In the mid 1990s, a man walked into a bank in Pittsburgh and robbed it in broad daylight. He wore no mask, used no disguise, and made no attempt to hide his identity. He even looked directly at the cameras. What is more striking is that he felt confident enough after the first robbery to walk into a second bank later that same day and do it again.
What makes this story worth examining is not the crime itself. It is the reasoning behind it. The man had covered his face in lemon juice. At some point, he had learned that lemon juice can be used as invisible ink. When applied to paper, it becomes visible only when heat is introduced. From that, he drew a conclusion. If lemon juice can make writing invisible, perhaps it could make him invisible to cameras as well.
He did not treat this as a guess. He tested it. He applied lemon juice to his face and took a Polaroid photograph. When the image appeared blank, he took that as confirmation. What he did not realize was what had actually happened. The lemon juice had gotten into his eyes, affecting his ability to see clearly. When he took the photo, he was not properly in the frame. The image appeared blank not because he was invisible, but because he had missed the shot entirely.
From his perspective, the experiment had worked. He moved forward with confidence. When he was arrested shortly afterward, he was genuinely surprised. As far as he was concerned, his reasoning had been sound.
This story has since been used to illustrate a well documented psychological pattern. People can feel most confident when they do not yet understand enough to recognize what they are missing. The issue is not ignorance in the traditional sense. It is incomplete understanding that feels complete.
That distinction is becoming increasingly relevant in how organizations approach artificial intelligence. AI tools are being adopted rapidly across functions. Teams are encouraged to experiment, explore, and integrate these tools into their workflows. The results are often immediate. Summaries are generated in seconds. Drafts appear fully formed. Analyses look structured and coherent. The outputs are not only fast, they are convincing.
This is where the risk begins. When outputs appear polished, they create a sense of confidence that can exceed the level of understanding behind them. A result that looks right is often accepted as right. Over time, this reduces the likelihood of questioning, validation, and critical review.
The challenge is not that people are using AI incorrectly. In many cases, they are using it exactly as intended. The issue is that usage does not automatically translate into judgment. Knowing how to generate an output is not the same as knowing how to evaluate it. This creates a gap between activity and capability.
Organizations often respond by focusing on training. They introduce sessions on prompting techniques, tool functionality, and use case examples. These efforts are valuable, but they tend to emphasize how to use AI rather than how to think with it. The more difficult questions remain unaddressed. When should AI be relied upon? What level of review is required? Where does human judgment remain essential? What defines a high quality outcome?
Without clarity on these questions, behavior becomes inconsistent. Some employees rely heavily on AI outputs with minimal oversight. Others hesitate to use it at all. Most fall somewhere in between, making decisions based on limited guidance and personal interpretation. The result is variability in quality, risk exposure, and overall performance.
Organizations that are navigating this shift more effectively are approaching the problem differently. They are not simply accelerating adoption. They are defining the role of AI within the work itself. They are clarifying where AI adds value, where it requires oversight, and where human judgment must remain central. They are making expectations explicit rather than assumed.
This does not eliminate uncertainty, but it changes how it is managed. When expectations are clear, behavior becomes more consistent. Decisions become more deliberate. Outputs become more reliable.
At its core, this is not a technology problem. It is a capability problem. AI makes it easier to produce, but it does not automatically make it easier to perform. When organizations equate output with improvement, they risk scaling inconsistency rather than effectiveness.
This is where AI literacy becomes critical. Not as a set of technical skills, but as a framework for understanding how AI fits into real work. It requires clarity around what is changing, what remains human owned, and what defines good performance in an AI enabled environment. It requires the ability to distinguish between outputs that are merely convincing and those that are truly useful.
The broader conversation about AI often focuses on speed, scale, and opportunity. These are important. But without a corresponding focus on judgment, clarity, and capability, they can create a false sense of progress.
The man who believed lemon juice made him invisible was not acting irrationally from his own perspective. He had a theory, he tested it, and he observed a result that appeared to confirm it. What he lacked was the context to interpret that result correctly.
Organizations face a similar challenge today. The tools are powerful, the outputs are compelling, and the potential is significant. But without a deeper understanding of how those outputs are generated and how they should be evaluated, confidence can outpace capability.
The question is not whether organizations will use AI. That decision has already been made. The question is whether they will develop the clarity and judgment required to use it well.
That shift from usage to capability is what ultimately determines whether AI becomes a source of advantage or a source of risk. It is also the focus of Implementing AI Literacy, which explores how organizations can move beyond activity and build the structures, expectations, and decision frameworks needed to translate AI use into meaningful performance improvement.
