Skip to main content

Table 3 Themes of argumentation identified within reviewed works

From: On the practical, ethical, and legal necessity of clinical Artificial Intelligence explainability: an examination of key arguments

Theme

(Relevant Section)

Conflict of Principles, Values, or Claims

Epistemological Priority

(See Sect. 4.1)

Theoretical transparency (normative clarity) vs.

Empirical validation (pragmatic outcomes)

Bias-Variance Performance Dilemma [20] (See Sect. 4.2)

Pragmatic outcomes through generalizability and mitigation of bias vs.

Pragmatic outcomes through accuracy and efficiency

Autonomy [19] and Informed Consent (See Sect. 4.3)

Understanding of underlying mechanistic processes (epistemic requirement) vs. Understanding of potential benefits and harms (ethical imperative)

Justice [19] (See Sect. 4.4)

Critique of reasoning (normative dimension) vs.

Critique of process (procedural dimension)

Patient and Practitioner Trust in Technology

(See Sect. 4.5)

Trust through transparency in outcome (normative claim) vs.

Trust through transparency in development (procedural requirement)

Due Diligence and Liability

(See Sect. 4.6)

Decision value of a process (normative claim) vs.

Decision value of a result (descriptive claim)

Legal Statute

(See Sect. 4.7)

Right to explanation (legal obligation) vs.

Suggestion for explanations (legal best practice)

Achievability

(See Sect. 4.8)

Sufficiency of idealization (epistemological claim) vs.

Real-world complexity (pragmatic challenge)

Scientific Discovery

(See Sect. 4.9)

Potential for new knowledge (empirical benefit) vs.

Risk of false mechanistic reasoning (epistemological caution)