What Are We Actually Evaluating When We Say a Belief “Tracks Truth”?

In a viral essay on the rationality forum LessWrong, philosopher Alex Glaucon challenges the centuries-old definition of knowledge as 'justified true belief' (JTB). He argues the 'truth' component is problematic because finite, bounded agents never have direct access to objective truth. Instead, Glaucon proposes knowledge should be understood simply as 'justified belief' (JB), where what we're actually evaluating is the quality of our epistemic procedures—the reliability of our 'map'—not correspondence with an inaccessible 'territory.'

Glaucon illustrates his point with a compelling thought experiment: an experienced guide, Chris, confidently states 'I know the ice is safe' based on twenty years of experience and current evidence. The group crosses safely, but unbeknownst to all, a hidden structural flaw exists. The ice later cracks. Glaucon argues nothing changed about the truth of Chris's statement between the safe crossing and the accident; the ice had been flawed for years. What changed was the 'live justificatory environment'—the emergence of a detectable 'defeater' to the belief. This shows knowledge claims are sensitive to available justification, not metaphysical facts.

The essay, gaining traction in AI and rationalist communities, suggests that for practical decision-making (especially relevant to AI alignment and reasoning systems), focusing on robust justification processes may be more valuable than chasing an elusive 'truth' criterion. It reframes the goal from 'tracking truth' to maintaining beliefs that survive rigorous scrutiny and updating procedures when new evidence appears.