 It's announced that this talk is going to be translated into English and Arabic. So if you're on here, you know how to access that translation. Okay. So every time I hear that the government wrote a computer program, I get chills. And especially when they say that they want to use that computer program to decide about people's lives, that's even worse. And Anabiselli has been looking into this question for one and a half years, how this works and how this process works. And she's going to tell us how computers basically contribute to decisions on asylum claims. Please give a warm applause to Anabiselli. Hello. It's nice to see you. I'm glad that you came. I can't really see you anymore because it's suddenly very bright here. But yes, we'll be talking about computers that make decisions on asylum or contribute to decisions on asylum. The relevant ministry is integrated identity management, plausibility, data quality and security aspects. These are systems that the Ministry for Migration and Refugees has been using for one and a half years. And they're supposed to make the entire process on asylum more efficient. We're talking about the following systems here. There's one system for transcribing names. So specifically Arabic names into Latin letters. There's language biometrics, basically dialect analysis to determine where someone is from. And it's about the analysis of mobile data devices. So basically this is analyzing people's mobile phones. We'll also be talking about these things in this order, which is what this training material for BAMF employees, in which it's presenting them. But I also want to tell you a few things about the history of this talk. I've been looking this into this topic for a while, since I was told, first told that the BAMF is using technology to determine people's dialects and where they're from. I'm originally a computer scientist. I turned journalist and from back in university where we were thinking about automated decision-making systems. I remember that of course you can find indicators for which language someone speaks, but that's really, really difficult. And so I started looking into this topic and the more I learned about it, the worse I thought this was. And I tried to get more information about it. Among other things, I submitted a freedom of information request in April 2018. And I asked them to send me the following. Basically the documents you use to train people, to train your own employees to use these assistance systems, these automated systems. Because I know that these systems are often very difficult to understand. They're very complex. People need to understand which algorithms it's based on, how people are supposed to interpret results, which are based on statistics. And I wanted to know how, to what extent are these people even able to do with these things, that was in April, as I mentioned. And then usually for your request, they're supposed to take care of pretty quickly. But I had to wait. I had to wait quite a while. I asked them again, please send them to me. Then it was June, it was July. And at some point in August, I was told that they couldn't, that they could send me the training documents, including the user manuals that they had. So that was really nice. But then they also said, we're going to send you these. But they're only for you. So only you can look at them. So it's only for your own personal desire to have access to this information, but you're not allowed to publish it, duplicate it, or let other people have access to it. And then I was like, well, but that's my job. So that's what we're doing now. And because this is 123 slides and 80 pages of user manuals, I didn't want to go through all of these in 60 minutes. But you can find all of these on fragnerstaat.de and netspolitik.org. Both of these are German online portals. But let's start with the first part. So goals, I think my goals are probably different from the ministry's goals because I want to point out potential problems and create some conscience of what could go wrong. So this is an overview. So these are the different parts. There's picture biometrics. So they compare pictures with existing data that they have, where they essentially try to figure out whether someone had already been registered before, but that's quite an old system. And so it's not actually new. So we won't be talking about that so much. But before that, we're also going to ask ourselves, well, how did we get to all these systems at all? We know that 2015, 2016, there were a lot of people who came to Germany and asked for asylum and were looking for security. And there were these terrible headlines, these terrible pictures about people waiting in front of the ministries. People whose asylum quests were stuck somewhere in a queue. That's also when elections happened. So just because the hot phase of the election campaigning started, they essentially introduced these tools. And also with the backdrop of this bad press reporting, so they essentially said, we can take care of all of these asylum requests with these automated tools. At the time, Markus Richter, who was in charge of IT at the ministry, said that they will contribute to a more secure and faster way of resolving asylum requests for asylum. Where does this idea come from? It comes from a number of management consultancy agencies. The ministry has one of the biggest budgets for employing management consultancies out of all the ministries in Germany. There was a similar person involved who also did something like that in the Ministry of Labor earlier, which was something that was also quite infamous in Germany. So that's how in 2016, McKinsey received between 20 and 30 million euros just for advising this ministry. And they were not the only management consultancy. But they were the ones who essentially pushed this digitization. So I'm going to talk about the first tool now, which is the so-called transliteration assistant, which basically transcribes names from Arabic into Latin. So someone enters their name in Arabic and then it's transcribed into a unified transcription system because people want to avoid a lack of clarity because in transcribing Arabic letters into Latin letters, there's not really a unified system. So that used to be done by translators who all have their own system that they learned. So I think that's actually quite easily comprehensible. And you get a pretty good idea of where it's coming from. And there's pretty good instructions for this here. You have to enter their number. And here in this next step, the person who's requesting asylum has to put in their name, all their data. So then that's saved and checked. And you have to say OK. But that's not everything this tool does. It's not just about transcribing the name, but the system is also supposed to figure out whether this name is plausible. So that means here, you can see this, this is the recommended transliteration. And then it says several countries that have percentages behind them. And it's supposed to say how likely it is that this person is from this particular country to determine whether it's plausible that they come from the country they say they're from to then be able to tell whether, well, for example, in this country, this name is pretty common in this other country, this other name is pretty common. And this whole thing, then maybe they realized this whole idea about percentages is not such a great idea because this is still from training materials. But reality is usually different. So this is something that is, this is a working directive, which basically says the name is relatively rare in this such and such country. And it's also pretty rare in these other countries. And if you look at how the software works, so basically they've adapted this to be a bit more realistic. So if you look at the software, it's using a commercial database that has millions of names in there and the countries that are associated with them, probably by crawling telephone books. And there are 20,000 entries in Arabic. If you consider how many names exist in any given country, that 20,000 is not a lot. But considering that they had to pay a lot for access to the system, which is several thousand euros, this is also quite different from what Markus Richter, who is now vice chief of this ministry and who was in charge of RT at the time, he said, we basically scanned all telephone books in the entire Arabic-speaking world. And I think I would say that there should be more entries in there than 20,000 entries per country if they had actually done that. And I would also argue that you can't necessarily tell based on a telephone book whether someone is telling the truth about their country of origin. The ministry, Banff, has also realized this now. There was recently a parliamentary inquiry by one of the parliamentary groups where the Ministry of the Interior basically responded that the success rate varies. And they're saying, well, we have a success rate of maybe 85% to 90% for countries like Syria, but the countries of the Maghreb, like Morocco, we have maybe a 35% success rate. You could say they have 65% failure rate, where they get complete bullshit. And that's partially because they obviously also are other influences, such as, for example, Italian or French names. So if you were looking at Germany, if your name is hoarse, everything is fine. But once you're called Nancy, you're going to get into trouble. So that sounds funny, but what you have to imagine is that this is a person who's requesting an asylum and who's part of this process and who's giving people their name. And this name may not be the most common name for the region they're from. But then the person who they work with essentially is told, well, this name is not typical for the country that they're from. So basically, you're generating mistrust. And the program is giving you some kind of piece of advice and is basically implying this person might be lying. This becomes slightly more drastic if we're looking at the linguistic biometric parts. So these biometrics are supposed to detect dialects and languages. And the way this works is that someone essentially speaks something into a telephone speaker for around two minutes, which BAMF is advertising as a very slim front end in other places. And the ministry actually has won a prize for this tool for the best e-government tool in German bureaucracy. You may believe that, you may not believe that, but they're very proud of the system. And this slim front end is pretty simple. So basically, you dial 72099, then you press a number, you put in a particular number, and then someone speaks for two minutes. So basically, someone is sitting there and the person has pictures in front of them. And the pictures have typical scenes. One person I talked to said that these pictures usually have, maybe there was a house and there was a garden. There was a mother with children and the person was supposed to describe the scene in their native tongue. And in an ideal case, they were supposed to talk for two minutes. And then you press another button and you basically get a report via email. This system is supposed to be able to detect and distinguish different dialects and languages. And in particular, the ministry is interested in different dialects of Arabic. So you can see this on this slide. There are different regions where Arabic dialects differ. There are five different groups of dialects that the BAMF distinguishes. Of course, they are way more. But so this is supposed to speed up this whole process and to make decisions on asylum based on a unified, have it happened based on a unified basis. Actually, the system that they bought is from, could actually technically do even more languages, including Bulgarian. It can do American English, Hindi, Farsi. But so the decision makers who are part of the asylum process were essentially told to only use it for people who speak Arabic and who are requesting asylum, which is also what the ministry says. But it didn't actually take us that long to find a few people where this wasn't the case. So for example, I found someone who is actually a Kurd, is from the autonomous Kurdish region in Iraq and speaks Kurdish, and then speaks a small Kurdish dialect. And this Kurdish dialect just doesn't appear in this list. But this person was asked to use the system nevertheless. The system would never have been able to detect whether he actually speaks the language which he says he speaks. And in that case, it basically said, oh, you speak Turkish. But maybe you speak Hebrew, entirely sure. And so this person was sitting there and was, what is that? He really didn't know. And so that's extreme. They're really obvious mistakes that happen as part of this process because the ministry's employees don't know what they can use the system for and what they can't use the system for. So there's a complete neglect of the existing guidelines for using these systems. We also had independent verification of this when the Ministry of the Interior responded to this parliamentary inquiry where they were told which countries people come from, who the system was used on, that included people from Kenya, that included people from Bangladesh, people from Venezuela. And so all of these are countries where people speak Arabic languages usually and where people are usually Arabic native speakers. So it's completely inappropriate. And then this is where it gets really interesting. So this is one of those analysis pieces. This is from the training materials. So on top, we can see the probability for different dialects. For example, Gulf Arabic was 76%. And then there are different Arabic dialects with lower dialects and then some unknown, other languages and unknown languages. And then it says how long the recording was. And then some technical details. And so for example, there's the interference distance, SNR. So that's basically an indicator of how good the recording was because usually you want this number to be large. So you don't have things that are happening in the background such as a truck passing by, which may have been louder than what the person was speaking into the recording. But what is BAMF is basically, the ministry is basically saying, it's basically saying these technical details are not relevant. So you don't have to look at this at all. One thing that you also look for this without any success, there's no information here regarding at what point such a result has any significance and actually meaningfully tells you something. Like saying that for example, only results from maybe 90% upwards are something that you can actually trust that are meaningful. So basically people are just by themselves, they get this piece of paper that says some kind of language that's on top of this list with the highest percentage. And then they might think, well, maybe when you ask yourself, what is this supposed to be? Another thing that I think is quite concerning is there's no quality control here whatsoever. The ministry itself says that they have a failure rate of 15%. So every fifth to sixth person that requests asylum, this system is wrong. And the ministry knows that the system is wrong, but they're basically not giving any kind of indication of how people are supposed to deal with that. They also still don't have any scientific advice or accompaniment to this system, despite them having said for more than a year that they want that for a long time, they've been trying to sell us on this idea that they worked with UPenn in the United States. So we asked UPenn and it turns out, well, they bought a language package for 3000 euros and that was essentially the scientific operation that with UPenn. All of that aside, language isn't static. It's not straight. Language changes, it might change throughout your life. And we have a video here that we might not be able to play, but I think we can play the audio. So this is a video from the German parliament, from the Bundestag. Oh, so when I came as, became to the office of being the head of my town. So there's, the good thing is, if you know this accent and dialect of German, which I actually understand due to my grandparents, that he's speaking right now that comes in quite handy for when you want to talk to the locals around. And he now speaks in Bavarian dialect. He talks about the colleagues that's standing next to him being a politician where you can figure, like she's basically showing different dialects of German and showing, it's quite revealing and telling how your own develop, like how your own language develops throughout your own history and your own life. And you can't really take this as a serious and significant system of telling and determining where people are coming from and stem from. So the statistic marker should not be there to categorize people. You could say now, and the Banff does this all the time, that this is obviously only one marker amongst a large pool of data that's deciding these things. I met another person where this person had in a refugee ID from Palestine. She had her brother certificate, her marriage documents from her parents from Syria. And she gave all of these to Banff and they did the language analysis. And then the language analysis determined that she's from Egypt. And so the Banff then said, well, if you're seeking asylum, you're not getting it if you're from Egypt. There was no other marker that showed that she could've disproved from where she was coming from and that she was being untruthful about this. So you can see these cases where the Banff, I mean, they can claim as much as they want to that those are, that they're taking into account all sorts of things. And this is not the only factor that's gonna be decisive. But in actuality, this all happens. So a decision maker, in this case, they're supposed to decide on three cases per day. So okay, you can listen to this, there's an app professional listening to it, like where you have maybe two hours, you have to really deal with their history and then you still have to make a decision. And then you're in the position where the computer said, but the computer said it gave me this lovely thing with all the percentages with a point after the comma even. And this is the problem with all these fucking systems is that it gives people security that's not actually there. And so this seeming security of making decisions on other people's lives, they can't, which they can't really make an informed decision based upon just the system. The problem really is even bigger. The decisions that are made based upon these results are that the person lied, the person seeking asylum didn't just try to seek asylum, but it is denied on the course of the fact that it lied. And it only has one week to appeal this decision and interpret this official administrative document and then appeal it and deny that they accept this. And I think about how long you take sometimes to understand bureaucratic German paperwork that you're getting. And if you are not familiar with this German bureaucratic system, it's almost impossible if you don't get enough support on how to deal with it. So this is quite a horrific way of keeping people away and turning them away. But I will tell you more about this later on. Well, let's just look how much this cost. This is supposed to be costing 2.1 million until 2019. This is the cheapest system that they're using of all the three. So remember 3.1 million. 2.1 for speech analysis. And then the one that's covered most in the media, analysis of mobile data devices, mainly cell phones. This system came with a law in 2016 and that was the law to better the process, to better the enforcement process of deporting people back to their home country. So this was a way for them to prove people's identity and to be honest, this was basically enabling the process of deportation. So if a passing comes to Germany and gets registered and does not have their passport with them, they can be asked to hand over their cell phone and then this cell phone is being connected to the system and all the data is taken from it and there's a report that's been done and this report, the interesting thing about the report, so if your actual country of origin cannot be proven in another way, this is something that they can do. If you look at this more closely, this is quite hilarious, if you think about it. This is quite an extreme overstepping and invading people's privacy to get all their data from their cell phone and the mild measurement that they are actually, that you can do, that the only thing that you can do to prevent this from happening is showing your passport. This is also in the training program. The actual... Hearing that people get, this decision of whether or not your data is taken from you from your cell phone is made before you even get any sort of hearing. So later on this has to then be okayed still. Like so the data that's taken from the cell phone is a different person than the person that then does the hearing afterwards and the person that does the hearing afterwards then has to decide if they want to read the report from the cell phone and then that has to be okayed from somebody from like a judicial department and you can take a look at what the system actually looks like, that the BAMF didn't really want to reveal and make transparent how they're using it and then all of a sudden this was part of some of their press releases and press images and so this includes the system, it's a system by MSAB and I actually have a video of this that probably is not gonna work. Unfortunately, we're gonna try this anyways. Nope. All right, so I'm just gonna quickly summarize. MSAB was a company that does mobile forensics and mainly does it for police, military and secret services and they're actually advertising with exactly this. So basically this is technology that's not something that you would necessarily want or suspect in asylum seekers but more within police investigations and so there's another company in Ceilid in Israel that likes to sell it to other quite questionable military dictatorships. So MSAB has up to this point a quite a clean slate. So the system does the following. It's quite simple. This video shows it really well, how it's easily dealt with and how it's so user friendly. So there's a couple of buttons and you register, you log in and then you enter your name and your password. I talked to the IT if you've forgotten it and then you have to confirm that you're familiar with the program that obviously only authors' personnel is allowed to use this and then you just have to click continue and then you click on the big button, take out the data, take out the data from the cell phone that you still register where this cell phone was obtained, when it was obtained and then they take a picture of it and they take the picture of the cell phone because if somebody scratches it, the person owning the phone obviously has to be reimbursed. So it depends on what kind of cell phone you have. BAMF can access and get different amount of data. Some of them, they can get more information and so they give their employees different types of devices that exist. It's a little strange. This is my most favorite slide of the whole thing and the whole training program. So there's cell phones. Cell phones don't always have to look alike so if you realize so that you can pick them out, I'll give you a few devices. So there's like a slide phone and then there's like a phone in the pistol and a gun. So if there's a refugee that comes with a gun, potentially the BAMF employee could potentially could then ask, is this your cell phone maybe? Could I quickly take a look at this? But yeah, so this is supposed to train people in identifying mobile devices. All right, so this is a Photoshop slide. It's not a real... At this time now you can actually get selfie guns but that's a different story. Also this explains that Android is the most common and that just exists as a smartphone. It could also be in a car, a radio system and then they also have to take a picture of that and then very important. The approval has to be done by the person who's seeking refugees. So that's why they're kind of legally can claim that the person owning the phone gave their approval on free will and they don't actually break into phones. And this is one of those moments where you usually get very skeptical and you kind of ask yourself how freely is this approval given? And then well like they have to unlock their phone and they have to help with this asylum seeking process. But if the person seeking asylum is not entering the password and this request for asylum can be regarded as taken back, retracted. And so basically this is how free willed this approval is actually given. Yeah, so Apple devices are actually easier in taking their data from it. So what happens with all this data? What is BAMF actually looking at? Just about that, the call history, the contact details, messages where they are going and what languages they're written in, general information on the device and usage of apps. They claim that they're not analyzing any content but if you are checking what language is used for writing messages that is quite close to content analysis and then this is summarized and this report can look something like this. There's nice little pie diagrams that show how what calls were dialed to the most and where calls were stemming from country-wise and how long you were on the phone with them and where did you send messages to and what languages were used to send messages. So this is done for messages that are sent and as soon as Arabic as a language is detected it's categorized in what kind of dialect is being used. So we don't just have this on a spoken level, we also obviously have this in a written language level. So in the Arabic there's different dialects of chat lingo. So it looks a bit something like this so you have Arabic symbols but if you try to type them into a regular Latin keyboard you have to sometimes substitute symbols with letters with numbers. So depending on the country and dialect this can differ and vary. So according to Banff this can detect 130 different dialects but we're not really sure how good of a job they're actually doing with this. So the other thing that's obviously analyzed is the contact details. So which countries did you enter for contacts? What websites were you accessing and what location data is your cell phone giving? So what we don't know is that we don't even know which kind of locations they're using. So we know they use geolocation data from photos, they use data that's collected by apps but apparently they can't analyze all this information, they can't access all this data and the system basically creates a picture like this and then they're asking well why are there no errors between them and then they're saying well we took those out but if you look at this you can maybe imagine that someone didn't go to Greece to Germany and then to Greece and then back to Greece again. And so these are account names which just tells you so you can analyze a bit further what this person does because you have their email account, you have their Facebook account, you may have their Fiber ID, I'm not sure, Viber ID, that's another messenger and if you have access to all of these apps and so this is basically the system where the Banff is giving its employees the highest number of support for interpreting these things. It's basically saying well if you have a feature phone like one of these old Berk phones then you can't look at that much information, you also have to look at how often this person seems to use this device, how long it may have belonged to them and what kind of information can you get and then obviously if there's not a lot of information on there then they haven't been using it that long or they may not be using it that much but if certain types of data do not match other types of information then maybe they have been using different types of SIM cards and maybe many different people have been using this which may not be that uncommon with people who fled and then there are results where they're saying that applications used related to identity is there are many different accounts that people essentially use and so basically they're assuming that if you have a lot of different apps then they're assuming that this is very meaningful I mean a very meaningful type of information and so for example they're saying that booking.com data is better than dating apps but it's not clear to me why booking.com is supposed to be better somehow and we can see this system can look at a lot of different things by this company MSAB and it can actually break into smartphones MSAB is the only serious competitor for sell it right and they could open phones up. Allegedly they don't do it because they're not allowed to but the system could also simply analyze and save everything that's saved on a smartphone and that's something that ministry has not only wanted to do since recently but in November 2017 the BAMF chief Yota Court at the time said well we have this really nice system but we'd really like to access photos as well because then we can see much more about what people actually did and I really think that's just a matter of time and the Ministry of Interior in response to this recent inquiry also said that they're looking into potential ways of expanding this so it's not just we're not going to stick to just this where they just have statistics and say where people may have been but they're going to go much, much deeper into people's privacy and they're going to interfere with it much more in a much deeper way and then we need to ask ourselves how deeply does this violate people's basic rights and how extreme do we think this would be if this happened to someone in Germany who basically just goes to an agency and puts in some kind of bureaucratic request and then they basically have to give them your phone and it has to be analyzed thoroughly and I think that's how you need to realize this is how people are being dealt with in Germany who basically have no political lobby for themselves who nobody speaks up for and only if we think about how we would react to how we would react to this happening to people who have lived here for a very long time and I don't think we should accept this and we shouldn't even start using this because this is basically a test balloon for doing further things in this direction. Round of applause for that. Okay, then the perspective. So analysis. So we set 3.1 million for transit iteration, 3.2.14 language biometrics. So what do you think for smartphone analysis? Prospectively 11.2 million euros until the end of 2019 and this goes towards paying for hardware because you need to analyze this somehow then for software that analyzes and then for tech support and because there's an increasing number of device types obviously the system has to be updated constantly as new smartphones are being created and the data needs to be read out and needs to be analyzed. And if you look at this just from so if you just ignore all these basic rights concerns so if you ask yourself how sensible this is, how much sense does that make? Does it make the bump vice chief Markus Richter said we have analyzed 27,000 smartphones? That's the country, the number that they were using to brag about this about a week ago. Around a week ago, the Ministry of the Interior said, well we've actually analyzed 9,710 phones. Even this number is something that you need to look at again because just taking the data off these phones and then analyzing these phones means that the system generated one of these reports but it doesn't mean that someone actually saw this report. So how many times has someone seen these reports? And the Ministry of the Interior has said that this was in 2,845 cases. So the number is going further and further down. So we started with 27,000 phones and now we're down to 2,845. And then you ask yourself, where did this happen? So I did a nice graphic here where you can see that the different parts of BAMF use this in very different ways. So right in the center that's Giesen they seem to like this quite a lot. And then also if you compare this to the number of people who arrive at these different parts of the Ministry then that's still not quite disproportionate. So 2,845 reports were read. And how many cases was this usable? 35%. So in 65% of cases they just didn't find any useful information, didn't get any useful information from these reports. Maybe because the person only had the phone for a short time, maybe because it's been shared with other people. Maybe because the phone broke because you couldn't actually get the data from a certain phone. They also said that they have problems with proprietary connectors. So 35% is information that they could have used but what the Ministry is interested in is in how many cases this actually gives you indicators that people, there might be contradictions and what people say about themselves. 2%. So 2%, then ask yourself 2% of 2,845. That's less than 60 cases. So you're saying they've spent, and look at this again, they're spending more than 11 million euros on this until the end of 2019. This is a number that you could use, amount of money you could use to actually hire a lot of people who know about people and you could talk to these people and you should also ask yourself, do you maybe really wanna spend that amount of money to keep people out of Germany or do you maybe just wanna let people come into Germany and realize you're a fucking rich country that is spending an insane amount of money on stupid software just to keep people out. Another round of applause. Yeah, so I'm just gonna abuse last year's motto again, which is do something, or which was do something. All of this sounds like it's there and as if it was something that we have to live with but it isn't. I would like to ask all of you to resist the use of these systems that maybe if you're affected by it, but also if you can support people who are affected by these systems. Because I already mentioned this earlier, I met some of these people who had problems with the system and this whole shitty situation has one good aspect. It doesn't have the status of an official report because previously there were people who were able to issue a specific report and who essentially then issued an expert opinion and advisory opinion that could be used in front of a court. And this stuff with the software, if you can say this software essentially has a failure rate of 15%, then you can't use that in front of court because if people actually run into problems with their asylum process, you can actually take this to court and quite successfully and one of the people I talked to in their case, it actually worked already because they could say in front of court, this is the only indicator that they may not have told the truth and that's completely unreliable. So the only thing you need to pay attention to is that this happens quickly because this person has one week to contest this decision and one week to file a suit and I think that's where if you're working with refugees and if you're supporting refugees or can someone talk to them, then you really need to tell them that they can defend themselves against this and you do not have to accept this. Another thing that you can do is that you can inform yourself and you can inform me. So at some point in my description, there's talk description, there's probably contact information. So if you have any further information about these beautiful systems, I'm gonna keep publishing these and with those, with that, I'm gonna thank you for your attention. Then thank you Anna for the talk. I would ask you to ask me to put the microphone in front of you. So if you have questions, then please come get to the mics and ask your questions and signal Angel first. Hello. Hello. Are there indications that the connection to the smartphones installs any additional software on the refugees phones? No. Sorry, no. I'm not aware of any cases where that happens or happened. I'm also still looking where this happened in general because it's quite difficult to find these people where this analysis happened, where it didn't just fail on a technical level. I already found three people where it, there was just a technical failure, but I would like to talk to people who can tell me about how it looks like when it worked. Two short questions. Is there any lawsuit going on against reading data from the mobile phones or are the refugees simply too intimidated to do anything against it? Question number two. Did you try to... Did you... Did you try to play the videos that you've shown for further analysis? Well, that's an eternal mission that I have. I really want to... I've been asking BAMF for a year, at least a year. Please, please show me these systems and my secret goal. Now that they're definitely not going to let me in anyway, is that I want to prove that Jose Hofer doesn't belong in Germany. He's the leading member of the CSU, the Bavarian Conservative Party, Google them. But the last excuse that I've gotten from them why I can't enter these rooms is that they're too small. So, unfortunately, that's not a joke, but I will keep trying. Maybe at some point they're going to let me in. But the first question, lawsuits, if there are people who are interested in suing about the system and going to court, get in touch, there might be plans about this. As part of essentially suing against decisions on asylum requests, that's already happening. But I think you need to attack this on a much more fundamental level if there are people who are willing to go through with this. During the examination of the mobile phone data, there is also data analyzed about the location data to identify the origin of the mobile phones. And usually during refuge or during escaping from countries, it's usually a very, very hectic and incoherent development. So it's actually not surprising that a mobile phone is not from the country of origin. That's a good question. I think one thing, all of these analyses and these reports are supposed to happen before the hearing happens. So once you register, they're supposed to take this picture of you. They look at your passport if you have one. If not, then they're looking going to go into this whole chain. And then in your hearing, you're supposed to have the chance to make your own statement regarding this. Well, smartphone analysis is something that they don't seem to have used a lot in these old processes, but there were processes in cases where they essentially used language and linguistic analysis, including people who may have been already in Germany since 2015, 2016, where their cases were just pushed back further and further and where they got language analysis at some point, where one of these traditional ones, where someone has to speak to a translator for half an hour, and then someone has to look at this. And that's incredibly expensive. And then at some point, the ministry essentially gave out an order where they said, well, please just look at these cases and consider whether we can maybe replace these technical analyses. And that's where they did these analyses after the hearing. And I found three people who actually didn't get an opportunity to respond to these reports. That's supposed to happen, but I think especially in all these old files and with all these old cases, that doesn't necessarily happen. Your analysis with regard to the efficiency of the extraction of data from the smartphones, it shows really clearly that it's really bad. So shouldn't the BAMF realize by itself that this is really nonsense? Don't they look at the data or the results? Well, the ministry published just different numbers. The number that they're given is that they're saying, well, we analyzed 27,000 phones, but that they're not telling you that that's only the cases where they plugged a cable into a phone, and not cases where they got any useful information from it that they don't tell you that. In our experience, really, you get information about these systems and they're planning a report. And then maybe a day before there's a report in another kind of medium where the ministry says how well these systems are working and how successful they are. And I think that it really is the thing, it's easy to ignore how useless these systems are because you're so focused on how helpful they are because, for example, they said just before the elections in 2017, oh, we have these systems and we're really on the cutting edge of e-government in Germany and maybe even the entire country, in the entire EU, we want to provide the system to other countries in the EU and we essentially want to be on the cutting edge of digitalization of bureaucracy in Europe. I would recommend all of you go on the ministry's website. They have great, amazing advertisement videos where they also show how they're trying to implement scrum and they have a scrum master and they're sitting in a room with graffiti on the walls and they have sofas in the room and I think that's where you, maybe how you can tell that this is the problem, right? They want to be cool, they want to be digital despite being part of a bureaucracy and we can't really get away from that anymore. If I understand correctly, you said that the tools and the system were used based on consulting by external companies. Do you think they were consulted incorrectly or do you think they would work better if they would be used more correctly and unlike? Well, what does it mean because of management consulting agencies? We know that in this time, there were a lot of management consultancies that were employed by a bump that have recommended more or less questionable things. They, for example, also recommended that you need to separate refugees into different classes, including people who have a good perspective and a good chance of remaining here and those probably gonna be deported anyway because they come from a country where you have a very low rate of people who deserve protection. So this was really about efficiency and this was about digitization and digitization and creating paper-free offices but it was also about the system and the system, whether you can build systems that can work. Well, it's not a question you should ask yourself. I think there's an ethical question here. We're all complained because there's this stupid Shufa credit score with algorithms which we don't understand, which uses weird data points about ourselves which we don't know about and that's why we don't get our smartphone contract. So that's software where there are algorithms that you don't understand they're in transparent databases and in these cases people don't get asylum because of that. And I think if we're saying that credit stores in Germany can't be made in such an in such an in transparent way then we also shouldn't be using the systems in the case of asylum at all. We really need to ask ourselves that question. I'll pause again. Then microphone number two, please. Again to the assignment of languages or assignment of dialects. There was a TV report that people were considering using the system for the police and do you know if this technology is also used by other German institutions? No, I don't have any information but I don't believe that anyone else is using it. I think the MDR, a German public TV station also wrote that. I think what you rather have in those cases is software that tries to identify speakers. So if someone says something you're trying to identify whether someone who made a threat on the phone is this person but they don't really do dialect analysis in that way because I think for most other German bureaucratic institutions it's not that interesting. Where it's interesting is obviously are obviously other migration agencies in other European countries and the BAMF is really trying to establish this as something that's on the cutting edge of the model. Is it, in your opinion, is it wise to delete all the data before you enter Germany? Or is that rather bad or makes it suspicious? That's a good question, I can't answer that. Well, I really can't answer that question. I don't know. I really don't know. The ministry is aware that there might be, that there are fake smartphones that circulate where someone put a plausible story onto that phone and obviously the market adjusts. You know there are systems that these systems analyze certain things that the market adjusts and you can basically create smartphones that have the information that these systems want to see. So you could say that there might be people who might quote unquote lie but so you're not gonna find those people, might not gonna find those people anyway. So in cases where there were contradictions maybe someone just found a phone and took it with them and then there was already data on there. But of course it's easy to fool the system and the ministry knows that. I wonder if those 200,000 euros per success are not also something that the financial court of Germany should look into if you disregard the ethical issues. What is your opinion? Well I think the Bundesrechnungshof, the federal court of auditors is supposed to look at the ministry anyway but if someone has that report from this year I would love to look at that. I mean I think if you look at this from a financial point of view you can definitely ask that question as well and should you can also ask yourself how many people could you employ? How many people could you enable to live in Germany using this money? I really think this is a complete waste of money. So if you have reports please get in touch with Anna. I was in approximately two years, two months ago with a youth at the Banff and together with a youthful refugee and the mobile phone data was extracted and I was wondering where the data was stored and how secure the data is being stored. You can ask the Banff ten times and you're gonna get the same answer every single time and it's in a secure data safe. So no. So no I don't know. I don't know. The ministry says this. So essentially unlocking this phone has to happen through a lawyer and the people who work in the Banff the employees themselves can't access this data that's stored in that database but how secure it is in a physical way. I don't know. And the announcement for the talk there was a sentence that the Banff is also doing a project with blockchain. Is that too much to ask but do you also want to explain this part? Oh yeah. So there's gonna be more about this in January. Banff is doing things with the blockchain as well and it's called blockchain at Banff. This is not a translation. It is called blockchain at Banff. I don't want to say too much about this. It's yeah, of course it's cool to do things with blockchain now. So the Banff is also doing things with blockchain and they're trying to figure out how to process asylum requests on the blockchain. Apparently there was a proof of concept and allegedly there's currently a pilot project happening in Dresden and so not far from where we're at Congress and what their goal is to essentially enable other ministries to also access the state of affairs in different asylum cases more quickly and more immediately. So basically once something happens you're gonna put it on the blockchain and then other ministries can also access that but there are still a lot of questions. There's also a white paper that's published on the Banff website because at some point they realize that this whole blockchain thing is kind of stupid because it's quite difficult to get them back off the blockchain again although they're supposed to delete data after 10 years and then you can do something pseudonymously and maybe you can delete things on the blockchain after all so it's gonna continue to be exciting. I'm not that afraid of this project because it's something with blockchain and it's gonna be over at some point. Gonna get over that. But a colleague of mine and I are looking at this right now. Let's say that. As long as the blockchain does not exist are there any lawful guidelines how long the data has to be stored? Yeah, they exist. I don't know them by heart. Basic data can be stored for 10 years. Other data that is more sensible such as health related data. A lot of information that is also collected as part of asylum requests such as, for example, what your vaccination status is. Stuff like that. I think health data can be stored for one year but don't pin me down on that. I'm not entirely certain. There was a regulation on this but yeah, I think that you can read up on this in a lot. I don't know this off the top of my head. Hello, I'd like to ask I know from my work that it's very common that there are mistakes when it comes to first names. So, and sometimes there is just what sometimes the name, the first name is just the name of a city and then it's an obvious mistake. But once it has been processed by the BAMF, there is only with a lot of formal proceedings and a lot of hearings possible to change things afterwards. And I wonder if this program has an error rate of such a high error rate. How can you deal with this? Well, it has a failure rate of 15% because this name transcription program is only supposed to help you transcribe the name that a person is inputting themselves. So, the person is supposed to input it themselves unless they can't write, then a translator does that. And so, it's only supposed to transcribe those names in a unified way. So, that seems quite sensible actually in the first instance. But I think there's another problem there because you have different parts of the bureaucratic system that are responsible for things like this because you might actually end up in a situation where the police registers you and then I think correcting data and cleaning data after someone has been registered there that's quite difficult. And for the so-called central registry for foreigners, the BAMF is technically responsible for that and they're responsible for keeping this data clean. But basically they're essentially putting this responsibility away from themselves and they're essentially saying whoever enters that data is responsible for ensuring that this information is correct. That's not a problem that's going to go away anytime soon. Thanks to Anna for this great talk and please give her a warm round of applause. Applause in the audience.