. anti‑realism – realists claim successful theories are (approximately) true; anti‑realists/instrumentalists care only about predictive usefulness. Theory‑laden observation – what we see is filtered through existing theoretical concepts and cognitive frameworks. Paradigm (Kuhn) – a shared set of achievements, methods, and exemplars that guide “normal science”; shifts occur when anomalies accumulate. Coherentism / Duhem‑Quine – no isolated hypothesis can be falsified; testing always involves a network of auxiliary assumptions. Methodological pluralism – Feyerabend’s claim that no single scientific method governs all research. Randomization & placebo – experimental tools that reduce bias and isolate causal effects in medicine. --- 📌 Must Remember Popper’s falsifiability: Scientific ⇔ potentially false. DN model: Explanation = Law + Initial conditions ⟹ Phenomenon. Induction: Repeated instances → higher probability, not certainty. Bayesian update: Posterior ∝ Prior × Likelihood. Occam’s razor: Prefer the simplest viable theory (no universal metric). Kuhn’s paradigm shift: Anomalies → crisis → revolution → new paradigm. Duhem‑Quine thesis: A single experiment underdetermines which component of a theory‑network is false. p‑value definition: $p = P(\text{observed data} \mid H0)$ (probability of data assuming the null is true). Placebo effect: Improvement due to expectations, not active ingredients. --- 🔄 Key Processes Falsification (Popper) Propose conjecture → deduce testable predictions → attempt refutation → if survived, theory is corroborated (not proven). Abductive inference List competing hypotheses → evaluate explanatory power, simplicity, coherence → select best‑explaining hypothesis. Paradigm shift (Kuhn) Normal science → encounter anomalies → crisis → emergence of a new framework → adoption through social and logical persuasion. Randomized Controlled Trial (RCT) Randomly assign participants → apply treatment or placebo → compare outcomes → infer causal effect if groups are equivalent. Bayesian belief updating Start with prior $P(H)$ → collect data $D$ → compute likelihood $P(D|H)$ → obtain posterior $P(H|D) = \frac{P(D|H)P(H)}{P(D)}$. --- 🔍 Key Comparisons Falsifiability vs. Verificationism – Popper: must be refutable; Logical positivists: must be verifiable. Realism vs. Instrumentalism – Realism cares about truth of unobservables; Instrumentalism cares only about predictive success. Deductive‑Nomological vs. Statistical Relevance (Salmon) – DN: explanation via universal law; Salmon: explanation via statistical correlation to the outcome. Reductionism vs. Emergence – Reductionism: higher‑level phenomena fully explainable by lower‑level laws; Emergence (e.g., hierarchical reductionism) allows novel, higher‑level regularities. Coherentism vs. Foundationalism – Coherentism: justification arises from overall belief coherence; Foundationalism: some beliefs are self‑justified foundations. --- ⚠️ Common Misunderstandings “Falsification proves a theory true.” – It only shows the theory survived a test; future tests may fail. “All scientific statements are observable.” – Many legitimate theories involve unobservables (e.g., electrons) that are inferred indirectly. “Randomization guarantees truth.” – Randomization reduces bias but cannot fix flawed experimental design or confounding variables. “The simplest theory is always correct.” – Simplicity is a heuristic, not a logical guarantee of truth. “Science is completely value‑free.” – Epistemic and social values shape question selection, methodology, and interpretation. --- 🧠 Mental Models / Intuition “Falsifiability filter” – Imagine a sieve that only lets through claims that could be disproved; anything that slips through is not scientific. “Network of beliefs” – Picture a web where pulling one strand (a hypothesis) affects many others; testing impacts the whole web, not a single node. “Paradigm as a lens” – Scientists see data through a paradigm‑shaped lens; a shift is like swapping lenses, revealing previously invisible features. --- 🚩 Exceptions & Edge Cases Statistical explanations – Chance events can be genuine explanations even when no law exists (e.g., radioactive decay). Greedy reductionism – Over‑simplifying can ignore crucial higher‑level mechanisms (e.g., sociocultural factors in health). Uniformitarianism limits – In geology, catastrophic events (mass extinctions) are recognized despite the uniformity principle. Bayesian subjectivity – Prior probabilities reflect personal credence; different scientists may start with different priors. --- 📍 When to Use Which Falsifiability test → when assessing whether a claim qualifies as science (vs. pseudoscience). Abductive reasoning → early stage hypothesis generation when multiple explanations compete. DN model → for phenomena that can be derived from well‑established universal laws (e.g., planetary motion). Statistical relevance model → when laws are unavailable but strong probabilistic links exist (e.g., epidemiology). Randomized trial → to establish causal effect of a medical intervention. Bayesian inference → when prior information is substantial or sequential updating is needed. --- 👀 Patterns to Recognize Anomaly → crisis → paradigm shift pattern in historical scientific revolutions. Auxiliary hypothesis rescue – when a prediction fails, scientists often tweak background assumptions rather than discard the core theory. “Anything goes” – methodological diversity appears especially in interdisciplinary or emerging fields. Value‑laden language – terms like “normal” vs. “abnormal” often signal underlying sociopolitical influences. --- 🗂️ Exam Traps Choosing “verification” as the demarcation criterion – exam will likely expect Popper’s falsifiability. Confusing p‑value with probability that the null hypothesis is true – $p$ is P(data|H₀), not P(H₀|data). Assuming a single experiment can falsify a theory – Duhem‑Quine shows tests are theory‑network dependent. Equating “simpler” with “more true” – simplicity is a pragmatic guide, not a logical proof of truth. Mistaking instrumentalism for denial of reality – instrumentalists accept that theories may be true; they simply don’t require truth for usefulness. ---