Paper Series

Can AI smell (yet)? Anticipatory judgment, gap closure, and the Lovelace objection

7 October 2026

Vidhu Shekhar

Abstract

Since Lovelace, people have asked whether a machine can originate anything beyond what it is given. AI now produces historically novel solutions in coding, mathematics, and science, but much of this can be understood as gap closure: solving a problem whose objective, criterion, and domain someone else supplied. Gap closure leaves a different capacity untested. Experienced programmers can sense that an architecture will become brittle before they can name the requirement that will expose it. I call this capacity anticipatory smell and define it behaviorally: an agent exhibits it to the extent that, unprompted, its design choices later prove cheaper to adapt to changes absent from the task context, net of the cost of the flexibility they add. A framework of four modes, specification execution, gap closure, anticipatory smell, and self-originated inquiry, separates capacities that require different evidence and relocates the Lovelace objection from novelty of output to the source of direction. An informational result shows that systematically lower adaptation cost under unrepresented change requires information available at the time of choice that predicts differences in that cost; enumerating more futures does not supply it. I hypothesise that one source of such judgment is a learning signal linking design decisions to their later consequences, and propose a test comparing models trained on the same project histories with decision-consequence relations paired, left in their ordinary dispersed form, or shuffled.

Keywords: Artificial intelligence (AI), expertise, tacit knowledge, unawareness, creativity, Lovelace objection, software engineering,
AI evaluation

Note: This paper is being circulated as a working paper. A version of the paper is currently under consideration for journal publication.

Suggested Citation:

Shekhar, V. (2026). Can AI smell (yet)? Anticipatory judgment, gap closure, and the Lovelace objection. Working Paper, SPJIMR Paper Series, WP-2026-15. S.P. Jain Institute of Management & Research (SPJIMR). https://www.spjimr.org/paperseries/wp-2026-15/

Download PDF Preview PDF

Disclaimer: The SPJIMR Paper Series provides a platform for hosting and disseminating research outputs. The views, opinions, findings, interpretations, and conclusions expressed in these papers are solely those of the respective author(s). They should not be construed as representing the views, endorsement, or official position of SPJIMR.