Whose readiness counts? Disagreement within and between sectors in perceived AI and robotics preparedness
Citation. Wang, P. (2026). Whose readiness counts? Disagreement within and between sectors in perceived AI and robotics preparedness (arXiv:2608.23406). https://doi.org/10.48550/arXiv.2608.23406
Summary. This preprint examines how much information is hidden when perceived AI and robotics readiness is reported as a single sector-level score. In a survey of 982 people rating 17 concrete challenges, differences between broad challenge families explained little of the variation; respondents and applications within the same sector often differed more. The paper argues that readiness reporting should show which applications were assessed, who supplied the ratings, and how much the ratings disagree.
Key claims
- Across 15,200 ratings, the 17 challenges had substantial respondent disagreement: standard deviations ranged from 1.03 to 1.26 on a five-point readiness scale.
- In the study’s decomposition, stable differences between respondents accounted for 32.7% of observed variation, challenge differences for 7.3%, and remaining response-level variation for 60.0%; broad challenge-family averages accounted for about 2% in a separate descriptive breakdown.
- Manufacturing received the highest average readiness rating, but its robotics, process-optimization, and general decision-support applications were judged differently; a sector average should not be treated as a uniform preparedness judgment.
- Computer-science and AI/ML respondents reported higher readiness than non-technical respondents across challenge families, so the composition of a respondent panel can change the apparent readiness level.
- Ethical awareness and cyber security should both inform readiness decisions, but their distinct concerns should not be collapsed into one preparedness measure.
Evidence & limitations
- Card-based Prolific survey conducted in June 2026; 982 usable respondents rated 17 named challenges, and 864 completed all cards. The paper measures perceived community preparedness and available resources, not personal willingness or independently audited organizational capability.
- The preprint is not peer-reviewed or preregistered. Its 17 purposively selected challenges are not representative of all AI and robotics applications, and several readiness-related measures are single-item ratings collected in the same exercise.
- The authors describe the variance breakdowns as descriptive. With one rating per person-challenge pair, the remaining variation cannot separate person-by-challenge differences from measurement error.
- Public provenance was verified through the arXiv record and an exact SHA-256 match between the ingested Drive PDF and the public arXiv PDF.