The Role of Inclusion, Control, and Ownership in Workplace AI-Mediated Communication

TL;DR

This study investigates how writer-style bias in AI autocomplete affects workers' feelings of inclusion, control, and ownership in AI-mediated workplace writing. Using a three-arm online experiment with self-assured, hesitant, or no AI assistance, participants drafted a promotion-request email; AI reliance, inclusion, control, and ownership were measured, and exploratory analyses examined how inclusion relates to perceived agency, particularly for minoritized genders. The main findings show that AI style did not impact inclusion but did influence control and ownership, with hesitant style enhancing both; inclusion can buffer losses in agency due to AI reliance, and the link between inclusion and agency is stronger for minoritized genders. The authors propose a conceptual model separating AI-user alignment from AI-task alignment to explain how such factors shape perceived agency in AI co-writing, offering design guidance for coworking with LLMs and highlighting the importance of supporting diverse user identities in AI interfaces.

Abstract

Given large language models' (LLMs) increasing integration into workplace software, it is important to examine how biases in the models may impact workers. For example, stylistic biases in the language suggested by LLMs may cause feelings of alienation and result in increased labor for individuals or groups whose style does not match. We examine how such writer-style bias impacts inclusion, control, and ownership over the work when co-writing with LLMs. In an online experiment, participants wrote hypothetical job promotion requests using either hesitant or self-assured autocomplete suggestions from an LLM and reported their subsequent perceptions. We found that the style of the AI model did not impact perceived inclusion. However, individuals with higher perceived inclusion did perceive greater agency and ownership, an effect more strongly impacting participants of minoritized genders. Feelings of inclusion mitigated a loss of control and agency when accepting more AI suggestions.

Figures (6)

• Figure 1: Screenshot of the writing task. Instructions are given in the panel at the top. Participants can hit 'tab' to accept the suggestion or 'esc' to generate a new suggestion. Written text is in black and suggestions appear in gray text.
• Figure 2: Participants’ assessment of inclusion.Participants are more likely to say the assistant was made for them than sounds like them. Responses less than 5% are not labeled on the figure.Responses less than 5% are not labeled on the figure.
• Figure 3: Participants assessment of control. Participants assisted by the hesitant writing style model are more likely to feel greater control over the final version of the message and the writing process than participants assisted by the self-assured writing style model. Responses less than 5% are not labeled on the figure.
• Figure 4: Participants assessment of ownership. Participants assisted by a hesitant model are more likely to say that they wrote the message and the message sounds like them. Responses less than 5% are not labeled on the figure.
• Figure 5: Interaction Effects in the Linear Regression. Figure \ref{['fig:interaction']}a depicts the interaction between Inclusion and AI Reliance. Figure \ref{['fig:interaction']}b depicts the interaction between Gender and Inclusion.