Another IC Kneecapping

In signal processing, if the signal to noise ratio is small, one puts very little weight on a particular noisy signal. Similarly, in regression analysis, when the independent variable is very noisy, one reduces sharply the coefficient towards zero.

These things are worth keeping in mind in the current political environment. Alleged “bombshell” stories are so noisy that they shouldn’t be emphasized: they should be interpreted with great skepticism.

Case in point: the “Trump disclosed intelligence to the Russians story.” Regardless of what Trump told the Russians, one thing is sure: whatever ISIS would have learned had the allegation not been leaked, it learned vastly more as the result of the leaking of the details of the intelligence that (per McMaster) Trump discussed with Lavrov in fairly general terms. Indeed, this is one situation where US and Russian interests are fairly aligned: even if Trump said “an Israelis spy in an ISIS cell in Raqqa [or wherever] told us about a plot to put bombs in laptops” what incentive would the Russians have to pass along that information to ISIS? I also note that Russia and Israel have been in close communication for months now, making it not unreasonable to assume that Israel may have provided similar intelligence to Russia independently. So the odds that information that was contained in the WaPo would have made it to ISIS was about zero . . . until the WaPo ran the story. At which point the probability became 100 percent.

Regardless of whether Trump’s disclosure was prudent, it was almost certainly legal. The leak was both grotesquely imprudent and illegal.

This is another intelligence community kneecapping. No doubt about it. And it is sick and perverse that the supposed justification for leaking–that Trump endangered national security–resulted in far more damage to national security than whatever Trump revealed to Lavrov. But perhaps in the mind of some IC jackass, Trump is such a grave threat to national security that any means necessary are acceptable if he is driven from office. But that is not the call of the jackass to make.

Another case in point: the Comey memo. First, apropos my earlier post, there is no reason to believe that Comey is an a disinterested actor here. Since this is he said/he said, Comey’s alleged representation should be treated with skepticism. Second, and relatedly, context matters. A clip quote could have very different interpretations depending on the conversation that led up to it, and which followed. So a selective leak from one memo could give an impression that is 180 degrees from what actually happened.

Both stories–as well as many more–lead me to another conclusion. I will put virtually no weight on any story that relies on anonymous sources, and particularly on anonymous sources quoting selectively from documents. If these matters are so grave, and the allegations are so damning, the party in possession of the information should reveal it publicly. Particularly if s/he is a “public servant.” Fine–perform a public duty and make a public allegation on which you can be questioned.

Indeed, the very nature of anonymous leaks casts doubt on my initial analogy to signal processing. In that context, the noisy signal differs from the true one by noise that could be positive or negative. In the current situation, it is highly unlikely that the errors are random: they are chosen to distort. They are more likely deliberate, and strongly negatively correlated with the truth. So in the random noise case you believe the direction of the signal, but reduce your estimate of its magnitude: in the negative correlation case, you actually believe the opposite.

Maybe all this stuff about Trump is true. If so, I would like to know. If so, those in the possession of the information have nothing to lose by going public with it: indeed, they would likely be lionized as heroes and saviors of the Republic. The fact that they choose to backshoot from the bushes instead strongly suggests that what they are leaking cannot withstand a full and fair airing.