Logic is the study of inference. Instead approaching this subject from the verbal level - creating a verbal ontology of the concepts used and selecting the verbal inference rules or even implementing a probabilistic Bayesian approach - M-Logic starts with the idea of measurements made with concrete instruments. It builds from scratch a bottom-up theory of logic, with its concepts emerging from the M-Logic Machine that is being built. To my knowledge, this is an original approach.
M-Logic is question-driven. It only makes sense in a sensory-motor setting, although not all thought processes need to be tied to motor actions. All questions (and their answers) are made of a required context and an interrogation. For instance, "is the sky blue?" (actually the non-verbal question that corresponds to this verbal utterance) is answered first locating the sky (the required context), and then finding its color (the interrogation). The question-answer pair could be written, for instance, as (blue/sky; grey/sky). The non-verbal entities for "sky" and "blue" emerge from the sensory pathways and internal mappings of the machine. A logical value is generated when the question-interrogation is compared to the answer-interrogation found in memory records, up to the present moment. In this example the sky is found grey. If both instances of sky in the question-answer pair refers to the same memory, the agent then knows that the sky is not blue in that memory. Grey and blue are related because they share the same sensory pathways, instead of being related by some predefined verbal ontology. Since M-Logic works with cinematic memories (i.e. memories that record sequences of observations), questions are anchored in time: they can refer to the past, the present, or the future.
When no answer to a question can be found in the same cinematic memory where the question was placed, other records can often be used to generate beliefs. This is typical for questions that refer to the future. A special long-term memory, called the dominance-list memory, orders the cinematic records according to their predictive success. No probabilistic calculations are involved, just a reordering of the list. The belief is therefore justified, in the sense that it is the answer most frequently found correct. These beliefs are a form of inductive inference, while deductive inference is based on sequences of questions that were found to preserve truth.
The dominance-list memory handles exceptions to frequent situations in a simple intuitive way. For instance, if the required context is birds, it will usually generate a belief that birds fly. If the required context is penguins, it will generate a belief that penguins are birds that don't fly. Like a human set of beliefs, the M-Logic Machine set of beliefs often does not follow the strict coherency of classical logic.
The idea of truth and belief becomes operational, and internal to the machine. The machine inferences are made using its internal records. The inferences are adequate if the machine can perpetuate itself in a hostile world. The possible relation between the inner records and some external reality is irrelevant. In a very concrete sense, truth is life. M-Logic departs from the frequent AI understanding that a rational agent must optimize some performance measure. In that sense, the machine is often irrational.