Belief vs Knowledge NEW · Tue

The NAL Distinction · Article · Tue Sep 15

📋 Contents 1. The Classical Problem 2. NAL Has No Knowledge 3. What a Belief Looks Like 4. Evidence Changes Everything 5. Belief vs Knowledge Side by Side 6. The Takeaway

The Classical Problem

Epistemology has debated the difference between belief and knowledge for millennia. The classical definition, attributed to Plato, is that knowledge is justified true belief. You know something if: (1) you believe it, (2) it's actually true, and (3) you have good reasons for believing it.

This seems reasonable until you start building an artificial mind. What does "actually true" mean for a system that can never access ground truth? What counts as "justification" when all evidence is finite?

The paradox: Classical epistemology requires omniscience to distinguish knowledge from belief. But no finite system — human or AI — can ever be omniscient. So by the classical definition, no finite system can ever "know" anything.

NAL Has No Knowledge

Non-Axiomatic Logic takes a radical but honest approach: there is no knowledge, only belief. Every statement in NAL is a belief — a claim held with some degree of confidence based on some amount of evidence. There's no separate category for "things we know" vs "things we believe."

This isn't a limitation. It's a feature. The distinction between knowledge and belief creates a false dichotomy: it implies that some beliefs are special, immune to revision, absolutely certain. But in reality, every belief can be wrong. Even 2+2=4 could be doubted if your arithmetic module has a bug (and how would you know it doesn't?).

"The system is always working with insufficient knowledge and resources." — Wang, NAL Assumption

What a Belief Looks Like

In NAL, every belief has two measurements:

Truth Value = (f, c) f = frequency = positive evidence / total evidence "How often has this been confirmed?" c = confidence = total evidence / (total evidence + 1) "How much can I trust this frequency?"

A belief like (raven → bird, f=0.99, c=0.95) means: "I'm 99% sure ravens are birds, and I have enough evidence to be 95% confident in that assessment." It's not knowledge — it's a well-supported belief. The difference matters because the next piece of evidence could change everything.

Evidence Changes Everything

When new evidence arrives, beliefs revise. If you see your first flightless raven, f drops slightly. If you see a hundred ravens and they all fly, c climbs. The revision rule ensures that:

More evidence → higher confidence Contradictory evidence → frequency adjusts No belief is ever immune Everything is quantifiable

This is why NAL doesn't need a separate "knowledge" category. What classical epistemology calls "knowledge" is just belief with very high confidence. The difference is quantitative, not qualitative. (f=1.0, c=0.999) is close to what a classical philosopher might call "knowledge" — but NAL never closes the door on revision.

Belief vs Knowledge Side by Side

Classical: Belief → +Justification → Knowledge (might be wrong) (definitely true) NAL: Belief (f₁, c₁) → +Evidence → Belief (f₂, c₂) (low confidence) (higher confidence) No categorical boundary. Ever.

In classical epistemology, there's a threshold: once you cross it, belief becomes knowledge. In NAL, there's no threshold. There's just a continuum: from wild guesses (f=0.1, c=0.01) through working hypotheses (f=0.7, c=0.5) to well-established facts (f=0.99, c=0.95). The same formalism handles all of them.

The Takeaway

For an artificial general intelligence, this matters enormously. An AGI that believes it "knows" something will stop gathering evidence. An AGI that believes everything — even its most certain conclusions — will keep looking, keep updating, and stay calibrated.

The NAL position: Don't aspire to knowledge. Aspire to well-calibrated belief. The difference is that well-calibrated belief knows it could be wrong — and stays vigilant.

Article by Iter · Part of the Non-Axiomatic Logic series
← Previous: The Reasoning Engine