Computer Models Find Patterns In Asymmetric Threats

The Narrative Fallacy brings us a story about a project by University of Alabama researchers to develop a database capable of anticipating targets for future guerrilla attacks. Quoting Space War: "Adversaries the US currently faces in Iraq rely on surprise and apparent randomness to compensate for their lack of organization, technology, and firepower. 'One way to combat these attacks is to identify trends in the attackers' methods, then use those trends to predict their future actions,' said UA-Huntsville researcher Wes Colley. 'Some trends from these attacks show important day-to-day correlations. If we can draw inferences from those correlations, then we may be able to save lives by heightening awareness of possible events or changing the allocation of our security assets to provide more protection.' Researchers reviewed the behavior signatures of terrorists on 12,000 attacks between 2003 and mid-2007 to calculate relative probabilities of future attacks on various target types."

Comments

game theory

[Anonymous Coward]

It's all part of game theory. If your enemy doesn't randomize their tactics, then you can take advantage of any statistical bias or pattern. Soon anyone buying a geiger counter, thermal noise diode, or even a lava lamp, will be a candidate for the terrorist watch list.

A step down more like

[EmbeddedJanitor]

The computer model throws a double six as you walk past the screening point. You get selected for The Glove Of Fun.

Computer models are only as good as their data: Garbage In, Gospel Out. That's a problem with climate modelling. The climatologists keep tweaking the models until they get what they expect and are then smug because the models "prove" their predictions.

If terrorist activity is truely random, then this thing does not stand a chance. However, terrorists, like most people, likely follow some sort of pattern and if the signature "tell tale signs" can really be detected then perhaps attacks etc can be predicted.

Re:A step down more like

[Protonk]

People like you freak me out. What do you think science is? How, magically, does meteorology differ from Physics? Do we hold the same opinions about black holes, dark matter and the big bang as we used to? Hell, the term "big bang" was originally pejorative--scientists didn't fit that possiblity into their models. The data changed and so that forced a change in theory. It's how fucking science works. For fuck's sake.

Climate science is no different. What is different is that there are consequences for our actions on earth that matter depending on the outcome of the model. Because there are huge stakes involved, people tend to forms groups at the poles of opinion. You have companies with large stakes in suggesting that climate change is not man made paying for climate research by scientists who feel similarly. You have news organizations and political organizations (who know shit about science) taking the barest of abstracts from a study and runnign with it. You have sceince dumbed down by both sides in order to explain it to voters and policymakers. this sort of thing doesn't happen that much in some branches of science.

Evolutionary biology, genetics, labor economics, sociology, antropology. Those are a short list of disciplines whose conclusions draw people into camps. They also happen to be the same disciplines (not an exclusive list) that people accuse of unscientific practice (and then in doing so, describe the scientific method perfectly, as you have done). That those disciplines and only those disciplines would suffer from a failure to understand the scientific model alone while scores of other disciplines would execute that model perfectly strains credulity.

Data Mining Principles Applied, Eigenbehaviors

[ngr8]

Although there are clearly many random elements, http://reality.media.mit.edu/eigenbehaviors.php/suggest that the "Circular Error Probable" may be improved, the site reads in part:

Eigenbehaviors allow us to identify the structure inherent in daily human behavior with models that can accurately cluster, analyze and predict multimodal data from individuals and groups. We show that it is possible to accurately model many people's lives with just a few parameters - thus allowing accurate prediction of their future behavior from limited observations of their current behavior - as well as to create a similarity metric between individuals and groups that allows accurate identification of group affiliation and behavioral 'style'.

It isn't whether it is an optimal strategy, but whether these tools improve materially the effectiveness of intelligence. "Discovery" AI/Expert systems were finding new materials processes during the 1980s.

Oh ye of little faith. Still, trust in god but lock your car.

Or get change

[EmbeddedJanitor]

Sir, that quarter you have in your pocket is a ramdom number generating device. Spread 'em.

Supposedly one of the better spies (I forget which) always carried a coin in his pocket that he'd flip every few minutes to make random decisions (get to a street corner: turn or go straight? Flip).

Re:Terrible idea

[FleaPlus]

Either they've found some hidden predictor of attacks, or maybe someone needs a course in basic Ramsey theory [wikipedia.org], which deals with conditions under which order (patterns) must occur even in random noise.

Sure, that's why you have test sets to determine if the models the system learns from the data are useful or not. I think it's safe to assume that the scientists working on this are familiar with the basics of learning theory and modeling.

Re:Terrible idea

[Rei]

Really, though, you think they're studying shopping mall attacks? They studied "12,000 terrorist attacks". Bet you didn't know there had been 12,000 terrorist attacks in Iraq in the past four years, did you, let alone 12,000 well documented enough to study? Assuming an average of 20 people killed per attack, that'd mean ~250k people had been killed in well-documented terrorist attacks without the media catching on to the overwhelming majority of it. With that many people being killed by terrorists, who needs insurgents?

Here's a wild notion: they're doing what the US government usually does and calling any insurgent attack a "terrorist attack". Which is why this research is being carried out for the DoD instead of the Department of Homeland Security.

Re:Because....

[timeOday]

Before you rush to judgement, I suggest you play a few rounds of rock, paper, scissors. There's nothing crazy about spending money trying to find patterns in the enemy's behavior, even if he's intentionally trying to be random... that's the entire field of cryptology!

He was eventually nabbed when...

[patio11]

counterintelligence realized most people do not flip when crossing the street.

No, I'm joking. Seriously though, one of the things the military does in Iraq when looking for the foreign jihadis is they watch for wrong turns off main thoroughfares. It is apparently pretty effective at sorting out people who aren't from around here -- if you know Main Street less well than the Americans, you just might be from out of town!