Queensland Police Service (QPS) in Australia to initiate artificial intelligence (AI) trial to determine future risk that some reported subjects engage in domestic violence.
Authors identified as "high risk" (based on previous reports, criminal records or others) will be visited at home by the police before domestic violence escalates and any crime is committed.
Domestic violence: prevention is better, but ...
I say it right away: we need to find better ways to improve the safety of women who are victims of domestic violence. However, using AI technology in this context can have unintended consequences, and this proposed plan raises serious questions about the role of the police in preventing domestic violence incidents.
The approach is based on an algorithm that was developed from existing data. All statistical algorithms must evaluate the risk on the basis of the available data. This in turn means that they are "only" as good as the data supporting them. It is no coincidence that experts criticize data-based risk assessment tools in police activities, pointing out the lack of transparency in the specific types of data analyzed.
Beware of playing "Pre Crime"
As mentioned, the key data acquired most consistently are information on past situations in which the police have been called and data on criminal background. Using this information to train an artificial intelligence algorithm could reinforce existing biases in the criminal justice system. It could create an endless feedback loop, a dead end vicious circle between the police and those citizens who have more contact with the police.
If ethnic discrimination also comes to mind, you are not far from the truth. In Australia, for example, the people who come into contact with the police most are the Aborigines. It is not difficult to imagine that under this new “anti-domestic violence” regime, Aborigines will be visited more by the police.
Police are keen to say that for this pilot project the attributes of ethnicity and geographic location were removed before training the AI model. Despite this, it is still likely that Aborigines will continue to be disproportionately targeted, as they are over-represented in all types of contact with the police.
Is crime risk based a good approach?
The goal of these police strategies based onartificial intelligence it is to prevent or reduce crime, through an assessment of the risk of future crimes. In theory, this means that the police would intervene early to prevent a crime from occurring. Who remembers the aforementioned "Pre crime" of "Minority Report"?
I say that with this approach there are risks that the police may even create crime. A "preventive" program of this type can increase resistance and circumvention mechanisms: it would not surprise me if this led to other crimes.
In summary, the plan proposed by the Australian police presents a clear risk. That of widening the network of criminalization both for the perpetrators and for the victims (who can be mistakenly identified as perpetrators.) How? For example, victims who used violence in self-defense were sometimes arrested in place of the perpetrator.
Could it bring further harm to victims of domestic violence?
The victim's role in such a program is also worrying. Any program that deepens the surveillance of the perpetrators also deepens the surveillance of the victims.
Victims don't always want the police to intervene in their lives. In some cases, this form of proactive policing may seem like an extension of scrutiny rather than help. What happens when the police visit and find that a perpetrator and a high-risk victim are living together again?
The risks are there
Victims may fear that child protection authorities are involved and feel compelled to hide the fact that they are still with the offender. And once a victim has been forced to lie, she may be reluctant to call the police the next time she needs police intervention. In other cases, the perpetrator or victim may decide not to follow the safety advice of the police officers they visit. It is unclear what the police might do in a situation where they ask an attacker to leave or try to rescue a victim, but they refuse.
The mission of any domestic violence intervention should be to restore power to the victims. But we know that interventions don't help all women (or men) equally. Structural inequalities, including race and class, mean that interventions are experienced differently by different people.
