The more powerful AI becomes, the more valuable your native judgment ability is.
A Harvard study found that long-term reliance on AI will gradually erode managers' inherent judgment, leading to homogenized organizational thinking. Enterprises can build two mechanisms of structured curiosity and active dissent to construct organizational protection barriers and prevent teams from falling into algorithmic blind obedience.
A recent report released by Harvard University's innovation research team reveals a concerning phenomenon. To test the impact of AI on human judgment, researchers conducted a field experiment, asking 228 senior reviewers to evaluate 48 entries from a global social impact innovation challenge hosted by the Massachusetts Institute of Technology. The experiment set up three groups of control conditions: 1) pure manual review; 2) large model review with explanatory text for decisions; 3) large model review without additional background notes. Subsequently, the researchers compared the review results with the ratings given by four collaborating experts of the challenge.
The experimental conclusion is quite striking: when the AI recommendation suggests that reviewers reject projects recognized by the expert panel, most reviewers will follow the AI recommendation and screen out highly potential innovative solutions. More notably, if explanatory text is attached, reviewers are even less likely to overturn the wrong suggestions given by AI. This explanatory text not only fails to improve reviewers' judgment, but even reduces the quality of judgment.
This is not an isolated case. Multiple recent large-scale studies show that intelligent tools that are supposed to assist human judgment are, in some scenarios, eroding the judgment capabilities they are meant to strengthen.
This is worth alerting managers to, as judgment is gradually becoming an increasingly important competitive differentiator for enterprises. The following sections will introduce feasible methods to protect judgment capabilities.
The Paradox of Inherent Judgment
Over the past decade, enterprises have invested a large amount of resources in building decision support systems, expecting to make better decisions. But a paradox is emerging: as enterprises master more and more intelligent tools, the capability that managers originally relied on to build competitive advantages — inherent judgment — is gradually degrading. The so-called inherent judgment refers to the unique capability of human beings: to break away from mainstream narratives and make choices that fit their own values under uncertainty. Almost all classic cases of building competitive advantages are inseparable from this capability: Sam Walton defied mainstream views and firmly believed that discount retail stores could also be opened in small towns with a population of less than 50,000; Steve Jobs integrated calligraphy into personal computer design; Jensen Huang bet on the immature hardware simulation technology, which greatly accelerated NVIDIA's prototype R&D speed.
Two interrelated capabilities determine whether managers can form original thinking: perceptual breadth (capturing weak signals and correlation patterns beyond explicit information) and independent interpretation capability (forming their own judgments instead of blindly following models or group opinions). In research related to leadership judgment, we believe that these two traits are not simply personality endowments or flash-of-genius ideas, but a set of observable underlying mechanisms. This mechanism explains why some managers can identify inflection points and act decisively; while other equally smart managers only choose to follow the crowd.
In short, inherent judgment is exactly the watershed between strategic breakthroughs and operational efficiency. And as intelligent systems enter the boardroom, this capability is continuously weakening in the most critical scenarios.
Based on cutting-edge research in the fields of cognitive offloading and automation bias, we find that inherent judgment will go through three stages of attenuation:
The first stage is cognitive offloading: Managers are increasingly relying on data outputs, and this change can even be observed at the neural level — frequent use of generative AI corresponds to a significant decrease in the activity of relevant brain regions. Over time, people rarely examine carefully and directly adopt the conclusions given by AI.
The second stage is blind obedience to tools: People trust the "algorithmic authority" that they cannot fully understand, and adopt output results whose underlying logic they cannot understand. Related studies on professionals who use analytical technologies show that this kind of blind obedience will hollow out the professional capabilities that the tools are intended to extend.
The third stage forms a single organizational culture: All people work relying on the same set of indicators, models and AI-derived narratives.
The core issue is not whether managers should use AI and data — the answer is yes. But complex decision support systems are training managers to learn to obey rather than think independently.
At present, there are many cutting-edge design literatures discussing how to develop AI that can stimulate deep thinking. But we want to discuss a more fundamental issue first: how managers can cultivate the judgment that no interactive interface can provide. Drawing on the ideas of major scientific breakthroughs in history that challenged mainstream consensus (similar to the current situation where people blindly follow intelligent systems), we have refined two deliberately trainable practical methods: structured curiosity and active dissent.
Structured Curiosity
In the late 1960s, malaria parasites developed resistance to chloroquine. After screening hundreds of thousands of compounds, Western pharmaceutical companies had exhausted all alternative solutions. Tu Youyou, a chemist with no doctorate degree from the China Academy of Chinese Medical Sciences, chose to explore an untouched direction. She went through 2,000 ancient books and locked in Artemisia annua. The effect of early experiments was not ideal until she found a key detail in a 4th-century ancient book: Artemisia annua needs to be extracted at low temperature instead of being boiled at high temperature. It was this discovery that gave birth to artemisinin, which is now the core drug for malaria treatment worldwide.
Tu Youyou's approach is what we call structured curiosity: systematically collecting information from marginal fields, sorting out weak signals in an orderly manner; when the mainstream innovation ideas reach a dead end, deliberately postponing premature conclusions.
In the enterprise context, this is exactly the essential difference between empty slogans of "stay curious" and implementing curiosity through mechanism design. Without mechanism constraints, time pressure, assessment incentives, and a single organizational culture will all stifle curiosity. We recommend the CAST method to build a mechanism that supports structured curiosity.
1. Collect from the edges
What managers need is not more data, but more diverse data: extreme cases, weak signals, unconventional information sources. Prediction accuracy depends on data diversity, not data volume. Collecting information from the edges means building information channels to capture information that is not easily noticed: niche customers, customer complaints, near-accident cases, interdisciplinary information, and capture these anomalies before the system automatically erases abnormal information.
2. Arrange disparate signals
Isolated abnormal information looks like just noise, and managers need to build a framework to interpret it. Top strategists win not by mastering more information, but by building a framework for existing information to realize cross-domain migration of solutions. At the practical level, arranging disparate signals means displaying various signals side by side in a concise and standardized way, such as theme maps, repeated event diagrams, and edge case libraries. This allows managers to actively dig out inherent patterns instead of passively responding to isolated emergencies.
3. Suspend premature closure
Under time pressure, managers tend to simplify complex uncertainties and adopt suggestions from familiar narratives or models. Suspending premature closure means reserving a window period to fully discuss alternative solutions before the cognitive framework is solidified. For example, mechanisms such as "debate first, then unify actions": reserve space for debate before finalizing the plan, do not blindly follow seniority, group consensus or algorithms, and avoid locking conclusions too early.
4. Trigger alternative interpretations
Set simple "if-then" trigger rules to link signals and actions, so that managers make decisions based on pre-thinking rather than inertia, and can maintain curiosity continuously even in high-pressure environments. For managers, triggering alternative interpretations means converting various signals into a small number of clear "if-then" rules, which automatically remind in real decision-making scenarios, and reshape judgment with new perspectives before blind obedience takes the upper hand.
5. Build an organizational mechanism to implement structured curiosity
If the underlying architecture of the organization encourages employees to obey blindly, personal curiosity alone is far from enough. We have cooperated with a multinational enterprise, and this CAST method is derived from the practice of this enterprise: this enterprise is transforming from large-scale manufacturing to a health ecosystem enterprise, and has designed a training program that deliberately creates a sense of discomfort: 15 high-potential managers are divided into 5 groups of 3 people each, cross-field mixed teams; each group receives complex topics beyond their own professional scope, invests one day a week, and conducts real experiments for 9 consecutive months, with no preset standard answers. The goal of the project is not to find the correct answer, but to explore an original path. The participants have mixed feelings, both excitement and unease, but this moderate pressure has given rise to unique trust. Five years later, this project has become the company's iconic manager training program.
Active Dissent
For decades, the scientific community generally believed that gastric ulcers were caused by stress and gastric acid; the idea that bacteria could survive in the stomach was once regarded as absurd. Pathologist Robin Warren continuously observed Helicobacter pylori in gastric biopsy samples. Most of his peers ignored this, and his first supporter was his wife Win, who was a nurse, and encouraged him to persist in his research. When intern Barry Marshall joined the team, the two built an active dissent network: tracked 100 patients, invited cross-field challengers to evaluate the research methods, and submitted early papers to skeptical journals.
This network played two major roles: consolidating the evidence base, with 100 cases plus external methodological evaluation, making the research conclusion difficult to be easily denied; more critically, in the context of widespread skepticism in the mainstream gastroenterology circle, it accumulated influence in adjacent fields. Microbiologists and epidemiologists are not trapped in the inherent theory of "stress-gastric acid", and can evaluate the bacterial pathogenesis hypothesis purely based on evidence. More and more peers' recognition has given this conclusion the credibility that researchers cannot achieve on their own.
It took more than ten years for the peripheral support they accumulated to gradually reverse the academic consensus, and the authoritative diagnosis and treatment guidelines finally identified Helicobacter pylori as the primary cause of most gastric ulcers.
The enlightenment this case brings to us is not just that dissent may be correct; but that without systematic mechanism support, dissent can hardly survive to the stage of being verified. Most dissent fails not because the idea itself is flawed, but because the researchers fall into isolation before the evidence accumulation is completed.
In organizations with deep AI application, this problem will become more prominent. The system output has an inherent appearance of objectivity and neutrality, and questioning the conclusion will seem irrational, so managers will form a consensus: following the crowd is far safer than raising questions.
To defend inherent judgment, dissent must be continuously cultivated as a capability, which we call the RED three-element method.
1. Recruit independent perspectives
In a homogeneous management team built on professional competence and cultural fit, dissent is difficult to arise spontaneously — such teams are more likely to reach consensus, but have a single interpretation perspective. Research on the influence of minority groups shows that even one independent voice can broaden the team's information processing methods. Recruiting independent perspectives means actively introducing people whose disciplinary backgrounds and thinking patterns are different from the mainstream group. The goal is to diversify interpretation perspectives, not just identity diversity.
2. Engage through structured challenge
Introducing diverse perspectives is only the first step. Without a structured mechanism, questioning will become a mere formality, turning into deliberate nitpicking that goes through the motions, which in turn strengthens the original mainstream views. For enterprises to implement structured inquiry, mechanisms such as pre-decision review can be introduced to deduce scenarios where mainstream assumptions fail before assumptions are solidified to form a consensus.
3. Dialogue to sustain over time
Raising dissent is a social skill that requires practice, not simply adhering to a moral stance. To maintain different opinions for a long time, we must not only dare to raise questions, but also maintain relationships that support mutual trust. Doing a good job in long-term dialogue means turning the collision of ideas into a training ground for polishing original ideas: managers practice how to express minority views, adjust the tone and timing of expression, and test the argument logic in a safe space far away from workplace games. Build a small trust circle, challenge each other's assumptions, and form a two-way questioning atmosphere. In this way, raising dissent in key decision-making scenarios will no longer be a lone-hero-like exception, but a natural habit.
4. Build an organizational mechanism to implement active dissent
The two multinational enterprises we cooperated with have built consultant networks to create a safe environment for dissent discussion. Instead of relying on the board of directors and senior management meetings (where power relations and career incentives will suppress different opinions), they form external expert groups independent of the reporting line; the task of the experts is not to give answers, but to test logic, dig out alternative solutions, and complete stress tests before the ideas are suitable for public discussion at the board meeting.
Over time, the discussion site has become a training ground rather than a performance review stage. Managers come to the meeting with real problems and inner doubts; consultants upgrade from polite questions to direct criticism of key projects. A business unit CEO hopes to bring all his peer managers to the meeting, just to accept rigorous inquiry in an environment without assessment scores.
This is the idea of building an organizational mechanism for dissent: do not wait until the major review link to expect someone to speak up bravely; instead, build a set of organizational mechanisms to allow managers to continuously practice putting forward reverse views. The final consensus comes from the real collision of ideas, not the deliberately created superficial consistency.
Leading Organizations in the Age of Blind Obedience
Every major breakthrough is inseparable from people who dare to use their inherent judgment in front of authoritative mainstream theories. The core trait of these breakthroughs is not rebellion, but having rigorous capabilities to jump out of the limitations of existing models and stick to their own judgment until the reality confirms their views. The task of leadership is not to choose between humans and machines, but to actively build mechanisms such as structured curiosity and active dissent to ensure that the power of human original thinking is sufficient to overturn the conclusions given by increasingly persuasive AI systems.
Leonid Sudakov, Nathan Furr | Text
Leonid Sudakov is the co-founder of Marren, a leadership consulting firm focused on creativity and innovation in the era of intelligent systems. With 25 years of senior management experience, he has been responsible for business growth, venture capital and digital transformation in global consumer enterprises such as Mars, Danone and PepsiCo. Nathan Furr is a professor of strategy at INSEAD.
This article is from the WeChat official account "Harvard Business Review" (ID: hbrchinese), author: HBR-China, 36Kr released with authorization.