A Universal Turing Machine can take every possible Turing Machine and see which ones produce a given observation as output. This produces a set of Turing Machines that produce the given output. In other words, this produces the set of hypotheses that successfully explain an observation.