I take the stance that when praxis and outcome are aligned, the question of truthfulness becomes somewhat moot—at least for me and I believe this is an aspect of epistemic humility.
In some East Asian cultures, particularly from a religious or philosophical standpoint, knowledge is understood as serving a societal function rather than existing in isolation, un-tethered from the mechanics of the world. You could say that this stance on knowledge is instrumental in a sense, since it is understood that knowledge exists in service of human needs. All this to say, it is as fallible as any other aspect of the human condition.
I think this perspective can also be supported from other theoretical frameworks, in which the reliability of something does not entail its being a constant. The conditions under which knowledge is formed are not static. The demand for absolutism implied by requiring “100%” certainty may reflect a desire for control rather than careful observation or logical analysis.
From this perspective, I am not convinced that the example of AI hallucinating 10% of the time meaningfully captures epistemic reliability. One could argue that human knowledge is inherently unreliable, particularly given that we still lack a universally agreed-upon definition of knowledge. A reliable way of forming beliefs, in my view, is one that follows logical procedures; those procedures may be considered reliable insofar as their application successfully transfers across contexts and does not contradict other fundamental truths. In this sense, a student’s failure to fact-check is not so much an unreliable epistemic method as it is a case of epistemic negligence.
Finally, I would argue that generative AI does not “know” anything in the epistemic sense. Its knowledge base is effectively the accumulated content of the internet, and what it provides is better described as instrumental rather than epistemic knowledge. This may help explain why generative AI has gained particular traction in many Asian contexts, where the emphasis of discussion tends to be less on the system’s truthfulness and more on its instrumentality. I think due diligence and epistemic responsibility ultimately, is ours.