Westerholm om AI-verktyg i polisen: ”Tekniken klarar inte alla hudtyper”

The government has presented a legal advisory proposal that gives the police the ability to perform real-time facial recognition from surveillance cameras You have previously been critical of such systems due to their inherent flaws What do you think about this First of all it's good that the police get additional tools in their toolbox to combat crimes I don't have a problem with that I also don't have an issue with the privacy concerns that many others may experience Then I ask in return what is the alternative There is no alternative The police must have good tools This is a well-suited tool However there are significant flaws in this particular type of software that you are working with In the USA where this software has been used for some time it turns out these systems cannot handle all skin tones We Caucasians you and me no problem Asians also no big problem Folk of Semitic and Arabic origin no problem either But the big problem is blacks from sub-Saharan Africa It's the dark skin tone or the dark skin color that causes these programs to have a very high error rate in identification There are also court cases in the US where this type of video depositions are used in criminal trials Then there was a guy he was actually arrested but later during the next court hearing a few days later So he could then prove that no he had not been in that part of Florida and robbed a gold shop She had been the opposite at her mother's house 30 to 40 miles away One could clearly see when looking at these video assignments this was not right at all Today there is significant doubt about using these systems All things depend on how much the police can decide later and especially whether prosecutors can use these at all in their evidence We now have free evidence in Sweden but how reliable is this technique There is as mentioned documented doubt about it I now know that Gunnar Strömmen says this will develop this will surely become good We simply can't have that kind of tool allowing certain types of interventions In a few years these tools will certainly handle it so we allow it already now If it doesn't work now why should we allow it then So that's what I'm doubtful about it's simply that bad These systems can then lead to people being incorrectly identified It's obviously a disaster for the individual but what does it mean for problems on a larger scale It is therefore an uncertain investigative method it is simply not legally secure It can lead to incorrect readings and other issues I think this system should be seen as a tool among others This tool should not be used individually to decide on different types of further actions It's pretty much what I see and if this tool is used individually it's immediately a rejecting one as there are other sources One could say this could lead the police but the police must not proceed with arresting anything This is a human-performed optical identification So it's about that level where you have to do it to get the experiences from the US It is not reassuring in any way rather this is going to happen The American police have had a pretty tough time with this It also becomes an issue at least in the US if the police have conducted A coercive measure in the form of an arrest but on the wrong person and subjecting this person to suffering will become an objec For extensive damage compensation processes Maybe we shouldn't have that The police have other things to worry about than paying damages For wrongful arrest or being in court because a decision has been made On the wrong side Then we shouldn't have this tool Either it depends on how far the police can act on this type of information It will require a decision from another authority To enable the police to use this tool But this can also have consequences when it comes to systems and conclusions What are the consequences for a database if you have this type of error margin This issue will come into play as it is naturally necessary to prosecute the decision It serves as the basis for using this type of investigative means The same applies to secret coercive measures There is a certain probability that the sentencing would result in at least two prison terms To use the secret coercive measures It is the same here that this type of surveillance equipment is allowed to be used If one suspects that one can commit a crime with a certain penalty value The important thing is that this is about when it rings in the system and now we see person X here Should person X be arrested now for having been prese en the person X must be the subject of an arres Or an object of an arrest there must be other sources that confirm this is indeed a person I think it was a high error rate an error rate of nearly 30% In these identifications especially Black people Then it's quite high and then other sources are needed Simply so that one can proceed with an arrest or detention or remand This can be a bit like a safety net It sounds like something is happening in the system now Men if it is based on a touch screen it cannot grasp anyone but it must take other actions such as optical identification This is a mistake that the US has had for a long time How is this still happening I haven't followed this development but it's sufficient For what happened 1.5 to 2 years ago the legal situation was still quite uncertain regarding these systems It's good that the government is moving forward with this type of legislation but before the legislation is needed We must have a framework that makes this tool effective Secure in the process of the police's investigative measures And we are not there yet The police must get the legislation and equipment in place but then need a go-ahead to start using this legislation I am far from being able to give the gift for it has not been presented in a satisfactory manner This is also something that Gunnar Strömmen has joked about I think you were the one who said that technological development is moving forward and it will soon be okay here That is not answer I want I want better precision