What Comes After Falsification? New Epistemologies

KARL POPPER'S FALSIFICATIONISM (1934)

THE DEMARCATION PROBLEM: ┌─────────────────────────────────────────┐ │ Question: What makes science scientific?│ │ ↓ │ │ Not: "Can you prove it?" │ │ (Nothing provable with certainty) │ │ ↓ │ │ Instead: "Can you disprove it?" │ └─────────────────────────────────────────┘

THE CRITERION: ┌─────────────────────────────────────────┐ │ Scientific theory = Falsifiable theory │ │ ↓ │ │ Must make predictions that could be │ │ proven wrong │ │ ↓ │ │ Must be testable in principle │ │ ↓ │ │ If no conceivable test can refute it: │ │ NOT SCIENCE │ └─────────────────────────────────────────┘

EXAMPLES: ┌─────────────────────────────────────────┐ │ SCIENTIFIC (Falsifiable): │ │ • "All swans are white" │ │ → Find black swan = falsified │ │ • "Light bends near massive objects" │ │ → Measure during eclipse = test │ │ ↓ │ │ NOT SCIENTIFIC (Unfalsifiable): │ │ • "God exists" │ │ → No test can disprove │ │ • "Everything happens for a reason" │ │ → Explains everything, predicts │ │ nothing │ └─────────────────────────────────────────┘

THE APPEAL: ┌─────────────────────────────────────────┐ │ Clear criterion │ │ ↓ │ │ Separates science from pseudoscience │ │ ↓ │ │ Captures scientific practice: Test │ │ theories, reject failures │ └─────────────────────────────────────────┘

CHALLENGES TO FALSIFICATIONISM

PROBLEM 1: STRING THEORY ┌─────────────────────────────────────────┐ │ Most promising quantum gravity theory │ │ ↓ │ │ But: Makes no testable predictions │ │ (at accessible energies) │ │ ↓ │ │ Unfalsifiable = not science? │ │ ↓ │ │ But: Logically consistent, solves │ │ problems, guides research │ │ ↓ │ │ Feels like science even if unfalsifiable│ └─────────────────────────────────────────┘

PROBLEM 2: MULTIVERSE ┌─────────────────────────────────────────┐ │ Follows from inflation theory (tested) │ │ ↓ │ │ But: Other universes beyond horizon │ │ (can't observe) │ │ ↓ │ │ Unfalsifiable by definition │ │ ↓ │ │ Yet: Logical consequence of good theory │ │ ↓ │ │ Reject because untestable? Or accept as │ │ implication? │ └─────────────────────────────────────────┘

PROBLEM 3: AI BLACK BOXES ┌─────────────────────────────────────────┐ │ Neural networks make accurate │ │ predictions │ │ ↓ │ │ But: Can't explain HOW (black box) │ │ ↓ │ │ Can't formulate falsifiable hypothesis │ │ (no explicit mechanism) │ │ ↓ │ │ Yet: Produces scientific knowledge │ │ (protein structures, drug candidates) │ └─────────────────────────────────────────┘

PROBLEM 4: COMPLEX SYSTEMS ┌─────────────────────────────────────────┐ │ Climate models, ecosystems, brains │ │ ↓ │ │ Too complex to test every prediction │ │ ↓ │ │ "Validated" not "falsified" │ │ ↓ │ │ Many parameters, partial tests │ │ ↓ │ │ Doesn't fit clean falsificationism │ └─────────────────────────────────────────┘

PROBLEM 5: HISTORICAL SCIENCES ┌─────────────────────────────────────────┐ │ Evolution, cosmology, geology │ │ ↓ │ │ Can't repeat past (one history) │ │ ↓ │ │ Can't "test" in experimental sense │ │ ↓ │ │ Yet: Scientific by consensus │ │ ↓ │ │ Different methodology than physics │ └─────────────────────────────────────────┘

BAYESIAN APPROACH

THE SHIFT: ┌─────────────────────────────────────────┐ │ FROM: "Can we falsify?" │ │ ↓ │ │ TO: "How probable given evidence?" │ └─────────────────────────────────────────┘

BAYES' THEOREM: ┌─────────────────────────────────────────┐ │ Update belief based on evidence: │ │ ↓ │ │ P(Theory|Evidence) = │ │ P(Evidence|Theory) × P(Theory) / │ │ P(Evidence) │ │ ↓ │ │ Start with prior probability │ │ ↓ │ │ Update as evidence accumulates │ └─────────────────────────────────────────┘

HOW IT WORKS: ┌─────────────────────────────────────────┐ │ Never "prove" or "disprove" │ │ ↓ │ │ Instead: Increase or decrease confidence│ │ ↓ │ │ Example: │ │ • Evidence for evolution: P → 99.9% │ │ • Evidence against flat Earth: P → 0.001%│ │ ↓ │ │ Science = accumulating probabilistic │ │ warrant │ └─────────────────────────────────────────┘

ADVANTAGES: ┌─────────────────────────────────────────┐ │ • Handles uncertainty naturally │ │ • No sharp falsification required │ │ • Captures actual scientific practice │ │ • Works for probabilistic evidence │ └─────────────────────────────────────────┘

DISADVANTAGES: ┌─────────────────────────────────────────┐ │ • Subjective priors (starting beliefs) │ │ • No clear demarcation (what │ │ probability = science?) │ │ • Can rationalize pet theories │ └─────────────────────────────────────────┘

INSTRUMENTALISM

THE IDEA: ┌─────────────────────────────────────────┐ │ Science's goal: Predict phenomena │ │ ↓ │ │ NOT: Reveal "truth" │ │ ↓ │ │ Theory = instrument for prediction │ └─────────────────────────────────────────┘

CRITERIA FOR GOOD THEORY: ┌─────────────────────────────────────────┐ │ • Accurate predictions │ │ • Broad scope │ │ • Simplicity/elegance │ │ ↓ │ │ NOT required: │ │ • Correspondence to "reality" │ │ • Mechanistic explanation │ │ • FalsifiabilityThe property of a claim that allows it to be tested and possibly proven wrong. A claim that cannot fail is not yet in contact with reality. │ └─────────────────────────────────────────┘

EXAMPLE: QUANTUM MECHANICS ┌─────────────────────────────────────────┐ │ Predicts with incredible accuracy │ │ ↓ │ │ But: No one agrees what it "means" │ │ ↓ │ │ Copenhagen? Many-worlds? Pilot wave? │ │ ↓ │ │ Doesn't matter for predictions │ │ ↓ │ │ "Shut up and calculate" (instrumentalism│ │ in practice) │ └─────────────────────────────────────────┘

EXAMPLE: AI SCIENCE ┌─────────────────────────────────────────┐ │ Neural network predicts protein │ │ structure accurately │ │ ↓ │ │ We don't understand how │ │ ↓ │ │ Instrumentalist view: Doesn't matter │ │ ↓ │ │ It works = sufficient │ └─────────────────────────────────────────┘

STRENGTHS: ┌─────────────────────────────────────────┐ │ • Pragmatic │ │ • Sidesteps metaphysics │ │ • Works for black-box AI │ └─────────────────────────────────────────┘

WEAKNESSES: ┌─────────────────────────────────────────┐ │ • Abandons understanding │ │ • Multiple incompatible theories could │ │ predict equally well │ │ • Unsatisfying (science = prediction │ │ machine?) │ └─────────────────────────────────────────┘

COHERENCE THEORY

THE PRINCIPLE: ┌─────────────────────────────────────────┐ │ Theory is scientific if: │ │ ↓ │ │ • Coheres with established knowledge │ │ • Multiple independent lines support it │ │ • Explains diverse phenomena │ └─────────────────────────────────────────┘

CONSILIENCE (E.O. WILSON): ┌─────────────────────────────────────────┐ │ "Jumping together" of knowledge │ │ ↓ │ │ Different fields converge on same │ │ conclusion │ │ ↓ │ │ Example: Evolution supported by: │ │ • Fossils (paleontology) │ │ • DNA (genetics) │ │ • Anatomy (comparative biology) │ │ • Biogeography (distribution patterns) │ │ ↓ │ │ Consilience = strong warrant │ └─────────────────────────────────────────┘

APPLICATION TO UNFALSIFIABLE THEORIES: ┌─────────────────────────────────────────┐ │ String theory: │ │ • Coheres with quantum mechanics │ │ • Coheres with relativity │ │ • Unifies forces │ │ ↓ │ │ Even if untestable, coherence gives │ │ epistemic warrant │ └─────────────────────────────────────────┘

HISTORICAL SCIENCES: ┌─────────────────────────────────────────┐ │ Can't re-run history │ │ ↓ │ │ But: Multiple lines of evidence converge│ │ ↓ │ │ Example: Asteroid killed dinosaurs │ │ • Iridium layer (chemistry) │ │ • Crater dating (geology) │ │ • Fossil record (paleontology) │ │ ↓ │ │ Consilience substitutes for repeatability│ └─────────────────────────────────────────┘

BEYOND TESTABILITY

THE CONTROVERSIAL CLAIM: ┌─────────────────────────────────────────┐ │ Some theories scientific even if │ │ empirically untestable │ │ ↓ │ │ Criteria: │ │ • Logical consistency │ │ • Explanatory power │ │ • Mathematical beauty │ │ • Solves theoretical problems │ └─────────────────────────────────────────┘

THE ARGUMENT (Dawid, Ellis, Silk): ┌─────────────────────────────────────────┐ │ String theory may never be testable │ │ ↓ │ │ But: Only known quantum gravity theory │ │ ↓ │ │ Solves theoretical inconsistencies │ │ ↓ │ │ Mathematically consistent │ │ ↓ │ │ This is enough for scientific status │ └─────────────────────────────────────────┘

THE BACKLASH: ┌─────────────────────────────────────────┐ │ Critics: "This abandons empiricism!" │ │ ↓ │ │ Post-empirical = not science, just math │ │ ↓ │ │ Slippery slope to anything goes │ │ ↓ │ │ Defenders: We're already at physical │ │ limits, need new criteria │ └─────────────────────────────────────────┘

HISTORICAL PRECEDENT? ┌─────────────────────────────────────────┐ │ Einstein's relativity (1905-1915): │ │ • Developed largely on theoretical │ │ grounds │ │ • Predicted perihelion of Mercury, light│ │ bending │ │ • But motivation = resolve theoretical │ │ tensions │ │ ↓ │ │ Theory-driven, then tested │ │ ↓ │ │ Post-empirical claim: What if testing │ │ impossible? │ └─────────────────────────────────────────┘

REDEFINING SCIENCE

THE CLASSICAL DEFINITION (Popper): ┌─────────────────────────────────────────┐ │ Science = Falsifiable theories │ │ ↓ │ │ Clear, but excludes much modern science │ └─────────────────────────────────────────┘

ALTERNATIVE 1: METHODOLOGICAL ┌─────────────────────────────────────────┐ │ Science = Systematic inquiry using │ │ rigorous methods │ │ ↓ │ │ Broad, but what methods? (too vague) │ └─────────────────────────────────────────┘

ALTERNATIVE 2: SOCIOLOGICAL ┌─────────────────────────────────────────┐ │ Science = What scientists do │ │ ↓ │ │ Descriptive, but circular (who are │ │ scientists?) │ └─────────────────────────────────────────┘

ALTERNATIVE 3: PRAGMATIC ┌─────────────────────────────────────────┐ │ Science = Knowledge that works reliably │ │ ↓ │ │ Instrumentalist, but abandons truth │ └─────────────────────────────────────────┘

ALTERNATIVE 4: PLURALIST ┌─────────────────────────────────────────┐ │ Science = Family resemblance concept │ │ ↓ │ │ Different sciences use different methods│ │ ↓ │ │ No single criterion defines all │ │ ↓ │ │ Physics ≠ Biology ≠ History, all science│ └─────────────────────────────────────────┘

THE UNCOMFORTABLE TRUTH: ┌─────────────────────────────────────────┐ │ No sharp line between science and │ │ non-science │ │ ↓ │ │ Spectrum of epistemic warrant │ │ ↓ │ │ Better question: "How reliable?" not │ │ "Is it science?" │ └─────────────────────────────────────────┘

THE STRING THEORY DEBATE

THE SITUATION: ┌─────────────────────────────────────────┐ │ 50 years of development │ │ ↓ │ │ Thousands of physicists │ │ ↓ │ │ Zero experimental confirmations │ │ ↓ │ │ No testable predictions at accessible │ │ energies │ └─────────────────────────────────────────┘

CASE FOR "SCIENCE": ┌─────────────────────────────────────────┐ │ • Only known quantum gravity framework │ │ • Resolves theoretical inconsistencies │ │ • Predicts gravitons │ │ • Unifies forces │ │ • Mathematically consistent │ │ ↓ │ │ Guided by physical principles │ └─────────────────────────────────────────┘

CASE FOR "NOT SCIENCE": ┌─────────────────────────────────────────┐ │ • Untestable (in practice, maybe │ │ principle) │ │ • Makes no novel predictions │ │ • Unfalsifiable │ │ • Many possible versions (10^500 │ │ vacua) │ │ ↓ │ │ Math, not physics │ └─────────────────────────────────────────┘

THE IMPASSE: ┌─────────────────────────────────────────┐ │ If falsifiability = science criterion: │ │ String theory NOT science │ │ ↓ │ │ But: Community treats it as science │ │ ↓ │ │ Either: │ │ • Physicists doing non-science, or │ │ • Falsificationism wrong │ └─────────────────────────────────────────┘

THE AUTHORITY PROBLEM

IF WE KEEP STRICT FALSIFICATIONISM: ┌─────────────────────────────────────────┐ │ Much frontier physics = not science │ │ ↓ │ │ String theory, multiverse = philosophy │ │ ↓ │ │ But: Loses research guidance, unifies │ │ knowledge │ └─────────────────────────────────────────┘

IF WE ABANDON FALSIFICATIONISM: ┌─────────────────────────────────────────┐ │ What distinguishes science from │ │ pseudoscience? │ │ ↓ │ │ Without clear criterion, slippery slope │ │ ↓ │ │ "My unfalsifiable theory is science too"│ │ (astrology, homeopathy, etc.) │ └─────────────────────────────────────────┘

THE BALANCE: ┌─────────────────────────────────────────┐ │ Need criteria flexible enough for modern│ │ science │ │ ↓ │ │ But rigorous enough to exclude nonsense │ │ ↓ │ │ Difficult balance │ └─────────────────────────────────────────┘

THE PRAGMATIC ANSWER: ┌─────────────────────────────────────────┐ │ Science = community practice │ │ ↓ │ │ Peer review, institutional quality │ │ control │ │ ↓ │ │ Imperfect, but works │ │ ↓ │ │ Trust process, not single criterion │ └─────────────────────────────────────────┘