The Moment the Magic Died for Me

I remember exactly when I stopped being dazzled by AI.

It was when I understood what a transformer was. When I learned that “intelligence” was next-token prediction. When I saw that hallucinations weren’t rare bugs but a fundamental architectural property. When I understood embeddings, attention heads, and how RLHF works.

Before that, every ChatGPT response felt almost mystical. “How does it know this?!” After, it became: “Ah, it’s the most statistically likely distribution given the context. Makes sense.”

The spell broke. I didn’t stop using AI — I use it more than ever. But I use it with utilitarian skepticism, not with awe. And that internal shift, which I thought was just my professional maturation, has a name in science: the Magic Effect.

And a study published in the Journal of Marketing proved I’m not the exception — I’m the rule.

The Study That Defies Common Sense

The paper “Lower Artificial Intelligence Literacy Predicts Greater AI Receptivity,” published in the Journal of Marketing (volume 89, number 5), was conducted by Stephanie Tully (USC), Chiara Longoni (Bocconi University, Milan), and Gil Appel (George Washington University). The American Marketing Association highlighted the results in November 2025. Harvard Business Review covered it. Marginal Revolution discussed it.

The study involved 4 surveys, cross-country data from 27 countries, and 6 additional experiments with thousands of participants. The conclusion is unequivocal:

People with lower AI literacy are typically more receptive to AI. And this relationship is not explained by differences in perceptions of AI’s capability, ethicality, or feared impact on humanity.

The real mechanism is different: people with lower AI literacy are more likely to perceive AI as magical and experience feelings of awe in the face of AI executing tasks that seem to require uniquely human attributes — empathy, humor, creative insight.

When someone with low technical literacy sees ChatGPT write a poem, the reaction is: “This is incredible. It’s almost human.” When someone who understands the architecture sees the same thing, the reaction is: “It’s the statistical distribution of token sequences in the training corpus.” Same screen. Opposite reactions.

The Illusionist Analogy

The analogy the researchers use — and that haunted me for weeks — is perfect:

Before knowing how the trick works, everything seems incredible. Almost supernatural. You’re dazzled. But the second someone reveals the trapdoor, the double mirror, or the sleight of hand, the spell breaks immediately. It stops being magic and becomes just a man on stage.

That’s exactly what happens with AI. Consumers with low literacy see AI responses as manifestations of an almost mystical intelligence. They experience awe and reverence. And they’re extraordinarily receptive to integrating these tools into daily life.

Those who understand the mechanism see the same screen and see only what’s really there: a sophisticated statistical pattern-matching engine. Without the “magic” factor, the dazzle disappears and gives way to utilitarian skepticism.

The Executives’ Blind Spot

One of the study’s most revealing parts: researchers interviewed 36 C-suite executives from a major European insurance company and asked: “Which customer segment should we target for our new AI-based products?”

All of them, without exception, answered they should target consumers with higher literacy and technological knowledge. The logic seemed obvious: those who understand more, adopt more.

The data proved the absolute opposite. And this misalignment between executive intuition and actual consumer behavior is, in my opinion, one of the paper’s most practical findings.

Researchers validated the pattern across multiple demographic groups — university students, ordinary consumers, professionals — crossing national-level data from 27 countries. Always the same result: ignorance about what’s behind the code is the biggest fuel for blind acceptance.

The Dilemma Nobody Wants to Discuss

This is where the paper gets uncomfortable for the industry. The authors’ conclusion is clear: “Efforts to demystify AI may inadvertently reduce its appeal. Maintaining an aura of magic around AI could be beneficial for adoption.”

This deeply bothered me. Because it implies that transparency and adoption are in direct tension. The more honest you are about how AI works, the fewer people will want to use it. The more “magical” the experience seems, the more adoption — but based on false premises.

And when I connect this with other data I’ve researched this year, the picture gets even more complex:

The 76% of Americans who don’t trust AI (Quinnipiac, March 2026) are, paradoxically, those who use it least — but when they use it, they use it with healthier skepticism.

The 900 million weekly ChatGPT users include an enormous proportion of people who think AI is “almost magical” — and who are therefore more vulnerable to undetected hallucinations, excessive trust, and decisions based on unverified outputs.

The 88% of high performers reporting burnout are on the opposite side: they know how it works, use it massively, but suffer the consequences of a tool that’s “good enough to be useful, imperfect enough to be dangerous.”

What I Take from This (Personally)

After digesting this paper, I reached three conclusions that changed how I think about AI education:

First: AI education shouldn’t eliminate awe — it should redirect it. The problem isn’t that people are impressed by AI. The problem is that blind awe leads to blind trust. Ideal education doesn’t destroy the enchantment — it transforms “this is magic!” into “this is incredible engineering, but with specific limitations I need to understand.”

Second: transparency with design can resolve the tension. You don’t need to give a lecture on transformers for someone to use AI safely. You need UX that communicates uncertainty (citations, confidence indicators, “I don’t know”), without destroying the experience. Claude saying “I’m not sure about this” is transparency with design.

Third: “utilitarian skepticism” isn’t a loss — it’s an upgrade. When I lost my awe for AI, I gained something more valuable: effective use. I know when to trust and when to verify. I know when AI is the right tool and when it’s a trap. That judgment — which only comes with knowledge — is literally the most valuable skill of 2026, as I’ve repeated in post after post.

Conclusion: The Industry’s Great Challenge

The long-term challenge isn’t creating models that seem more “magical.” It’s building tools that deliver pragmatic value so unquestionable that people keep using them even knowing exactly how the trick works.

As the population becomes more digitally educated, the trapdoor effect will wear off. And when the awe ends for the majority — as it ended for me — only one question will remain: does the tool deliver real value, or did it only seem to while I didn’t know what was happening behind the scenes?

That’s the question that should guide every AI company. And the answer will determine who survives the next phase.

Share if this resonated:

I used to think AI was magic. Now I know it’s engineering. And paradoxically, that’s exactly why I get more value from it than ever.

Read Also