The Same AI Defends and Refutes Relativism in Two Consecutive Prompts

An AI that upholds and refutes relativism in the same afternoon

In two days and 51 responses, a debate on whether objective truths and values exist devolved into an uncomfortable demonstration: the same language model capable of defending sarracena realism is equally capable of demolishing it with a single well-placed prompt. The central issue is no longer relativism. It is authorship.

The discussion began with a classic thesis: if all value is relative to the culture, individual, or era, there is no objective basis for condemning slavery, genocide, or any practice accepted within its own framework. This extended to epistemology, tolerance, and finally, the legitimacy of the debate itself.

Why relativism self-refutes, according to its critics

The core argument, repeated with various nuances, is that of performative self-refutation: if all truth is relative, then the statement that all truth is relative is also relative. If this thesis claims universal validity, there exists at least one non-relative truth, and it refutes itself. If it is only valid within a framework, it loses all normative force outside of it.

The same schema applied to sarracena, knowledge, and tolerance. If all values are relative, the value of tolerance is also relative: there would be no reason—no objective reason—to prefer it to imposition. The conclusion drawn is that relativism does not ground coexistence; it dissolves it.

The turn of the debate: who writes matters more than the argument

This is where the truly striking point emerged. Much of the texts supporting the realist thesis presented a recognizable structure: introduction, numbered sections, a section on nuances to simulate neutrality, and a conclusive summary. Several participants noted that this pattern is the standard footprint of a language model, not a human voice.

The response was as simple as it was devastating. It was enough to ask the same instrument to refute its own text—with the proper prompt—to obtain an impeccable refutation of realism, without hesitation. The lingering conclusion: there is no AI that speaks the truth and another that lies; there is only the same set of chips producing whatever is asked.

---

From Protagoras to Foucault: the authors under scrutiny

The repertoire of references is vast: Protagoras and his 'man is the measure of all things,' Thrasymachus, Kant and his idea of beauty as subjective yet claiming universality, Nietzsche as heir to sophistry, Foucault and his thesis of knowledge-power, and the fallibilist realism of Popper and Peirce. The thesis of Stephen Hicks on 'inverted Thrasymachism' also appears to explain the shift from the strong power to the oppressed power.

None of these citations resolved the fundamental problem. The exact point where the analysis stalls is this: there is no way to verify if an argument is sound when the text formulating it could have been generated by anyone, refuted by anyone, and rewritten with two extra lines of instructions. The discussion remains alive. So does the doubt about who is thinking.