Frames that guide professional judgment

They analyzed 57 interviews with professionals and millions of articles and posts, identifying three main debates (method, mind, morality) that structure how responsibility is assigned and how fast to move. This shows that decision makers must recognize those frames to anticipate conflicts and align expectations.

Faced with the rise of language models, product and policy teams constantly have to decide whether to speed up or slow down development, and those decisions are rarely based on technical data alone. What really drives the conversation are the ideas each group has about how AI is built, what kind of mind they attribute to it and what they consider right or wrong in its use. When those views clash, projects stall or conflicts over responsibility arise that make it hard to move forward.

To understand these dynamics, the authors of Proceedings of the ACM on Human-Computer Interaction carried out a mixed study that combined computational analysis of millions of press releases and social media posts with fifty-seven semi-structured interviews with AI professionals conducted in 2021 and 2023. Using topic models based on dictionary learning over vector representations, they identified four recurring cognitive challenges: keeping up with the pace of information, translating concepts between different groups, assigning responsibility for social impacts and balancing competing priorities. Three main framing debates emerged from those challenges: the method of development (between top-down expert systems and bottom-up emergent capabilities), the mind attributed to AI (from passive tool to human-like entity) and the morality of its use (the choice between slowing down or speeding up its progress).

Leaders can use that map of frames to spot early where tensions appear and act before they turn into blockages. Instead of imposing a single narrative, they can design communication that respects the different readings and offers common ground, for example by pointing out how the same advance is perceived both as a productivity gain and as a risk to job autonomy. Making assumptions about responsibility and pace explicit leads to better alignment of expectations among engineers, regulators and users, and avoids costly readjustments later.

Conclusions

• Professionals adopt top-down or bottom-up development schemes to define who controls the results.
• They see systems as passive tools or as digital minds, which shapes how blame is assigned.
• They debate whether to speed up or slow down progress, and that stance determines the policies they support.
• Recognizing these frames makes it possible to anticipate disputes and design more robust decision processes.