# Philosophy's implicit value signals Philosophy papers often rely on unstated evaluative standards — and this actually helps explain why LLMs can be good at philosophy. ## The Observation Many philosophy papers don't explicitly state their "value function" (what counts as success). Authors proceed as if the relevant virtues are shared and obvious: - What counts as a decisive objection - What's "too ad hoc" - Which costs are acceptable - What needs formal treatment vs intuitive case - Which "theoretical virtues" matter in this sub-literature ## Why This Isn't a Problem "Implicit" doesn't mean "absent." The value signal is *distributed across practice*: > The "value function" is often implicit in the same way the rules of good conversation are implicit: you can violate them without anyone quoting the rulebook, and you can follow them without ever naming them. ## Connection to LLM Philosophy If philosophical evaluation were always explicit and formal, an LLM without tools would struggle unless it could do rigorous checking. But because so much evaluative practice is *tacit and stylistically encoded*, LLMs can pick up a social prior for what "good philosophy" looks like. That can be enough to generate genuinely useful work — especially when a human supplies: - The target question - The taste about what matters - The "no, not that — this" steering ## Candidate Value Signals For [[Move 37 and philosophical novelty|Move 37]]-type philosophical novelty, plausible "value signals" include (from weakest to strongest): 1. **Internal clarity signals** — Does it yield sharper theses, remove equivocation, specify scope? 2. **Dialectical resilience signals** — Does it survive the best objection? Avoid obvious counterexamples? 3. **Unification signals** — Does it explain why two debates are actually one debate in disguise? 4. **Fruitfulness signals** — Does it generate new questions, new tests, new distinctions downstream? 5. **Expert preference signals** — Would competent philosophers judge it a serious contribution? Signals (1)–(4) can be approximated inside conversation. Signal (5) is fundamentally social. ## The AlphaGo Contrast In Go, value = probability of winning (crisp, external). In philosophy, value is a "fuzzier, socially mediated bundle": clarity, non-question-begging-ness, dialectical resilience, unificatory power, fruitfulness, fit with the target debate. > Those standards are real — just not numerically crisp. The best analogue to AlphaGo's value network in philosophical dialogue is: a dialectical evaluation loop (including you, including multiple model "voices", including other humans). Source: [[Web Clippings/chat with ChatGPT about Floridi, LLM reasoning, Move 37|ChatGPT conversation on Floridi et al.]]