What exactly is 'hidden work' when humans collaborate with AI?
Hidden work is the unacknowledged human effort required to make AI systems function as intended. A 2023 study of waste workers in the U.S. during the pandemic coined the term 'patchwork' to describe this: workers performed continuous acts of calibration, troubleshooting, and repair to smooth the relationship between AI robotics and their environment [1]. This labor was invisible to managers and system designers, yet essential for the AI to work at all.
The same study found that this hidden work is not a one-time fix but an ongoing burden. Workers had to constantly adjust and fix AI systems because the technology could not handle real-world variability [1]. This means that without explicit design for transparency, human-AI collaboration inevitably creates extra, uncredited work for humans.
Can AI be designed to avoid creating hidden work?
Yes, but it requires proactive design that makes the AI's limitations and actions visible. A 2025 controlled experiment with 36 participants in temporary design teams showed that AI voice agents with proactive intervention strategies—designed to manage conflict escalation and de-escalation—improved interpersonal relationships, communication quality, and collaboration experience without adding hidden work [2]. The key was that the AI took on the burden of managing team dynamics, rather than requiring humans to compensate for its shortcomings.
A 2025 comprehensive review of human-AI collaboration proposes a framework with five levels of human integration, emphasizing that systems must be flexible and allow user control to avoid shifting hidden work to humans [5]. The review argues that when AI systems are designed as collaborators rather than tools, with transparent decision-making and user involvement, the risk of hidden labor decreases significantly.
However, a 2023 lab experiment with 156 participants found that negative attitudes toward AI can lead to reluctance to accept AI advice, resulting in lower performance [4]. This suggests that even well-designed AI may create hidden work if users distrust it and spend extra effort second-guessing or overriding it.
What are the ethical risks if hidden work is ignored?
Ignoring hidden work raises serious ethical concerns about responsibility and autonomy. A 2022 article argues that in human-AI collaboration, co-supervision is necessary for shared responsibility; without it, humans are left solely responsible for the AI's actions, which is unfair and creates hidden work [3]. The article proposes that AI systems should be programmed to supervise human actions as humans supervise AI, creating a balanced partnership.
A 2023 study of conversational agents (CAs) in virtual teams identified that these AI tools can collect user data, reduce worker autonomy, and foster social isolation—all forms of hidden work where humans must compensate for the AI's negative impacts [7]. The study derived 14 ethical guidelines for introducing CAs, including ensuring transparency about data use and maintaining human control, to prevent hidden labor from being offloaded onto team members.
A 2025 research platform designed to study human-AI collaboration found that classic computer-supported cooperative work (CSCW) principles—like information pooling and shared awareness—still apply when collaborating with AI agents [6]. This means that hidden work can be identified and measured using existing frameworks, giving researchers and designers tools to prevent it.
About These Sources
This answer is built on 7 peer-reviewed studies — published from 2022 to 2025, 3 from 2024 or later, 4 in Q1 journals, collectively cited 146 times — selected as the most relevant from 7 studies that passed quality screening, drawn from 55 papers retrieved from a database of over 500 million.
Sources used in this answer
1
Patchwork: The Hidden, Human Labor of AI Integration within Essential Work
Through participant observation and interviews in two U.S. waste labor sites, this study found that frontline workers performed continuous 'patchwork' labor—calibration, troubleshooting, and repair—to compensate for AI shortcomings, creating hidden, undervalued work [1].
2
Maintaining "Balanced" Conflict: Proactive Intervention Strategies of AI Voice Agents in Online Collaboration of Temporary Design Teams
In a controlled experiment with 36 participants in temporary design teams, AI voice agents with proactive intervention strategies improved interpersonal relationships, communication quality, and collaboration experience without adding hidden work, by managing conflict constructively [2].
3
AI and Ethics When Human Beings Collaborate With AI Agents
This conceptual article argues that co-supervision between humans and AI is necessary for shared responsibility; without it, humans bear sole responsibility for AI actions, creating hidden work [3].
4
Working with AI: How Attitudes Shape Human-AI Collaboration
A lab experiment with 156 participants found that negative attitudes toward AI led to greater reluctance to accept AI advice, resulting in lower performance, suggesting that distrust can create hidden work as humans override AI [4].
5
A Multifaceted Vision of the Human-AI Collaboration: A Comprehensive Review
This comprehensive review proposes a framework with five levels of human integration, emphasizing flexibility and user control to prevent hidden work and improve human-AI collaboration [5].
6
Through the Lens of Human-Human Collaboration: An Configurable Research Platform for Exploring Human-Agent Collaboration
This study introduces an open research platform for human-AI collaboration, demonstrating through three case studies that classic CSCW principles (e.g., information pooling) still apply, enabling measurement of hidden work [6].
7
Ethical Challenges for Human–Agent Interaction in Virtual Collaboration at Work
Through 15 expert interviews, this study identified ethical challenges of conversational agents in virtual teams, including data collection, reduced autonomy, and social isolation, and derived 14 guidelines to prevent hidden work [7].
Related research questions
