Human-AI collaboration breaks down without strong team communication

Across every measure, self-managed teams scored higher. The difference was especially pronounced during machine breakdowns, where rapid interpretation of system indicators was essential. The study’s structural equation modeling demonstrated that communication fully mediates the relationship between team organization and performance. In other words, the hierarchical or self-managed label alone did not directly affect productivity or quality. What mattered was how the structure shaped communication.

A recent scientific investigation reveals that self-managed teams outperform traditional hierarchical teams when handling automation failures, a result that carries major implications for Industry 4.0 workplaces where AI guidance is now part of daily operations.

Published in AI & Society, the study titled "How team organization influences the ability to solve automation failures: an experimental study on human–AI decision-making in teams" shows that communication is the decisive factor that determines whether teams succeed or falter when automated systems break down.

Backed by experimental evidence from a realistic smart factory simulation, the research offers a close look at how team structure shapes communication patterns, problem-solving capability and overall performance. 

Self-managed teams gain an edge in high-automation settings

The researchers designed an immersive laboratory experiment that reproduced the workflow of a modern, AI-supported production line. The researchers sought to understand how teams cope with real-time disruptions when they must diagnose machine failures, verify information and decide whether to trust AI-generated recommendations.

The experiment involved pairs of participants operating a simulated lens production system consisting of two connected machine stations. The environment was engineered to mimic the conditions of an Industry 4.0 factory, including automated processing steps, real machine interactions, material flow and an AI-based assistant that proposed solutions to equipment problems. Each team was randomly assigned either a hierarchical structure, with one member appointed as leader, or a self-managed structure, where both members shared responsibility for decisions and troubleshooting.

Across 87 experiment runs, the researchers measured productivity, error rates and the communication behavior of each team. The results reveal a clear pattern: self-managed teams consistently communicated more often, shared information more constructively and demonstrated superior performance when handling machine breakdowns. Hierarchical teams, by contrast, were far more likely to fall into low-communication patterns that reduced their ability to diagnose problems effectively.

Managing automation failures requires intensive information exchange between team members, especially because AI recommendations were sometimes incorrect. Self-managed teams showed a stronger tendency to verify information collaboratively, cross-check machine indicators and reach shared judgments about the best course of action. This behavior led to faster production completion times and fewer errors in responding to failure scenarios.

The researchers note that as automation rises, work shifts from routine processing to exception handling. Teams increasingly face situations where they must intervene when automated systems malfunction, and the success of these interventions depends heavily on communication quality. In high-speed production settings, a delay or error in diagnosing the source of a breakdown can cascade into larger operational disruptions, making team structure a strategic concern in modern manufacturing.

Communication emerges as the critical mediator of performance

Observers rated how often team members exchanged information, how they interacted during disruptions, whether their exchanges were constructive and how they responded to AI-generated suggestions. Participants also rated their own communication experience.

Across every measure, self-managed teams scored higher. The difference was especially pronounced during machine breakdowns, where rapid interpretation of system indicators was essential. The study’s structural equation modeling demonstrated that communication fully mediates the relationship between team organization and performance. In other words, the hierarchical or self-managed label alone did not directly affect productivity or quality. What mattered was how the structure shaped communication.

This statistical result confirmed the researchers’ hypothesis that team communication is the bridge between organizational design and operational outcomes. Self-managed teams naturally encouraged more horizontal information flow, while hierarchical teams limited exchanges by placing responsibility primarily on the designated leader. This reduced the frequency of input from the second team member and increased the risk of oversight when diagnosing causes of machine errors.

The cluster analysis included in the study further illustrated this divide. Most self-managed teams fell into a high-communication group, while hierarchical teams were split nearly evenly between high- and low-communication groups. This variability underscores how hierarchies depend strongly on the leadership style and behavior of the appointed supervisor. If leaders fail to foster open communication, teams rapidly slide into low-information patterns that degrade performance.

The researchers’ findings also align with established literature in human–autonomy teaming, which has shown that automation reduces situational awareness and increases the cognitive burden on humans during failures. Teams must be ready to compensate for this reduced awareness by sharing information quickly and accurately. In this context, self-managed teams appear better suited to the demands of Industry 4.0 environments, where troubleshooting requires collective interpretation of machine data, AI suggestions and production indicators.

The study’s experimental design also reinforces the growing recognition that AI assistance introduces its own complexity. Participants were given an AI tool that produced recommendations with associated probability values indicating the likelihood of accuracy. In two of six failure scenarios, the AI recommendations were incorrect, testing the team’s ability to challenge or override the technology’s guidance. Self-managed teams were more likely to question these suggestions and re-evaluate machine indicators, reducing the risk of compounding errors. Hierarchical teams, however, showed a greater tendency to rely on the leader’s interpretation, which limited the opportunity for collaborative cross-checking.

This dynamic reflects a key insight of the study: teams working with AI systems must also learn to manage the fallibility of AI. When communication is constrained, whether by hierarchy or other organizational patterns, teams may be more prone to automation bias and overreliance on AI suggestions. Well-functioning communication, by contrast, supports a balanced approach where AI input is considered but not accepted uncritically.

A call for rethinking team structures in the AI era

Manufacturing firms face increasing pressure to integrate AI systems into their production processes, but technological upgrades alone do not guarantee performance gains. As automation grows more complex, human teams must resolve problems that machines cannot, and the structure of these teams becomes a decisive factor in operational reliability.

The researchers argue that current industrial practices are still dominated by lean production concepts, which rely on hierarchical team design and centralized decision-making. While such structures may have worked in earlier eras of standardization and predictable workflows, they may be mismatched to modern AI-supported environments where decision-making must be distributed, adaptive and informed by rapid sharing of situational cues.

The study indicates that self-managed teams offer a strong alternative model. They show greater resilience during disruptions, exhibit higher communication quality and maintain consistent performance even under pressure. In highly automated settings, these characteristics become essential for sustaining production flow.

However, the study also acknowledges that self-management is not a simple organizational fix. Successful implementation may require structural adjustments, such as assigning micro-roles within teams, developing new communication protocols or designing AI systems that actively support collaborative decision-making rather than individual directives. For hierarchical teams, the study suggests that improved training for team leaders could help mitigate communication bottlenecks. Leaders must learn to encourage information exchange, invite peer verification and create an environment where team members feel empowered to question AI-generated guidance.