Background: Contemporary debates on disinformation are dominated by two influential approaches: Fallis’s functional (teleological) model and Simion’s purely epistemic Disinformation as Ignorance-Generating Content (DIGC) model. Their respective strengths and weaknesses become salient in real-world settings such as the Backfire Effect, the spread of bullshit, and disputes about the epistemic status of Large Language Models (LLMs).
Aims: We aim to (i) critically evaluate the explanatory and classificatory utility of the functional and purely epistemic models across these scenarios, (ii) diagnose key failure modes (especially DIGC’s over-generation and the functional model’s difficulties with non-intentional sources), and (iii) propose a more extensionally adequate framework.
Methodology: We conduct a comparative conceptual analysis of both models and test their classifications against several cases. In particular, we use empirical findings on the Backfire Effect to examine whether a purely consequence-based criterion misclassifies accurate, well-intentioned scientific information. We also incorporate Frankfurt’s distinction between lying and bullshit to refine how epistemic malice is characterised.
Findings: Fallis’s functional model captures complex forms of disinformation (including true and adaptive disinformation) by tying disinformation to a misleading function, but it struggles to classify outputs from non-intentional sources such as autonomous AI. DIGC broadens coverage by removing intentionality and focusing on dispositions to increase ignorance, yet this purely epistemic stance yields an Over-generation Problem: under Backfire conditions, it can wrongly classify accurate and well-intentioned scientific communication as disinformation. To address these limitations, we propose a hybrid teleological framework, Functional-Contextual Disinformation (FC-DIGC), which combines DIGC’s consequence criterion with a teleological constraint requiring a misleading function. This synthesis better separates malicious deception (disinformation) from unintended epistemic harm (contextually harmful misinformation) and helps clarify how LLM outputs should be categorised.
Conclusion: A hybrid teleological approach improves extensional adequacy by preventing over-generation while retaining coverage for non-intentional systems. FC-DIGC provides a principled way to distinguish disinformation from contextually harmful misinformation and, by integrating the lying–bullshit contrast, captures a broader spectrum of epistemically motivated malice relevant to contemporary information environments, including AI-mediated communication.