L’AI te lo legge in faccia: fisiognomica macchinica tra decodifica aberrante e retorica dei dati

The widespread use of machine learning tools for automated facial analysis has recently shaped, in several widely publicized articles, a new form of physiognomy, the pseudoscientific discipline which attempts to infer inner traits of human beings from their physical appearance. The rhetorics of these articles pairs pseudoscientific narratives with a key epistemological assumption: namely, that the machine–based construction of scientific facts can legitimately go beyond measurement and data processing, instead involving the very definition of the model which interprets the data. What is proposed here is that the semiotic category of aberrant decoding, as formulated by Eco – with some theoretical adjustments to make it suitable for interpreting the alien semiosis process in AI models – lends itself to deconstructing this mechanism, providing a useful critical element which serves as a technical counterpart to the analysis of the narratives at play.