Facebook risk algorithm
The announcement that car insurer Admiral intended to price car insurance for new drivers based on an algorithm which analysed their Facebook posts made for interesting reading recently (thanks to @GrahamTRuddick for the Guardian article).
The unprecedented move gave an interesting insight into how many companies are already using online personal data. The idea was that Admiral Insurance would analyse the Facebook accounts of first-time car owners to look for personality traits that are linked to safe driving. For example, individuals who are identified as conscientious and well-organised will score well.
In contrast, evidence that the Facebook user might be overconfident – such as the use of exclamation marks and the frequent use of “always” or “never” rather than “maybe” – would count against them.
The initiative was called firstcarquote and was officially meant to launch on 1 November but was abruptly pulled when Facebook got wind of the idea and realised that this intrusive use of personal data would rebound on them.
Although the proposed scheme was voluntary, and Admiral was seeking to reassure people that they would only offer discounts rather than price increases, it is clear that they were considering expanding the scheme in the future in order to price individual risk more accurately.
The scheme has been shelved for now but it is tempting to wonder whether Admiral or other insurers will continue to explore the idea.
Criminal justice system
Since the assessment of risk is the cornerstone on which modern offender management is based, I wondered if the same approach could be applied by criminal justice agencies.
There are a number of parallel initiatives already in operation:
The analysis of social media is already embedded in the growing “science” of predictive policing; examples include:
Beware which is being piloted in several US States. It analyzes people’s social media activity, property records, and the records of friends and family to assign suspects a so-called “threat-score.” That “threat-score” is then be used by police to pre-determine if a suspect is going to be dangerous, and to adapt their approach accordingly.
The CIA’s Directorate of Digital Innovation recently revealed it is using analysis of social media big data to predict social unrest overseas which it reckons it can forecast up to five days in advance. It would be surprising if the FBI weren’t working on similar initiatives to predict social upheaval or riots in the USA itself.
There have certainly been a number of studies into the use of social media in the 2011 London riots and other urban disturbances such as earlier this year in Baltimore.
Assessing individual risk
But what about assessing the risk posed by individual offenders by developing a similar approach to that developed by Admira to identify safe and risky drivers.
Once you entertain the idea, it no longer sounds so far-fetched. Police and probation already routinely scrutinise offenders’ Facebook pages when they are investigating crimes or are concerned about harm in situations where repeat victimisation is common — domestic violence or sex offending for instance.
Given the increasing sophistication of many of the big data analytical techniques now, it would seem relatively straightforward for offenders convicted of violent offences to have their Facebook posts searched for any threatening or abusive words or phrases.
The reason that commercial organisations are so devoted to big data (think Tesco clubcard etc.) is that the end-user (customer or offender) supplies the information for free and, once the programme is written, the data can be analysed at very little cost.
Those working in probation know that OGRS and OASys are relatively poor predictors of risk on an individual basis and numerous probation inspections have highlighted the difficulties that many areas are having in doing timely Risk of Serious Harm (assessments).
Perhaps all these systems could be underpinned by a Facebook algorithm which would at least have the advantage of being dynamic and easy to update…