After examining regular probability, he examines the Bayes approach to probability. He shows the conditional probability equation first to provide a comparison. This equation is P(H/E)=P(HupsidedownUE)/P(E). The Bayes approach to probability is P(H/E)=P(E/H)P(H)/P(E). P(H/E) is revised probability after reviewing evidence. P(E/H) is the conditional probability of E in case of H. P(H) is the probability without any evidence. The Bayes approach allows for the estimate to change as more information comes in.

In intelligence analysis, the equation given is R=PL with R being the estimate of conditional probability of the hypothesis H after revising evidence E. P or prior estimate times the likelihood of an event or L. He advocates that this approach is best used in strategic warning; for example the probability of a terrorist attack. This approach forces analysts to quantify their estimates and reduce cognitive bias using competing hypotheses. The weaknesses of this approach in the vulnerability to false evidence, time constraints, and limited information.

He uses historical events such as the Cuban missile crisis and the tension between Russia and China during the Cold War to show the differences between conventional probability and the Bayes approach. The estimates were the same, but the analysts using the Bayes approach arrived at their conclusions faster than the conventional analysts.

Critique: Dr. Davide was very thorough in his study of the use of Bayesian networks in intelligence analysis. However, it was a difficult read because he uses mathematical language very often. His examples were also slightly outdated. An updated version of his study would be useful in the modern intelligence field.

Dr. Barbieri Davide. https://www.researchgate.net/publication/257933578_Bayesian_Intelligence_Analysis.

I agree that the study needs updated but I would also try and argue for a better method of evaluating analytic confidence. He doesn't seem to think that analysts have the ability to quantify there analytic confidence, but they do in most cases. So I agree that Bayesian would be great when applying it to intelligence studies, but in combination with current methods used for evaluating analytic confidence.

ReplyDeleteBayesian can provide analysts with the ability to quantify their estimates. The inability to quantify analyst's confidence leaves room for bias and makes the analyst unable to see things differently since they have to rely on past experiences in making estimates.

DeleteI agree with Matthew. Bayes should be added once other analysis was conducted to add (counter)weight to what we already found.

ReplyDeleteDo you think the weaknesses inherent in Bayesian Statistics can weaken the analysis because they add external pressure?

The issue at hand with this methodology is that it fails to remove cognitive bias in my opinion. It remains a challenge for analysts to objectively quantify there analytic confidence. Better methodology needed.

ReplyDeleteThat's attractive that Bayesian Statistics can net similar results in less time. However, as the above comments have highlighted, the issue of quantifying analytic confidence stands out to me.

ReplyDelete