Semantic probabilistic inference of predictions

Prediction is one of the most important concepts in science. Predictions obtained from probabilistic knowledge, are described by an inductive-statistical inference (I-S inference). However, such an inference encounters a problem of synthesis the logic and probability that consists in the rapid decreasing of the probability estimates of predictions in the process of logical inference. The procedures for calculating estimates in the Probabilistic Logic Programming do not solve the problem. From our point of view, prediction can not be well combined with a logical inference. Logical inference should be replaced by calculations. The paper proposes a semantic approach to the calculation of prediction, when the inference is considered not as verification of the truth of some statement on the model, but as a search for facts in the model, predicting the statement with a maximum probability. To do this, the work defines a semantic probabilistic inference forcalculating the predictions. In the process of semantic probabilistic inference, estimates of predictions strictly increase. We prove in the paper that prediction estimates obtained by the semantic probabilistic inference are certainly not worse than the estimates obtained by the logical inference with the parallel calculation of these estimates.