summary: New research claims that AI failures in the workplace are rarely due to a lack of “intelligence” but rather a lack of “cognitive coordination.” This research suggests that treating AI as a “plug-and-play” tool creates friction, as humans and machines use fundamentally different logics to process information.
To succeed, teams must move to “hybrid cognitive alignment.” This is a gradual process in which humans and AI develop shared expectations through experience. The study emphasizes that the value of AI lies not in its standalone capabilities, but in its ability to act as a collaborative partner that understands its own limitations.
important facts
- “Logic gap”: AI relies on statistical patterns from data, whereas humans use judgment, social cues, and experience, leading to natural inconsistencies in task performance.
- Hybrid cognitive adjustment: This new process involves humans recalibrating their trust and adapting their interaction styles as they learn how the AI behaves over time.
- Dynamic tasks: Role sharing between humans and AI only works if the tasks are stable. In practice, if unforeseen events occur (such as a market crash), a fluid change of responsibility is required.
- Collaboration over performance: This research suggests that rather than simply pursuing raw performance, AI developers should prioritize “designing for collaboration” – ensuring that the system communicates its limitations.
sauce: Stevens Institute of Technology
In the iconic Star Wars series, Captain Han Solo and the humanoid droid C-3PO boast strikingly contrasting personalities. Han Solo is driven by emotion and bold self-confidence, often ignoring C-3PO’s logic-based caution. This relationship between humans and droids is exemplified by Solo’s famous statement, “Never tell me the odds!” He dismisses C-3PO’s advice against navigating through an asteroid field where the chances of survival are 3,720 to 1. This probability was carefully calculated by this shining companion.
In Hollywood classics, this comedic relationship makes for compelling drama, but in everyday reality, such dynamics do not work for a successful human-machine relationship.
As AI becomes part of many individuals’ daily lives today, humans and machines must learn how to work well together, says Bei Yang, an assistant professor at the Stevens School of Business who studies human-machine teamwork.
“Companies are using AI alongside humans, but it’s difficult to get them to work well together,” she says. “Humans think differently than AI. Humans use experience, judgment, and social cues. AI uses statistical patterns learned from data.”
These differences can complement each other, she added, but only if they are well coordinated. If not, users may overconfident the AI’s output, misuse the system, or waste time on fixes and workarounds.
“In these cases, AI does not reduce effort; it increases friction,” she says. “This mismatch often causes teamwork between humans and AI to underperform” and sometimes fail completely.
When companies analyze AI failures, they attribute them to one of two pitfalls. Either the technology is not powerful enough, or it is too powerful to be trusted. But Yang suggests another reason. This means that machines and humans are not well-coordinated to work together. “AI failures occur because there is a mismatch in how humans and machines understand tasks, roles, and responsibilities.”
When bringing AI into the workplace, companies tend to actively divide tasks between humans and AI, Yang notes. This only works if the task is stable, predictable, and does not change over time. However, this is not the case in most work environments.
As an example, Yang deploys AI to quickly monitor markets and use high-frequency trading algorithms to identify trends and opportunities. However, certain unexpected events, such as sudden market declines, major policy changes, or the release of inflation data, can distort AI’s understanding of the market.
“AI is not really designed to understand such events because the algorithms are trained with preset rules, which could change the entire market and even lead to a crash,” she says.
The title of her new paper is Mind-machine synchronization: Hybrid cognitive coordination as an emergent coordination mechanism in human-AI collaboration.Published in management academy In the March 18, 2026 issue of the Journal, Yang argues that an effective partnership between humans and AI should be built differently.
They must rely on a process called “hybrid cognitive adjustment.” This is about progressively developing shared expectations about what AI is for, how it should be used, and when human judgment should take precedence.
“This adjustment does not happen automatically when the system is deployed,” Yang says. “Instead, it emerges over time as people learn how AI works, adapt how they interact with it, and readjust their trust based on experience.”
For example, AI is currently being used in medical settings to analyze X-rays and CT scans. Because it’s trained on millions of images, it can often identify cancers and other problems better than doctors can see. However, without human input and oversight, the analysis is not as powerful because a particular patient’s medical history and patient response to medications are not well known.
Similarly, in a customer service setting, AI is trained on thousands of past interactions and can search internal policy documents in record time, but it may not be able to understand specific customer issues or needs. Without training people on how to properly use AI, many such efforts may not yield positive results.
So what should companies do when implementing AI? “We need to focus more on how tasks and roles are divided between humans and machines, and how that changes over time,” Yang says.
“Training that emphasizes how AI should be used and time for teams to adapt is essential,” she emphasizes. “Trying AI as a ‘plug-and-play’ solution often backfires. Treating AI as a new collaborator yields better results. For managers, these impacts are immediate,” she points out.
AI developers can also learn from this paper. This finding highlights the importance of designing for collaboration, not just performance. “A system should clearly communicate its capabilities and limitations, support user learning over time, and help users develop a strong partnership with the system,” she says. “Ultimately, the potential of AI lies not in making machines smarter on their own, but in making human-AI collaboration work better. It is coordination, not raw intelligence, that transforms AI from a source of frustration to a source of value.”
Answers to key questions:
answer: It’s probably a “mismatch” of expectations. If the AI doesn’t understand the specific context of a task as well as you do, it will waste time “avoiding” the task instead of processing it.
answer: This study suggests that power is not the issue, but alignment is. We overtrust or misuse AI because we don’t spend enough time learning its particular “personality” and limitations in real-world environments.
answer: In many cases, this is not the case. Most AI is trained based on preset rules and historical data. In the event of a “black swan” event, human judgment must prevail as AI lacks the “mental bandwidth” to understand the changes.
Editorial note:
- This article was edited by the editors of Neuroscience News.
- Journal articles were reviewed in full text.
- Additional context added by staff.
About this AI and neuroscience research news
author: Lina Zeldovich
sauce: Stevens Institute of Technology
contact: Lina Zeldovich – Stevens Institute of Technology
image: Image credited to Neuroscience News
Original research: The survey results are displayed below Academy of Management Journal.
