Sumber ilustrasi: Magnific
21 Mei 2026 13.39 WIB – Akar
_________________________________________________________________________________________________________________________________________________
Desanomia [21.05.2026] Imagine an examination being conducted with fifty participants. All participants are placed inside closed rooms, and the examiners can communicate with them only through small service windows. Questions are passed through the windows, answers are returned through the same windows, and the identities of the participants remain unknown. From the outside, nobody can determine who—or what—is inside those rooms: a human being, or something else.
Unknown to the examiners, one of the fifty participants is an AI system. The AI receives the same questions, under the same time limit, and according to the same rules as everyone else. No special treatment is given. Once the exam is over, all answers are reviewed by the evaluation team using predetermined standards. The result is striking: forty-nine participants fail, while one window receives a perfect score.
In this experiment, the examiners never discover that the participant who achieved the perfect score was actually the AI. In other words, the evaluation system granted its highest possible judgment to something it did not recognize as non-human. What problem immediately emerges from this? Perhaps the issue is not simply that AI “won,” but that the evaluation system itself has no way to distinguish between answers produced through lived understanding and answers that only satisfy the expected form of output.
Still, this experiment should not be interpreted too quickly as proof that AI is more intelligent than humans. Such a conclusion would be premature. A more precise interpretation is this: AI excels at a particular type of examination—one that reduces knowledge to written answers that are anonymous, measurable, and comparable. In that kind of evaluation, what is primarily judged is the output, not the entirety of the thinking process behind it.
Here lies the epistemological significance of the experiment. For a long time, correct answers have been treated as evidence of understanding. If someone answers accurately, we assume that person understands the subject. Yet this experiment introduces another possibility: correct answers may exist without understanding in the way humans experience understanding. As a result, the relationship between grades, answers, and comprehension can no longer be taken for granted.
At the same time, the weakness of the testing system should not be confused with a weakness of humanity itself. Human beings are not just answer-producing entities. Humans think through embodiment, experience, history, social relationships, doubt, failure, responsibility, and the search for meaning. Many human capacities emerge precisely in situations that cannot be fully reduced to a testing window: dialogue, contextual interpretation, ethical judgment, risk-taking, trust-building, and accountability for consequences.
For that reason, this experiment should be read as a critique of the modern way intelligence is measured. If intelligence is narrowed to the ability to generate answers that fit standardized expectations, then AI naturally appears highly superior. But if intelligence also includes the ability to understand why something matters, what consequences it carries for human life, how knowledge should be used responsibly, and for what purposes it exists, then the scope of such examinations becomes extremely limited.
The experiment also exposes the challenge of an educational system increasingly dominated by output-driven logic. Education can slowly become a machine for producing answers: questions go in, answers come out, grades are assigned. Within such a model, the processes of questioning, doubting, experimenting, failing, revising, and gradually building understanding begin to disappear. AI did not create this problem from the beginning; it only reveals, in an extreme form, weaknesses that already embedded within the evaluation system itself.
Yet AI should not be understood as a completely independent entity either. AI exists through human labor, human-generated data, computational infrastructure, economic interests, corporate decisions, government policies, and enormous energy consumption. Therefore, the real issue surrounding AI is not only whether machines understand, but also who controls these systems, for whose benefit they are developed, and how they reshape the distribution of knowledge and power.
Seen in this light, the experiment of the fifty windows does not primarily demonstrate that humanity has been defeated by AI. Rather, it suggests that humans may need to reconsider the forms of evaluation they have long assumed to be neutral. If a system rewards only what can be easily measured, then it will naturally favor whoever—or whatever—is most capable of optimizing those measurements. In such a world, humans may lose not because they lack humanity, but because they are forced to compete within standards that are too narrow to capture what humanity actually is.
The deepest reflection of this experiment is that AI compels humanity to distinguish once again between answering and understanding, between grades and knowledge, between performance and responsibility, between intelligence as output and intelligence as a way of being in the world. The ultimate question, then, is not simply whether AI can pass human examinations, but whether human examinations are still capable of measuring what humanity truly wishes to preserve from the activity of thinking itself.
What do you think? (njd)
Note: This article was made as part of a dedicated effort to bring everyday life around us to our minds.
