Probabilities beyond Belnap-Dunn logic: dealing with gaps, gluts and reliability

Logic and probability have long been seen as two different languages for describing the world. Logic is the study of how we reason, determining what must be true if other things are true. Probability is the study of uncertainty, measuring how likely an event is to happen. For centuries, these fields have operated under a strict assumption: that every statement is either true or false, and that our knowledge of the world is complete enough to assign a single number to the likelihood of any event. This works well for coin flips or weather forecasts, where the outcome is clear. But it struggles when we face information that is contradictory, incomplete, or simply unreliable. If a witness says a car was red, and another says it was blue, classical logic breaks down because the car cannot be both. If no one has seen the car at all, classical logic has no way to represent that total lack of information without forcing a guess.

To handle these messy realities, researchers have developed systems that allow for "gaps" (where we know nothing) and "gluts" (where we have conflicting information). These systems, known as non-classical logics, treat truth as a spectrum rather than a switch. However, a major challenge remained: how do you assign probabilities to statements in these flexible systems? If a statement can be both true and false, or neither, how do you calculate the chance of it happening? This question is crucial for artificial intelligence, legal reasoning, and any field where machines or humans must make decisions based on imperfect, conflicting, or unreliable data. Without a robust way to measure uncertainty in these complex scenarios, our tools for reasoning remain stuck in a world that is too simple to be real.

A team of researchers has now bridged this gap by introducing a new framework for probability based on a sophisticated six-valued logic system. Their work, published in a recent study, moves beyond the traditional four-valued systems that only account for true, false, both, or neither. By adding a crucial fifth and sixth dimension—representing the reliability or "classicality" of the information itself—they have created a method to measure not just whether a statement is true, but whether we can trust the source of that truth. This allows for a much finer-grained analysis of belief, distinguishing between a firm conviction, a shaky guess, a contradiction, and a total lack of evidence.

The researchers built their system on a logic called LET+K, which expands upon earlier models by introducing an operator that acts like a "reliability check." In everyday terms, this operator allows a system to say, "I have evidence for this claim, and that evidence is solid," or "I have evidence for this claim, but the source is untrustworthy." This distinction is vital. In a standard system, if you have a contradiction, you might be forced to treat it as a total failure of reasoning. In this new system, you can acknowledge the contradiction while simultaneously noting that the information causing it is unreliable, allowing the reasoning process to continue without collapsing. The team defined probability functions that work across six distinct regions of possibility: reliable belief, unreliable belief, conflict, unreliable disbelief, reliable disbelief, and total uncertainty.

To make these abstract concepts work, the authors developed a new way of visualizing these probabilities using what they call "twist structures." Imagine a map where every possible state of the world is divided into six specific zones, rather than just two. When a piece of information arrives, it doesn't just land on "true" or "false"; it illuminates specific parts of this six-part map. The researchers proved that their new probability functions are mathematically sound and complete, meaning they follow strict logical rules that prevent contradictions and ensure that the numbers always add up correctly. They showed that these functions can be defined in three different ways—one-dimensional, three-dimensional, and six-dimensional—and that all three approaches are perfectly equivalent, just looking at the same data from different angles.

A key achievement of the paper is the extension of "Jeffrey's update," a famous method for changing beliefs when new evidence arrives. In the classical world, if you learn a new fact, you update your probabilities based on that fact being certain. But in the real world, new evidence is often uncertain. The researchers adapted this update rule to their six-valued system. They demonstrated how to adjust probabilities when you learn that a statement is "somewhat reliable" or "somewhat contradictory," rather than just true or false. They provided both a mathematical definition and a logical proof that these two methods of updating are the same, ensuring that the system is consistent whether you look at it from the perspective of raw numbers or logical rules.

The study also explored how this system handles the classic "Bayesian update," which is used when evidence is certain. They showed that their new system naturally includes the old, classical methods as special cases. If the information is perfectly reliable and the contradiction is impossible, the system behaves exactly like the classical logic we are used to. But when the information is messy, the system flexes to accommodate the complexity. This means the new framework does not replace our existing tools; it generalizes them, offering a more powerful toolkit for situations where the world refuses to be simple.

The implications of this work are significant for any field dealing with uncertain data. By providing a rigorous way to measure the reliability of information alongside its truth value, the researchers have given us a way to reason about contradictions without losing our minds. They have shown that it is possible to have a mathematical theory of probability that respects the fact that some information is conflicting, some is missing, and some is simply untrustworthy. This is not just a theoretical exercise; it is a necessary step toward building artificial intelligence that can navigate the messy, contradictory, and unreliable nature of human knowledge. The paper concludes by outlining how this framework could be used to improve belief revision systems, allowing agents to update their views in a way that is sensitive to the quality of the evidence they receive.

In the end, this research offers a more honest way to think about uncertainty. It acknowledges that the world is not always black and white, and that our knowledge is often a patchwork of strong convictions, weak guesses, and outright contradictions. By creating a mathematical language that can describe all of these states simultaneously, the authors have provided a foundation for a more robust and flexible logic of evidence. Their work proves that we can measure the reliability of our beliefs just as precisely as we measure their likelihood, opening the door to a new era of reasoning in an imperfect world.