In most of these cases the defending side's intelligence did detect evidence of the impending attack; the failure was one of interpretation. It is difficult for an individual seeing the raw intelligence to recognize the evidence of a surprise attack, and even more difficult for him to persuade (in time) the organization that must respond. This is the downside of Bayesian inference--because surprise attacks are rare and have low a priori probability, they are difficult to recognize. The classic account is that of Wohlstetter .
Are such intelligence failures inevitable? The difference between intelligence and scientific research is that intelligence deals with a consciously deceptive adversary. The scientist constructs a hypothesis about nature and performs experiments to test its implications. The experiments may be technically difficult, the scientist may make mistakes and erroneous scientific results are common, but they can always be corrected by repeated and improved experiments. In contrast, an intelligent adversary is capable of performing the same analysis as the intelligence analyst, and therefore of knowing what false information to provide to thwart him. Of course, the adversary faces the same problem; the analyst is his adversary.
The intelligent adversary may be not only deliberately deceptive but deliberately unpredictable (for example, by choosing among alternatives only at the last moment), while nature does not confuse or deceive consciously or intentionally. The interpretation of secret intelligence is limited by two factors: