Abstract

In response to Luciano Floridi’s conjecture on a fundamental trade‑off between certainty and scope in symbolic and generative AI (Floridi, 2025), we firstly propose a shift in perspective by relocating the problem of quantifying the conjecture from the viewpoint proposed in the original paper, which considers AI machinery as static usage-independent artefacts, to a pragmatism-driven one which recovers the central role of the users in the evaluation of a system and transforms the quantities involved in the conjecture into achievable field measurements, taking into account the intrinsic probabilistic nature of modern generative AI. Secondly, based on this shift of perspective, we identify formal issues in the original definitions—specifically, the conflation of probabilistic and order‑theoretic notions in model correctness (C) and the opaque use of cumulative Kolmogorov complexity for mapping scope (S) and propose alternatives that fit the new conceptualization and resolve the identified issues. In summary, although retaining most of the original philosophical foundations and motivations, we (i) reframe the general problem introducing aspects related to usage in the conceptualization of the problem; (ii) introduce an Epistemic Certainty Level that replaces a mixed supremum–probability measure with a coherent expectation over indicator functions on output space’s user-driven equivalence classes; (iii) substitute cumulative Kolmogorov complexity with Shannon joint entropy. In our judgement, these revisions better quantify input/output mapping scope and make the conjecture more tractable in practice, by turning it from an aprioristic system evaluation into a field measurement.