The inevitable human residue and the measurement of artificial creativity

Generative AI systems produce artefacts that can scarcely be distinguished from human works, eroding the boundary between anthropogenic and machine creativity – with far-reaching consequences for science, law, art, economics, and the political public sphere. This article advances a twofold thesis. First, every AI product, irrespective of its technical autonomy, retains a structurally irreducible residue of human influence – the Last Human Percent (LHP). This residue is not a correctable measurement error but a constitutive condition of machine creativity, anchored in the cultural, semantic, and intentional framing supplied by training data, architecture, and use. Second, it functions as a measurement limit from which the degree of humanity in a work can be quantified. Philosophically, the LHP is situated within a line of thought from Wittgenstein through Kant to Derrida. Methodologically, the article proposes the Human–AI Dissimilarity Index (HAIDI), which determines the AI share relationally against human and machine reference corpora. By analogy with the metre or the second, such a measure is understood not as a discovered but as a constructed truth, whose validity rests on intersubjective convention and whose establishment is an infrastructural imperative for the age of generative AI.

Keywords: Generative Artificial Intelligence, Human–AI Dissimilarity Index (HAIDI), Last Human Percent (LHP), Authorship Attribution, Human–Machine Creativity, AI Detection, Epistemic Trust.