the Free Will (TM) state machine uses this constantly updated training data to bias its decisions throughout its shelf life. so there are rules (like you stated) but there is also a huge set of data that alters the rule's parameters, to such extent that some times, new rules are created and old ones get discarded.