Skip to content

Epistemology & Wissenschaftstheorie

This document details the epistemological foundations, philosophy of science criteria, and metatheoretical models that justify the architecture of Episteme. It draws primarily from the formal epistemology of Gerhard Schurz (2014), Imre Lakatos' methodology of scientific research programmes (1978), the structuralist framework of Joseph Sneed and Wolfgang Stegmüller (1976), and Paul Thagard's computational model of explanatory coherence (1989).

Methodological Characteristics of Scientific Theories

In formal philosophy of science (Wissenschaftstheorie), an isolated theoretical axiom (e.g., Newton's \(F = m \cdot a\)) possesses virtually no empirical content on its own; it cannot be directly verified or falsified in isolation. Empirical content and explanatory power emerge strictly through the systemic architecture of the theory.

Schurz (2014) identifies four primary methodological criteria that distinguish genuine scientific theories from speculative or ad-hoc systems:

System Character and Empirical Creativity

The systemic interaction of multiple hypotheses generates empirical creativity (Empirische Kreativität): two or more theoretical hypotheses combined in a system yield novel empirical content that strictly exceeds the set-theoretic union of their isolated empirical contents.

  • Formal Operationalization: Let \(E(H)\) denote the empirical content (observable deductive consequences) of a hypothesis \(H\). A theoretical system exhibits empirical creativity if and only if:

[ E(H_1 \wedge H_2) \supset E(H_1) \cup E(H_2) ]

When evaluating extracted theoretical premises from scholarly literature, the pipeline verifies whether introducing a new premise into conjunction with the background theory deduces novel observable consequences. If it does not, it represents empirical deadweight or metaphysical surplus.

Globality and Empirical Unification

A hallmark of progressive science is globality (Globalität): the capacity of a theory to explain qualitatively disparate empirical phenomena (e.g., planetary orbits, ocean tides, and free fall in Newtonian mechanics) through the identical theoretical mechanism. This unification power (Vereinheitlichungsleistung) reduces an unmanageable multitude of isolated empirical generalizations to a concise set of fundamental principles.

  • Formal Operationalization: Unification cannot be operationalized merely by counting post-hoc explanations of already known data. Instead, it is measured by the theory's capacity to generate novel predictions—predicting qualitatively new phenomena of which no instance had previously been observed (e.g., predicting that artificial satellites can maintain a stable orbit). In our graph representation, unification is reflected in high out-degree centrality of core axioms connecting across diverse empirical clusters.

Holism of Meaning and Testing (Duhem-Neurath-Quine)

Because empirical content emerges only globally within the system, the meaning of a theoretical term is determined by the entire network of interconnected laws (Holism of Meaning).

From this follows the Holism of Theory Testing (the Duhem-Neurath-Quine thesis): an empirical anomaly refutes via Modus Tollens only the global conjunction of all involved premises; it does not indicate which specific premise is false. There is no definitive experimentum crucis.

  • Formal Operationalization: Because isolated axioms cannot be falsified individually, we operationalize theory evaluation comparatively over the set algebra of empirical successes (\(E\)) and empirical failures (\(M\)). A theory version \(T_1\) represents rational scientific progress over \(T_2\) if and only if:

[ E(T_1) \supseteq E(T_2) \quad \wedge \quad M(T_1) \subset M(T_2) ]

This creates a strict partial order of theoretical superiority. Merely counting raw numbers of successes is epistemologically invalid; progress requires expanding verified successes while strictly shrinking the domain of anomalies.

Homogeneity vs. The Tacking Paradox (Klebeparadoxon)

Karl Popper famously demanded "boldness" in scientific theories in the form of maximized unexamined empirical content. However, unconstrained maximization leads to a fatal logical trap: the Tacking Paradox (Klebeparadoxon).

If one takes an empirically well-confirmed physical theory \(T\) and conjoins an arbitrary, untested claim \(H\) (e.g., "Telepathy exists"), the conjunction \(T \wedge H\) strictly possesses a vastly increased empirical content. Yet this is not progressive science; it is trivial conjunctive padding.

  • Formal Operationalization: We demand that a scientific theory must be homogeneous (non-factorizable, nicht-faktorisierbar). A theory \(T\) is unacceptably heterogeneous (factorizable) if it can be decomposed into two disjoint sub-theories \(T_1\) and \(T_2\) such that its empirical content \(E(T)\) also decomposes into disjoint sets \(E_1\) and \(E_2\):

[ T \equiv T_1 \wedge T_2 \quad \text{such that} \quad E(T) = \operatorname{Cn}(E_1 \cup E_2) \quad \text{with} \quad T_1 \vdash E_1, \; T_2 \vdash E_2 ]

An increase in empirical content is methodologically legitimate only when newly added axioms logically interact with the existing core, preserving structural and semantic homogeneity.

Schurz's Four-Dimensional Theory Statics (Theorienanalyse)

To decompose complex philosophical and scientific texts into machine-processable theory graphs without losing structural integrity, the pipeline implements Gerhard Schurz's four-dimensional framework of theory statics:

flowchart TD
    subgraph D1 ["1. Linguistic Basis (Vocabulary)"]
        LE["Pre-theoretical Empirical Language (L_E)"]
        LT["Specific Theoretical Language (L_T)"]
    end

    subgraph D2 ["2. Logical Architecture"]
        AX["Irreducible Natural Axioms Ax(T)"]
        CN["Derived Consequence Set C(T)"]
    end

    subgraph D3 ["3. Semantic Stratification"]
        P_TH["Pure Theoretical Sentences (L_T only)"]
        MIX["Mixed Sentences / Zuordnungsgesetze Z(a,b)"]
        EMP["Empirical / Pre-theoretical Consequences (L_E only)"]
    end

    subgraph D4 ["4. Epistemic Status (Lakatosian Dynamics)"]
        CORE["Hard Theory Core (K_i)"]
        PERI["Protective Belt / Periphery (V_j)"]
    end

    D1 --> D2 --> D3 --> D4

Linguistic Basis (Das Vokabular)

The terminology of the theory is sharply separated into two distinct sub-vocabularies:

  • Pre-Theoretical Empirical Language (\(L_E\)): Terms presupposed as already understood and empirically grounded (observation terms or terms established by unproblematic background theories).
  • Theoretical Language (\(L_T\)): Novel, theory-specific constructs introduced by the theory itself (e.g., quarks, subconscious, categorical imperative).

Logical Architecture (Axiome vs. Konsequenzen)

The deductive statement system is organized into:

  • Natural Axioms (\(Ax(T)\)): The irreducible, non-redundant core laws that define the theory. Redundant conjunctive appendages must be eliminated to prevent artificial inflation.
  • Consequence Set (\(\operatorname{Cn}(T)\)): The set of all theorems, explanations, and predictions logically or probabilistically entailed by natural axioms (\(Ax(T) \vdash \operatorname{Cn}(T)\)).

Semantic Stratification (Satzarten)

Axioms and consequences are classified by their semantic composition:

  • Pure Theoretical Sentences: Principles formulated entirely within \(L_T\) (e.g., Newton's actio = reactio).
  • Mixed Sentences (Zuordnungsgesetze): Crucial bridge principles connecting \(L_T\) with \(L_E\). Without explicit mapping laws, theoretical terms remain empirically disconnected and decay into speculative metaphysics.
  • Empirical Consequences: Testable assertions formulated strictly in \(L_E\) that establish contact with observable phenomena.

Epistemic Status & Theory Dynamics (Hard Core vs. Periphery)

Following Imre Lakatos, theories are structured into differential epistemic strata:

  • The Hard Core (\(K_i\)): Axioms that define the historical and paradigm identity of the theory. If these are abandoned, the paradigm collapses.
  • The Protective Belt / Periphery (\(V_j\)): Auxiliary hypotheses, boundary conditions, and specialized parameters surrounding the core. When anomalies occur, Modus Tollens is redirected to modify peripheral hypotheses in the protective belt, shielding the hard core from premature refutation.

Lakatosian Theory Dynamics & The Degeneration Index

When scientific theories evolve over time across multiple text versions or historical editions, Episteme evaluates whether the evolution represents a progressive or a degenerating research programme.

Successes vs. Failures

Empirical interactions for a given theory version \(V_j\) are partitioned into:

  • Empirical Successes (\(S\), or \(E\) in German Erfolg): Phenomena correctly predicted or explained by the theory.
  • Failures Type a (\(F_{\text{Typ a}}\), or \(M_a\) in German Misserfolg Typ a - Direct Refutations): Well-established empirical observations in direct logical contradiction with the theory.
  • Failures Type b (\(F_{\text{Typ b}}\), or \(M_b\) in German Misserfolg Typ b - Ad-Hoc Immunizations): Anomalies that contradicted an earlier version of the theory, but were "resolved" in the current version by introducing an ad-hoc auxiliary hypothesis that produces no novel testable predictions (\(\operatorname{Cn}(K_{i+1}) \setminus \{a\} \subseteq \operatorname{Cn}(K_i)\)).

The Degeneration Index (\(D\))

Schurz operationalizes Lakatos's criterion through the Degeneration Index \(D(V_j)\):

\[ D(V_j) = \frac{|F_{\text{Typ b}}(V_j)|}{|S(V_j)| + |F_{\text{Typ a}}(V_j)|} \quad \left( \equiv \frac{|M_b|}{|E| + |M_a|} \right) \]
  • Progressive Programme (\(D \to 0\)): New versions expand empirical coverage, resolve anomalies via hypotheses that generate verified novel predictions, and keep ad-hoc immunizations minimal.
  • Degenerating Programme (\(D \gg 0\)): Anomalies are continuously patched by circular auxiliary clauses that shield the core without expanding empirical creativity, turning the theory into an empirical patchwork.

Explanatory Coherence (Thagard's ECHO Model)

To evaluate the mutual acceptability of contested hypotheses within the extracted Theory Graph, Episteme implements principles from Paul Thagard's Theory of Explanatory Coherence (TEC):

  1. Symmetry: If proposition \(P\) explains \(Q\), then \(P\) and \(Q\) cohere with each other symmetrically (\(w_{pq} = w_{qp}\)).
  2. Explanatory Breadth: Hypotheses that explain a larger volume and diversity of empirical evidence receive greater excitatory support.
  3. Simplicity: The network penalizes complex, convoluted explanations. The excitatory weight between a hypothesis and the explained data is inversely proportional to the number of auxiliary co-hypotheses required for the deduction.
  4. Unification: Core hypotheses that repeatedly explain diverse data without requiring case-by-case ad-hoc assumptions achieve high global coherence.
  5. Higher-Level Warrant: A hypothesis gains substantial plausibility if it is itself explained by a deeper, overarching theoretical principle.
  6. Data Priority: Verified empirical data nodes possess intrinsic acceptability, providing an independent source of activation energy that spreads through excitatory links.
  7. System Coherence (\(H\)): The global mathematical harmony of the entire network at state \(t\) is calculated via: [ H(t) = \sum_{i} \sum_{j} w_{ij} \cdot a_i(t) \cdot a_j(t) ] Where \(w_{ij}\) represents excitatory or inhibitory edge weights and \(a_i(t)\) represents continuous node activations.

Avoiding Retrospective Bias (Presentism)

A severe challenge in computational digital humanities and philosophy is retrospective bias (Presentism): the tendency to analyze historical theories using modern concepts, formalisms, or measurement instruments that were unavailable to the historical authors.

Episteme enforces historical neutrality: - The graph structure must reflect the author's native conceptual framework and explicitly stated background assumptions. - Theoretical claims are evaluated relative to the historical empirical base (\(B(t)\)) available at the time of authorship, rather than modern ground-truth datasets.

References & Bibliography

  1. Schurz, G. (2014). Philosophy of Science: A Unified Approach. Routledge.
  2. Lakatos, I. (1978). The Methodology of Scientific Research Programmes: Philosophical Papers Volume 1. Cambridge University Press.
  3. Sneed, J. D. (1976). The Logical Structure of Mathematical Physics. D. Reidel Publishing Company.
  4. Stegmüller, W. (1976). The Structure and Dynamics of Theories. Springer-Verlag.
  5. Thagard, P. (1989). "Explanatory Coherence." Behavioral and Brain Sciences, 12(3), 435–467.
  6. Thagard, P. (1992). Conceptual Revolutions. Princeton University Press.
  7. Quine, W. V. O. (1951). "Two Dogmas of Empiricism." The Philosophical Review, 60(1), 20–43.
  8. Duhem, P. (1906). La théorie physique: son objet, et sa structure. Marcel Rivière.