← All sources
Making AI Intelligible: Philosophical Foundations
Herman Cappelen; Josh Dever · 2021 · Oxford University Press (open access monograph)
background
medium priority
coded
Main argument
Thesis: philosophical theories of meaning/content can and should be applied to AI systems - the question 'can AI systems say things / have contentful states?' requires DE-ANTHROPOCENTRIZED metasemantics (adapting externalist theories like Kripke/Burge/teleofunctionalism beyond human thinkers); the book develops this for cases (a credit-scoring system 'saying' Lucie is a poor risk; a stop-sign detector detecting stop signs) and argues in closing that (a) attempts to bypass content via pure reliability talk (the No-Evidence view) fail because ML correlations lack the causal structure dispositions require - 'when you have a satisfactory theory of [dispositions/reliability/evidence]... you have in effect come very close to constructing a theory of content again'; and (b) explainable AI is a metasemantic project: Lucie 'is entitled to understand why' - an oracle's reliable-but-incomprehensible verdict 'isn't good enough'.
Why it matters here
The philosophy-of-language foundation: whether AI outputs can carry CONTENT at all (externalist metasemantics for de-anthropocentrized systems), with the closing argument that reliability-without-content ('No-Evidence' views) collapses back into needing a theory of content - and that explainable-AI's why-questions (Lucie's mortgage) are METASEMANTIC questions. Upstream of the moral-agency debate: before asking whether AI can be a moral agent, one must ask whether its outputs mean anything.
Reading notes
Targeted treatment (184pp monograph): intro + final chapter read; middle chapters (externalism applied to specific cases: SmartCredit, stop-sign detectors) mapped via TOC. Pre-ChatGPT (2021) - a clean 'before' snapshot for the philosophy-of-language side.
Cappelen, H., & Dever, J. (2021). Making AI Intelligible: Philosophical Foundations. Oxford University Press.
Close reading — 3 coded units
#1
· pp. 3
· claim
“Our focus is on one relatively underexplored question: Can philosophical theories of meaning, language, and content help us understand, explain, and maybe also improve AI systems? Our answer is 'Yes'.”
#2
· pp. 161–162
· argument
“in the machine learning case, we don't have any reason to think that there is a causal connection to support the disposition. [...] we have to fall back on the coincidental convergence of weird structural properties and our target properties on the training cases—but the coincidental convergence doesn't give us reason to treat the system as reliable for new cases. [...] When you have a satisfactory theory of that kind—one that responds to all these concerns—you have in effect come very close to constructing a theory of content again.”
#3
· pp. 162
· argument
“If an AI system tells us that Lucie should not get a mortgage, she is entitled to understand why she should not get a mortgage. To answer the why-question by simply insisting that the decision was made by a reliable but incomprehensible algorithm isn't good enough.”
Synthesis-matrix row
complicates
T5-AGENCY-DENIED-EVALUABILITY-KEPT
content question upstream: whether outputs mean anything is itself unsettled
Memos (1)
theoretical · unit #2
Cappelen & Dever supply the deepest layer of the dissertation's anti-agency stack: below 'no phronesis' (Noller), 'no deliberative capacity' (Millière), and 'no stable values' (five empirical studies) sits 'no settled account of CONTENT' - if it's open whether the system's outputs mean anything, ascribing moral judgments (let alone agency) is doubly premature. Their closing argument (unit 2: reliability theories collapse into content theories) also warns the dissertation's own experiment: interpreting LLM 'moral reasoning' outputs presupposes a metasemantic stance - state it (a modest externalism: outputs inherit content from human linguistic practice, per the Wittgenstein thread) rather than assume it. And unit 3's Lucie case is the philosophy-of-language ancestor of the intelligible-reasons requirement running through S&K, J&N, and Tasioulas - cite as the origin point.