Plausible Lies
Why do we come to trust AI explanations?
The fact that AI can generate inaccurate information is now common knowledge. In fact, when one opens an interface for systems such as ChatGPT or Gemini, warnings that responses may contain errors are explicitly presented as part of official company policy.
Asking why AI misinformation is dangerous is therefore no longer sufficient. The more fundamental questions lie elsewhere. Why do falsehoods produced by AI feel true to us? How does an explanation secure a user’s trust before any process of verification begins? Why do we treat these outputs not as tentative reference points, but as something approaching a correct answer?
Here lies the central argument. The most dangerous aspect of large language models, hereafter referred to as LLMs, is not that they occasionally state false information, but that they automate falsehood in the form of persuasion. LLMs are not systems designed to evaluate the truth value of information. Nevertheless, they implement with remarkable sophistication the linguistic forms of explanation, organization, judgment, and conclusion. In this process, persuasion does not depend on prior validation of truth. Once a rhetorical form achieves a sense of completeness, its content acquires credibility regardless of whether it is true.
This form of persuasion, however, is not a minor technical flaw or a temporary malfunction of artificial intelligence. It is the structural outcome toward which commercially successful AI products in contemporary society naturally converge. Systems that deliver explanations that are fast, accommodating, and easy to understand increase user satisfaction. Higher satisfaction leads directly to longer usage times and higher return rates. Within this process, the accuracy or intellectual honesty of explanations may be evaluated through indirect signals, but it is not the primary criterion. It is certainly not the only one. What ultimately becomes the key competitive advantage is how smoothly an explanation can take over the work of judgment itself.
ISBN: 979-11-997073-4-4 (Paperback)
ISBN: 979-11-997073-5-1 (E-book)