The Role of Expert Judgment in AI

Last month, a Fortune 100 CIO said their company spent millions on an AI decision system that their team actively sabotages daily. Why? Because it optimizes for data they can measure, not outcomes they actually need. This isn't isolated. After years advising tech leaders, I'm seeing a dangerous pattern: organizations over-indexing on AI for decisions that demand human judgment. Research confirms it. University of Washington studies found a "human oversight paradox" where AI-generated explanations significantly increased people's tendency to follow algorithmic recommendations, especially when AI recommended rejecting solutions. The problem isn't the technology. It's how we're using it. WHERE AI ACTUALLY SHINES: - Data processing at scale - Pattern recognition across vast datasets - Consistency in routine operations - Speed in known scenarios - But here's what your AI vendor won't tell you: WHERE HUMAN JUDGMENT STILL WINS: 1. Contextual Understanding AI lacks the lived experience of your organization's politics, culture, and history. It can't feel the tension in a room or read between the lines. When a healthcare client's AI recommended cutting a struggling legacy system, it missed critical context: the CTO who built it sat on the board. The algorithms couldn't measure the relationship capital at stake. 2. Values-Based Decision Making AI optimizes for what we tell it to measure. But the most consequential leadership decisions involve competing values that resist quantification. 3. Adaptive Leadership in Uncertainty When market conditions shifted overnight during a recent crisis, every AI prediction system faltered. The companies that navigated successfully? Those whose leaders relied on judgment, relationships, and first principles thinking. 4. Innovation Through Constraint AI excels at finding optimal paths within known parameters. Humans excel at changing the parameters entirely. THE BALANCED APPROACH THAT WORKS: Unpopular opinion: Your AI is making you a worse leader. The future isn't AI vs. human judgment. It's developing what researchers call "AI interaction expertise" - knowing when to use algorithms and when to override them. The leaders mastering this balance: -Let AI handle routine decisions while preserving human bandwidth for strategic ones -Build systems where humans can audit and override AI recommendations -Create metrics that value both optimization AND exploration -Train teams to question AI recommendations with the same rigor they'd question a human By 2026, the companies still thriving will be those that mastered when NOT to listen to their AI. Tech leadership in the AI era isn't about surrendering judgment to algorithms. It's about knowing exactly when human judgment matters most. What's one decision in your organization where human judgment saved the day despite what the data suggested? Share your story below.

Many people are rushing to use AI for data analysis, but too often, they treat it like a magical drop-in analyst. They upload messy datasets, skip defining the research question, and ask AI to "find insights." What they get in return may sound smart, but often lacks any real statistical or contextual foundation. This creates a dangerous illusion of analysis. It looks polished, but under the surface, it’s guesswork wrapped in jargon. AI models, especially large language models, are not trained to understand your domain, your business rules, or the causal logic of your data. Unless carefully guided, they do not check assumptions, test for statistical significance, or understand which results are valid and which are noise. They might even fabricate insights when uncertain, presenting plausible explanations that simply aren’t true. And yet, people trust these outputs because they look complete and confident. That does not mean AI has no place in data work. Used properly, it can significantly accelerate the analysis process. It helps with cleaning data, transforming variables, summarizing trends, translating code, and generating visualizations. It can also help non-experts understand findings in plain language. But we must be very clear about what AI can and cannot do. It is a support tool. It is not a replacement for human thinking, especially when it comes to defining the problem, choosing the right model, or interpreting results within a meaningful context. Good analysis requires more than output. It requires judgment. You still need to know what question you're answering, how your variables work, what your sampling biases are, and whether your model assumptions are satisfied. You still need to make sense of outputs in the context of your actual problem. No tool can think for you. If you’re integrating AI into your research or analytics work, here’s a better path. Start with clear intent. Know what you're trying to learn, who it’s for, and what decision it will inform. Use AI for specific, bounded tasks, and guide it carefully with detailed prompts. Then validate everything it gives you. Check the math. Check the logic. Ask whether the insight actually holds in your context. And always review outputs with a critical human eye. We are not in the age of AI-powered insight. We are in the age of AI-augmented judgment. Tools are evolving fast, but responsibility stays with us. The best analysts and researchers will be the ones who use AI to go faster, but never hand over the wheel.

A powerful partnership is the strategic integration of artificial intelligence and attorney expertise. While AI systems excel at reviewing thousands of documents in minutes and spotting patterns across vast datasets, the irreplaceable value of human legal judgment is becoming increasingly apparent. Forward-thinking legal organizations are now identifying critical "judgment points" throughout their workflows—those moments where contextual understanding, ethical navigation, and strategic thinking cannot be automated. The modern legal professional isn't threatened by AI but rather empowered by it. The modern legal professional is developing new skills to effectively question machine-generated insights and reconcile divergent assessments when human and AI analyses differ. This balanced approach creates space for the judgment, creativity, and wisdom that represent our profession's highest contributions. As technology handles routine information processing, attorneys can focus on what truly matters: providing the strategic guidance and ethical reasoning that clients ultimately seek. The question isn't if machines will replace lawyers—it's how legal practice evolves when professionals are freed from drowning in document review to focus on truly meaningful work. What aspects of your legal practice benefit most from this human-AI partnership? I'm Colin, author of The Legal Tech Ecosystem and General Counsel of Malbek, CLM for the Enterprise. #legaltech #innovation #law #business #learning

I’ve been spending a lot of time recently working with AI—specifically testing how it can support the creation of clinical care plans. And I have to say: it’s incredibly impressive. The speed, the clinical completeness, even the nuance at times—this technology has come such a long way in such a short time. It’s clear that AI will have a meaningful role in the future of medical decision-making. But the more I use it, the more I keep coming back to one unresolved concern. As physicians, we’re not just trained to interpret medical information—we’re trained to filter it. We constantly consider not only what to say to a patient, but how and when to say it. We weigh whether a piece of information will help or harm, reassure or confuse, empower or overwhelm. That act of filtering—of protecting patients from unnecessary harm while guiding them forward—isn’t just a soft skill. It’s a core principle of clinical practice. It’s baked into our training, our experience, and our professional oath to “first, do no harm.” AI doesn’t yet know how to do that. It doesn’t understand when not to speak, or how to hold back in the service of care. That’s not a fault of the technology—it’s simply not what it was designed to do. Right now, that’s not a crisis. Physicians are still in the loop, reviewing and tailoring AI-generated care plans with the appropriate clinical and emotional judgment. But as we move toward a future where AI plays a larger role in autonomous care delivery, I worry we may lose that essential human filter. And with it, one of the most important safeguards in medicine. The promise of AI in healthcare is real—and exciting. But if we’re going to do this right, we need to make sure we’re not just scaling information, but also protecting the art of interpretation.

The more we study human/AI collaboration the more we realize how difficult it is to speak in absolutes. We are easily sucked into the idea that #AIautomation will solve all of our problems, until it doesn't. Thx to my good friend Bas van de Haterd (He/His/Him) for sharing this excellent study "Falling Asleep at the Wheel: Human/AI Collaboration in a Field Experiment on HR Recruiters," by Fabrizio Dell'Acqua of Harvard Business School. The study explores the dynamics of human effort and AI quality in recruitment processes and reveals yet another paradox of AI: Higher-performing AI can sometimes lead to worse overall outcomes by reducing human engagement and effort. When it comes to hiring, this finding is pretty significant. Especially when one layers in the presence of bias that (hopefully) can be mitigated by the efforts of recruiters to be objective (We can dream can't we!). Here is a quick summary of the article's findings and implications. Key Findings: 💪 Human Effort vs. AI Quality: As AI quality increases, humans tend to rely more on the AI, leading to less effort and engagement. This can decrease the overall performance in decision-making tasks. 🙀 Lower Quality AI Enhances Human Effort: Recruiters provided with lower-performing AI exerted more effort and time, leading to better performance in evaluating job applications compared to those using higher-performing AI. 🎩 Experience Matters: More experienced recruiters were better at compensating for lower AI quality, improving their performance by remaining actively engaged and using their expertise to supplement the AI’s recommendations. Implications for Talent Acquisition Leaders: ⚖ Balanced AI Integration: While it may be tempting to implement the most advanced AI systems, it’s crucial to ensure that these systems do not lead to complacency among human recruiters. Talent acquisition leaders should focus on integrating AI tools that enhance rather than replace human judgment. 💍 Training and Engagement: Investing in training programs that encourage recruiters to critically assess AI recommendations can help maintain high levels of human engagement and performance. 🛠 Custom AI Solutions: Consider developing AI systems tailored to the specific needs and skills of your recruitment team. Custom solutions that require human input and oversight can prevent "falling asleep at the wheel" and ensure optimal performance.

Recently, I was working on a complex SQL query for a user funnel analysis. The query would grow to hundreds of lines, involving multiple event tables at different granularities, handling various user segments, and calculating numerous metrics - the kind of query that makes you reach for another coffee. Typically, I break such problems into multiple simpler queries and combine them sequentially as CTEs, validating results at each step. As the query grows, my confidence in the final results increases. This approach also helps me catch and fix edge cases that might silently corrupt the results. Instead, I gave Claude a few sample SQL queries referencing the source tables and asked it to create a new query. After I provided some context and the expected output, it quickly generated an impressive query with well-structured CTEs, window functions, and complex JSON extracting functions. Huge efficiency win! But something felt off about the results. Upon review, I found three issues: ▪️ The query wasn't accounting for order cancellations, which inflated certain metrics ▪️ There was a missing date join between tables at different grains that resulted in multiple duplicated rows ▪️ The query joined tables with timestamps in different time zones leading to misleading results All three were subtle errors that substantially skewed the results. But here's the interesting part: Even with time spent reviewing and fixing these issues, using AI reduced my development time from about 3 hours to 30 minutes. It's like having a quick assistant who can write solid starter code, though you need to double-check or even triple-check their work. 💡 AI is great for speeding up data science work, but it works best when paired with human expertise to catch these kinds of subtle issues.

What does AI in cybersecurity really mean? It’s more than just automation and faster threat detection. While AI is a powerful tool in the cybersecurity landscape, it’s not the whole answer. Here’s why human expertise still plays a crucial role: → AI Isn’t a Silver Bullet AI can analyze vast data sets and spot patterns—but it can’t grasp the nuances that a human can, like detecting complex social engineering attacks or understanding context in real time. → False Positives? AI Needs Human Insight AI often flags benign activities as threats due to its reliance on algorithms. Human analysts are essential to interpret these alerts and separate real threats from false alarms. → Humans See the Bigger Picture Cybersecurity isn’t just about technology—it’s about understanding human behavior and organizational dynamics. Experienced professionals can spot emerging threats AI hasn’t yet recognized. → Evolving Threats Need Human Adaptability Cybercriminals are constantly innovating, and while AI can handle known attacks, humans are better at adapting quickly to new, evolving threats and devising strategies to counter them. → Collaboration is Key AI should enhance—not replace—human decision-making. When used together, AI automates routine tasks, allowing cybersecurity experts to focus on complex, critical issues. The takeaway? AI is a powerful ally, but it’s human intuition and expertise that make cybersecurity truly effective. How is your organization balancing AI with human expertise? Let’s discuss how the combination can strengthen your cyber defense!

One challenge of using AI in healthcare is ensuring the right people use it in the right situations. In the 1980s, Berkley professors and brothers Hubert and Stanley Dreyfus explained how we change as we acquire skills in their book, “Mind Over Machine: The Power of Human Intuition and Expertise in the Era of the Computer.” As the figure shows, as we become more skilled, we rely less on rigid, context-free rules and more on experience and intuition. We move from being detached observers to involved practitioners. Think of different types of drivers. Teenagers learning to drive must continually consider what each button, pedal, or movement does. Newly licensed drivers can handle routine conditions but struggle to parallel park or manage city traffic. London cab drivers can easily navigate busy city streets during thunderstorms. We see this in clinical medicine, too. While patients know themselves best, they often lack the perspective and vocabulary to interpret clinical information in context. Novice clinicians can handle a lot but tend to pay attention to too many details and interpret rules too rigidly. Experienced clinicians use intuition to separate signals from noise and perform each move naturally. As AI enters clinical medicine, we must make sense of its output. Patients can sometimes do this independently, such as initially navigating care and some chronic disease self-care. Novice clinicians are needed to help manage routine clinical conditions. But, we will always need experts for the many less straightforward clinical scenarios that lack single right “answers.” Matching clinical needs with the right level of expertise has always been important. It will become even more critical in the future.

To use AI well, we need human expertise and judgment. But we’re cutting off the very pipeline that provides it. AI can convincingly generate responses that look brilliant, especially to the untrained eye, but this can also include fabrications and misinterpretation of nuance. This is why we need deep human expertise to can spot the difference and effectively wield these powerful tools. >>> This is the problem we’re racing toward: As we automate more of the foundational work that once built expertise, and plug junior talent into short-term AI training roles with no long-term arc, we’re not just accelerating AI. We’re hollowing out the very judgment we’ll rely on to keep it aligned. This is the real crisis: not that AI makes mistakes, but that we’re dismantling our ability to recognize them. That’s not just a workforce issue. It’s a strategic failure. We are solving for short-term efficiency and undermining the long-term capacity we’ll need to govern these systems wisely. >>> Here’s what’s happening: This generation enters a turbulent job market. They have education, but little experience. Businesses see an opening: smart, affordable talent to annotate and train models. But these roles rarely lead to career-building paths. Meanwhile, seasoned experts will retire—and we don’t have replacements in the making. The result? A fragile AI future. Fewer people who can challenge model outputs, who understand both context and consequences. >>> What we need to be exploring now: How do we bootstrap the next generation of expertise? And that takes all of us: 1. Industry: How can we ensure we don’t treat AI training roles as disposable? How can we create onramps? Fund apprenticeships? Link these jobs to richer skill development> 2. Early career professionals: Explore how to use your unique vantage point. You see how AI is evolving, you are working on it every day: use that to find what it will . Become the person who can do what AI can’t. 3. Everyone else: Let’s really use this moment to amplify the conversation. There is no playbook here, we’ve never had to grow human expertise in the shadow of a system this fast and powerful. If we fail to build human capability alongside machine capability, we don’t just lose jobs, we will lose judgment, and that cost will come due just as AI’s power peaks. Let’s not wait for that reckoning, let’s take a long view of what we will need. >>> Please share your thoughts, and let’s get this conversation going: > How do we grow real expertise in a world where “learn by doing” work is disappearing? > What new kind of  role or program could “bootstrap” the next generation of experts? > If you're early in your career: What do you wish leaders understood about what it’s like to navigate this moment? ____ 👋 Hi, I'm Alison McCauley. Follow me for more on using AI to advance human performance. https://lnkd.in/gYYUA_E6?

Some engineers worry that AI will replace complex decision-making, but that’s the wrong way to think about it. Agentic AI excels at retrieving and synthesizing information across vast systems at speeds humans simply can’t match. But judgment, intuition, and high-level problem-solving still belong to humans. Consider software engineering. So much of it involves toil: repetitive, interruption-driven tasks that don’t require deep human reasoning but demand system-wide knowledge. AI can take over the grunt work: scanning logs, monitoring system changes, and surfacing critical insights. Humans step in only when context and strategic thinking are required. The future of engineering isn’t about replacing human decision-making; it’s about elevating it. It’s also not about having fewer engineers but dramatically accelerating the technology output. Agentic AI will make humans exponentially more effective.