Send message to Authors

Paper short abstract

The proposal highlights the value of embedded, situated knowledge held by actors, examining how it is represented in synthetic data and how it can shape the intelligence of AI systems.

Paper long abstract

Does synthetic data capture the embodied knowledge of the actors it represents?

With this question, we intend to open a discussion around how the inclusion or exclusion of embodied, situated knowledge in synthetic data not only alters the intelligence of AI systems but also redistributes power over how their results are interpreted, trusted, and acted upon.

Suchman explains knowledge is "situated, embodied, and accountable to the specific positions from which it is enacted" Suchman 2002). Such knowledge holds diverse forms of human understanding that often remain undocumented and unarticulated, yet are tacitly used, shared, and developed over time through experience, conversation, and collaboration. It is crucial for forming agreements, building diverse opinions, and shaping dissent. In this sense, it is a defining characteristic of the individual who holds it.

Hybrid factory floors are rich sites of embedded knowledge, where workers continuously draw on tacit, embodied expertise to complete tasks. Now, this workspace has been disrupted by robots enabled by physical AI, often trained using synthetic data. It creates an important tension between simulated knowledge and the lived realities of workers' practices.

What remains unclear is whether the AI-generated synthetic data also incorporates workers' embodied knowledge. If not, then can synthetic representation be considered complete or accurate?  How is such an outcome getting consumed as intelligent? Moreover, if synthetic data succeeds in capturing embodied, lived, situated knowledge, does it become more than "copies" of missing actors? In light of these provocations, we warrant a careful epistemic examination of synthetically generated knowledge.