Are We Automating the Joy Out of Work? Designing AI to Augment Work, Not Meaning

Abstract

Prior work has mapped which workplace tasks are exposed to AI, but less is known about whether workers perceive these tasks as meaningful or as busywork. We examined: (1) which dimensions of meaningful work do workers associate with tasks exposed to AI; and (2) how do the traits of existing AI systems compare to the traits workers want. We surveyed workers and developers on a representative sample of 171 tasks and use language models (LMs) to scale ratings to 10,131 computer-assisted tasks across all U.S. occupations. Worryingly, we find that tasks that workers associate with a sense of agency or happiness may be disproportionately exposed to AI. We also document design gaps: developers report emphasizing politeness, strictness, and imagination in system design; by contrast, workers prefer systems that are straightforward, tolerant, and practical. To address these gaps, we call for AI whose design explicitly focuses on meaningful work and worker needs, proposing a five-part research agenda.

Figures (9)

• Figure 2: Proportions of tasks across occupational sectors in three datasets: Prolific (n=171), LM-annotated (n=10K), and U.S. labor statistics (n=18K). The Prolific sample covers 12 sectors (based on available occupations), the LM-annotated dataset covers 19 sectors, and the U.S. Bureau of Labor Statistics dataset covers all sectors. The distributions of Prolific and LM-annotated tasks are broadly similar across sectors, indicating that LM annotations capture sector patterns consistent with U.S. labor statistics.
• Figure 3: AI Exposure Gap by dimensions of meaningful work (rows). The higher the gap, the more strongly that dimension is associated with tasks likely to be exposed AI. This gap is computed as the difference of how important a dimension is between two groups of tasks: those that are more likely to be exposed to AI, and those less likely. We estimate the gaps and 95% confidence intervals with mixed-effects models. Bold names and corresponding black bars indicate differences that are statistically significant. Tasks rated likely to be exposed tend to involve novelty, creativity, happiness, and freedom in how workers do them. Tasks rated not likely tend to involve emotional awareness, in-person interaction, building relationships, and supporting social connection.
• Figure 4: Association of tasks exposed to AI in each of the sectors (in the rows) with subset of three dimensions of meaningful work (creativity, positive affect, and autonomy in the columns). Sectors are sorted by the average $z$-score across the three dimensions. Creative and socially-oriented sectors (arts, community service, education, life sciences) are associated with tasks exposed to AI that emphasize novelty, positivity, and freedom. In contrast, routine and manual sectors (office support, production, farming) score much lower.
• Figure 5: Association of tasks not exposed to AI in each of the sectors (in the rows) with subset of four dimensions of meaningful work (emotional awareness, in-person interaction, relationship building, and social connections in the columns). Sectors are sorted by the average $z$-score across the subset of four dimensions. Human-facing sectors such as community and social service, education, and healthcare consider their tasks not exposed to AI to emphasize emotional awareness, in-person interaction, and social connection. In contrast, technical and routine sectors (e.g., production, office support, computer and mathematical) score far lower, indicating that workers in these sectors view tasks not exposed to AI as less socially or emotionally significant.
• Figure 6: The three most misaligned AI traits, (top row) and the three most aligned traits (bottom row). Larger values indicate greater disagreement between workers and developers. Scores are the absolute differences between workers' ratings of how much they want an AI system to exhibit each trait and developers' ratings of how much they intend to incorporate that trait into the design of an AI system. Icons show which trait direction each group prefers (e.g., workers wish straightforward systems, while developers set out to design polite systems). Top contributing sectors and example tasks are listed for each trait.