Predictive counter-terrorism

Needless to say, the application of predictive policing to terrorist behaviour is much more challenging than its application to burglaries. Because the geographical clustering does not so readily apply to terrorism, the consequences of the omission of the decision processing dimensions that we have just been discussing become even more pronounced. If it can be used in counter-terrorism, predictive policing is more likely to be deployed indirectly, by targeting antisocial (or, possibly, social) behaviour that might be precursors to terrorist activity rather than terrorist activity itself. For example, racially charged vandalism, gang violence, gun offenses or drug dealing. In this regard, predictive policing in a counter-terrorism context may help to broaden some applications of ILP, which can sometimes be too narrowly focused on terrorism and fail to consider the terrorists’ involvement in crime (McGarrell, Freilich & Cherniak 2007, p. 151). But the limitations of a contagion or diffusion process applied without consideration of decision-maker agency are clear and will need to be addressed.

We can offer a few suggestions based on both orthodox and behavioural economics. These suggestions can be used to guide the interpretation and use of the results of predictive policing algorithms by analysts whose task it is to generate intelligence used to guide the allocation of police resources. We do not think it feasible to build a full predictive ‘profiling’ model.Therefore, rather than add behavioural variables to predictive policing models, we think it better to concentrate on improving the interpretation of the results by consciously considering various aspects of the decision-making processes that are relevant in each case. This includes police, criminals, modellers and the analysts themselves. Even so, the prospects for using predictive policing (on its own) in a terrorism context are limited. Our suggestions are as follows. Remember, though, these are pattern predictions and are not expected to hold in every single case:

  • 1. Hot spots might be better interpreted as reference points. If the activities there have been successful, the offenders are likely in the domain of gains and risk averse. If the activities have been unsuccessful, the offenders are likely in the domain of losses and risk seeking. The first type of offender drifts away, the second type stays.
  • 2. The likelihoods determined by the analyst and embedded into directions for police resource allocations are subject to the analyst’s probability weighting. High likelihoods have been underweighted and low likelihoods overweighted. This happens subconsciously without the analyst being aware of it. Even if the predictive policing algorithms gave absolutely correct probabilities about future offenses, the analyst would apply decision weights to those probabilities and distort them. Further distortion happens all the way down the line, including in the police cruiser doing its rounds on the dark, wet streets.
  • 3. Analysts who are invested in predictive policing will experience gains and losses from its success or failure.This shapes subsequent intelligence reports and adjustments that might be made to the underlying models, at times making the reports (and adjustments) too conservative while at other times too aggressive.
  • 4. Overconfidence in the use of predictive policing may lead to its overuse, beyond the point where its benefits begin to decline.

The key point is that predictive policing is not detached from decision-making processes. The fact that it appears to be is something that should be addressed before it is used more widely. The models, even the most basic ones, are based on choices. Human choices. Human decision-making processes have quirks.These quirks shape even the most basic choices. We value something more just because we have it. We overweight and underweight probabilities. We take more risk to avoid or recoup losses. We protect gains. We compartmentalise our resources into mental accounts. We assess gains and losses against a reference point rather than absolutely. Our reference points might be our status quo, an aspiration or goal or the product of our envy at what someone else has achieved and our desire to do better. We are overconfident, susceptible to information cascades and we can be pulled in various directions by our desires for career advancement and career stability. While analysts and researchers are often attracted to the big decisions and the spectacular events, it is in the everyday that all of these little decision-making quirks make themselves felt.

< Prev   CONTENTS   Source   Next >