A winning Siberian football team could betray a secret weapons city
Picture an American in 1950s Moscow reading that a tiny, snowbound Siberian town keeps beating the capital at football. The sensible leap is that the town is no longer tiny, and that the state has quietly packed it with talent for some hidden project. That leap from clue to conclusion is what logicians call inference.
Inference is a step from premises to what follows from them. Aristotle, in the 300s BC, already separated deduction from induction. Deduction draws conclusions guaranteed by premises known or assumed true; induction climbs from particular cases toward a general rule. Charles Sanders Peirce championed a third kind, abduction, which aims for the most likely explanation rather than certainty. The football puzzle is a textbook case: in a command economy that directs people where needed, an anomalously strong team from a remote place is best explained by a concealed, well-staffed programme.
Validity concerns shape, not truth. A correctly formed argument can start from a false premise and still land on a true conclusion: claim that all tall people are musicians, add that John Lennon was tall, and you correctly conclude he was a musician, for the wrong reason. The same valid form with a different false premise can prove he was French. What a valid form does guarantee is that true premises never yield a false conclusion.
People are not naturally good at this. Informal logicians keep long catalogues of fallacies, and cognitive psychologists trace many of them to mental shortcuts. Confirmation bias, demonstrated by the Wason selection task, leads us to hunt for supporting evidence when disconfirming evidence would tell us more. The Linda problem exposes the conjunction fallacy, in which a detailed, representative-sounding combination is judged likelier than one of its parts.
Machines took up the job early. Automated logical inference powered expert systems and later business rule engines, and the Prolog language checks whether a statement follows from a knowledge base by backward chaining. Told that all men are mortal and Socrates is a man, Prolog confirms Socrates is mortal, yet asked about Plato it answers no, because anything it has not been told is treated as false. Today the word also names the stage where a trained neural network produces predictions.
Source: Inference
