Back to Overview

The Manifesto

Good Sense for / by / to AI

Hume, Epstein, and Floridi as the Intellectual Foundation of GoodSense.ai — Research Synthesis.

TL;DR

  • Common sense ≠ sound judgement. "Good sense" (sound, context-sensitive practical and moral judgement) is conceptually distinct from "common sense" in both its senses — consensus/widely-shared belief and tacit world-knowledge. A system can hold encyclopedic commonsense knowledge and faithfully reproduce consensus belief while still lacking sound judgement.
  • Three pillars, one division of labour. Hume's corrected moral sentiment supplies the normative content of good sense; Epstein's anchoring/grounding supplies the social-ontological mechanism by which AI reshapes social reality; Floridi's infosphere / Fourth Revolution supplies the ontological stage and the recursive society-technology feedback loop.
  • The tripartite framework (Good Sense FOR / BY / TO AI) is defensible: FOR = prospective design desideratum, BY = the AI as operational exerciser of good sense, TO = retrospective evaluative standard on consequences.

SECTION 0 — The "Common Sense vs Good Sense" Distinction

The claim that common sense does not necessarily equate to sound judgement has deep philosophical pedigree. Descartes (1637) deployed irony noting that precisely because everyone believes themselves amply supplied with good sense, self-assessment is the least reliable instrument we have. Good sense (the power of judging well) must be properly exercised through method.

Gramsci (1929-1935) built an explicit dipole between senso comune (vague, commonly-held beliefs) and buon senso (the good, relatively coherent, critical kernel). Kant (1790) located sensus communis as a "community sense"—the faculty of reflective judgement, which judges by reference to a possible universal standpoint.

In contemporary AI, "common sense" means tacit world-knowledge (the frame problem). Our "good sense" is unambiguously the capacity for sound, corrected, context-sensitive judgement — explicitly NOT mere statistical regularity nor a knowledge stock.

PILLAR 1 — DAVID HUME'S ETHICS (Moral Sentimentalism)

Hume's corrected moral sentiment supplies the normative content of good sense. Reason is the slave of the passions; moral distinctions are derived from sentiment, not reason. However, raw sympathy is biased. To share and communicate moral judgements, we correct sentiment by adopting a "steady and general point of view."

  • Normative content of good sense = corrected sentiment, not consensus. The general point of view shows sound judgement to be corrected and generalized sentiment, precisely NOT majority opinion or statistical regularity (consensus/senso comune).
  • The is-ought gap as a structural limit on value learning. Descriptive behavioural data (what humans do prefer/reward) cannot entail what an AI ought to do, challenging standard RLHF paradigms.
  • Judgement vs. rules. Good sense is a character-like capacity that rigid rules cannot exhaust.

PILLAR 2 — BRIAN EPSTEIN'S SOCIAL ONTOLOGY

Epstein's framework provides the social-ontological mechanism where good sense must intervene. He distinguishes grounding (non-causal metaphysical determination between facts) from anchoring (how frame principles themselves are put in place).

  • AI as a new ground AND a new anchor. AI becomes embedded in social reality not just by grounding facts (e.g., "this applicant is high-risk"), but by anchoring—algorithmic systems put in place new frame principles for social kinds like "creditworthiness" or "verified."
  • AI changes frame principles, not just facts. The deepest societal impact of AI is that it re-anchors the frame principles of social kinds. Good sense must intervene at the anchoring level, not merely audit individual outputs.
  • Anti-individualism. Social facts are not exhaustively anchored in individual attitudes. A "good sense" lodged only in individual users' or designers' intentions is insufficient.

PILLAR 3 — LUCIANO FLORIDI'S PHILOSOPHY OF INFORMATION

Floridi's Fourth Revolution supplies the ontological stage and the recursive dynamics. ICTs are re-ontologizing technologies; they don't merely add tools to the world, they change what the world is (the infosphere).

  • The ontological stage. AI is a re-ontologizing force operating within the infosphere, among inforgs living onlife.
  • The recursive society-technology loop. AI shapes society, which shapes the data and values that shape AI. Whether this circle is virtuous or vicious depends on our moral standpoint and consequences (where "good sense TO AI" bites).
  • Agency without intelligence. AI can be a mind-less moral agent at an appropriate Level of Abstraction (LoA). Asking it to exercise "good sense" is not asking it to be conscious, but asking that its agency be structured by design to track corrected sentiment.

SECTION 4 — THE SYNTHESIS: GOOD SENSE FOR / BY / TO AI

1. Good Sense FOR AI — design desideratum.

Good sense by design, applicable to any system considered as AI. The thesis: building good sense in means engineering the Humean corrective process—generalizing and de-biasing sentiment toward a common point of view—rather than hard-coding rules.

2. Good Sense BY AI — the AI as exerciser of good sense.

The system running on a good-sense lens. An artificial agent can exercise good sense at an appropriate LoA without being intelligent. Hume supplies the content of the judgement; Epstein supplies the social-ontological objects the judgement is about.

3. Good Sense TO AI — external evaluative standard on consequences.

Good sense applied consequentially to systems that were NOT framed by good-sense-by-design. A retrospective, patient-oriented evaluation. Hume's general point of view is the standpoint from which we correct our sentiments and judge the system's downstream effects.

The three are unified by Floridi's recursive circle: FOR (design) feeds BY (operation), which produces consequences judged by TO (evaluation), whose verdicts feed back into FOR. Good sense is what makes the circle virtuous rather than vicious at each pass.