foreign good afternoon everyone thank you very much for returning to the applied AI stage just to recap for those of you who are new to the room my name is Gary Abella and I'm the pleasure of being your host for the rest of the afternoon I come from the II background and have a small bespoke agency here that helps companies take AI Research into production and developing and deploying AI Solutions so I've had a thrilling time this morning and no doubt we're going to have another great few sessions this afternoon so now that our tummies are full and our minds are re-energized I'm really going to jump straight back into our next talk and a vital vitally important session uh that's often really overlooked in that whole data science life cycle which is the importance of ethical and compliant data collection and now I've seen guest to be able to talk that through is co-founder and CEO of brighter AI Mr Marian Gläser please welcome me on stage thank you hello everyone I hope everyone
can hear me and the food has settled as much that also the ears are open so you have now I would say 30 seconds to leave the room if you want to because I will talk about the gdpr I will talk about the appi the Japanese data privacy regulation I will talk about the CCPA the California consumer Privacy Act and a couple of others I will make it fun so if you like had to attempt now to to leave the room or I I have the um the ambition to make it fun and I will have the ambition to make it very practical on the use cases that we are working on and not give a commercial presentation but an Insight a bit of how we reflect the industry how data privacy developed within the last couple of years worldwide the compliancy developed well um in the in the different regions and how to actually bring this together with AI with machine learning because this is where two and I don't have a slide for
this because I think that everyone in the room is aware of it most of the AI development relays on a lot of data on dictator and if you have a privacy regulation you roll your eyes and you say okay this is limiting us and if we are in Europe or in more and more regions in the world you need to find solutions that are within privacy for our values but also allow machine learning so without further Ado um I'm Marian Gläser co-founder and CEO from brighter AI privacy tech company here from Berlin about 60 people I'm also speaker for the federal Association of startups for AI as well as privacy speaker for the AI Association as well as a couple of others background is in automotive as well as in Consulting that's enough about me and a bit about the company um just so that you that you know what's what's the background of what we're doing founded in 2017-19 uh European's hottest startup by Nvidia with a data privacy topic and data privacy was
not really that hot at that time I'm not sure if it ever is but German Innovation award most relevant cyber security company in Europe and quite happy about the use cases that we're working on and the things that we do so first up is privacy a thing so is this really is this really something to talk about besides uh regulation I mean um in the industrial revolution in the French Revolution people are fighting for their rights and they are fighting for freedom and within rgbs we click them away with every sign up and so often this is not really a thing apparently or sometimes or is it um I will hopefully answer this question but I first want to go a little bit into the into the history and how old at least the enforcement of privacy and how it's being seen in the past evolved and how do we get to the current day it's actually quite long ago already in in 1361 that there was the first Privacy Law of not stalking basically
which was implemented in the law then a couple of hundred years later everyone knows it you're not allowed to open letters and another Milestone that is today very much accepted another one in 1888 there were first cameras introduced more than cameras a huge spark in the media what does this mean to privacy of the sun you can be captured everywhere and in 1948 the UN acknowledged that it's a fundamental human right and there was the rise of privacy it took a couple of hundreds of years and then there's the second part in Facebook 2010 they put the default of data from private to public and Mark Zuckerberg said privacy is no longer a society Norm in 2011 smartphones were released with the first tracking systems the New York Times declares the age of Big Data and in 2012 girls around me an app was forced to shut down which would tell you where are girls around you so it got a little bit absurd um within the last I would say 15 years and then
of course a couple of other things happened and then the gdpr came so this is the world how we how we look at it today or no it's 2018. uh this is how it looked in 2018 gddpr got in place specifically looking at data breaches at how to basically tackle daycare data leaks how to make sure principles such as data minimization are kept and so on and now we're in this in this complex world where we see privacy for every one of us right nobody wants that that actually kind of data gets out of finding people specifically in public having being exposed so it is not just a fundamental right but it's actually for every one of us something that is quite valuable and on the other hand you have the regulation gdprs to restrictive it's limiting Europe in the in the way going forward the gbr in a lot of ways really needs to have a do-over but it's there and it does support the first part so how to how to actually find
find a way to to to go around this and bring bring privacy and data together and I want to I want to give one more example because often when we think about privacy we think about like especially privacy in big Tech we think about accepting the the agbs and again French Revolution fighting for rights it's a little bit different emotionally but there are other examples where it does matter a lot and it's no surprise picking this example because this is the area we're working in as well it's one of the most critical one but also the one of the most exciting one right you have cameras all around the world you use them for safety but you can use them for Innovation you can use them for hundreds and thousands of different business application from cost saving for um within public transport for autonomous driving which are relying on camera data for new insights and store Concepts such as Amazon go so really really powerful medium one of the richest sensors in fact cameras and
cheap as well which is great and with that you can do a set quite a lot of things people flow iron analytics fi like understanding the accessories of the person the emotion of the person you can really in-depth understand this person who he's with and when he's with and what he's doing how how he's feeling towards everything around him or her this case him and you could say here also here well he's in a crowd right so he's pretty much autonomous so identity should be fine well in this age of of basically Ai and analytics there are technologies that are also evolving which can be used for good but they can also be used for privacy violation and this is when then any kind of breaches of big data sets actually really makes sense to protect them somehow this person for example buy a single shot of his face and a tour called Pim eyes you can find this person everywhere online his personal profile any kind of Linked In His meetups that he went
to his private details when he's on a photo with someone else that he even didn't content to and then it becomes actually a little bit scary because your face is an access code to your identity however you cannot change your face and that's similar for a lot of other personal identifiable things that that you have and I hope I I with this example answered at least the question why privacy really matters and it's something sometimes a little bit different than just accepting agps do you acknowledge this as well there have been almost 3 billion euros in fines against data privacy regulation and you remember this slide we actually spoiled it a little bit this is how it looks today so when there were voices in in 2018 that privacy would be privacy would be a Flink it it's not it became a worldwide aspect so let's go on to a journey let's go on to a journey with an AI company that's going on a journey with an AI automotive company because lots of corporates
they're not just basically in one nation doing their development doing their research doing their development of in this case self-driving cars but um doing this worldwide right and this is for every AI deep learning engineer This is highly simplified labeling training test run validation so if there are any questions that this flow is incorrect you misunderstood the presentation um and so um we're going on a journey with a tier one Automotive so classical company that basically supports the oam with bringing certain features to the market here here basically self-driving car assistant functions address functions and they're doing data collection in Europe they're doing labeling in Brazil uh they are doing the model training with another company in the Silicon Valley where else where where the big manufacturer the big chip manufacturers are sitting and they're doing also of course model validation in Japan they need to test that it actually worldwide works so they have a bit of a journey of data that travels through the world and I don't want to to bore you
with basically all these different um regulations in detail but just giving a gist of how privacy now complicates this entire flow so in in Europe in in the gdpr actually it's very consent driven so how do you get a consent when you go around with a mapping vehicle from a person that you're capturing that's practically impossible so you need to find different legal or technical ways that's a that's a big one another big one is of course the the penalties up to four percent of the annual turnover as a penalty for for companies breaching breaching um the gdpr so for this company even to start the data collection that's a that's a headache you start with but then the next part very expensive labeling you find a good company in Brazil and Brazil also has a quite strict privacy regulation um introduced I think in 2020 and it's also for them you're processing personal identifiable information it actually in their regulation also doesn't matter if it's Brazilian data or if it's European data and they're
basically impact and risk assessment is quite heavy the moment that you want to import the data and then with labels exported again they as well have heavy fines next the training that needs to be conducted in the in the in the in the US the model training of course there's also test cases that that people are going in it's not a model that is only specifically for one for one client as this company is quite good in providing these models to multiple tier ones and and oems and also there they're doing risk assessment they need to find ways to collect the data within California or maybe need to move to other states Virginia also released a new privacy regulation or hires on the verge on so you're getting the picture so if you're someone that manages this data flow you would say can we just get rid of privacy can we just focus on autonomous driving again I said you probably already got it Japanese then not making it even better in fact we've we've
been in in Japan early this year for a couple of months and for anyone who's been in Japan and in business relations um before um probably knows knows the situation for for the Japanese companies that we're working with it's actually not so much about the regulation but it's about ethics they're saying if we feel we're doing something good for our country for our Humanity basically doing something about data privacy then we should do it so in the end basically there was not a big aspect of looking at the regulation itself for those companies but um the aspects of doing data privacy so from this data Journey privacy regulation will stay and I could stay say it even more drastically Gardner predicts that 75 of the world's population will be regulated with some kind of data privacy regulation and it also means in order to collect huge amounts of data worldwide as a company you have to accept that this is part of the process such as for security nobody is questioning anymore that you need
a firewall or that you have to have access management these are all things that become normal and and regular and the same goes for for data privacy it's something that is is becoming and and will be even more an integrated part of data processing when it comes to Big Data worldwide and there are several uh several companies um that provide Solutions so from synthetic data where let's stay in the in our example where you try to argument data sets with 3D generated content there are companies with unstructured synthetic data where attributing the real world with artificial data in order to for example anonymize there are companies that do secure environment uh Federated learning differential privacy K anonymity so there's a wide range of of solutions and the key point is that these companies in our case smart cities and and Automotive they specialize on taking off the burden of the company of data privacy so we see it for ourselves we are successful at a company if our clients can forget about all these regulations
can ignore uh basically a session like this because they know data privacy is taken over from someone else who who does this on a on a continuous basis in 2018 and this is the only slide where where I talk specifically very briefly about our solution per se um in 2019. in 2019 the French the French tier one Valeo which is worldwide collecting data they wanted to publish basically a data set that they aggregated in order to flourish the community this is why I also can talk quite openly about it because it's it's public knowledge and it's I think 15 000 images from public 15 000 potential images from a Thomas that that can be captured there so it's highly critical to actually have uh have this data privacy compliant when you publish it openly so what we do as a company and we supported them with is anonymizing this data with artificial faces and with artificial license plates and why is this relevant for anyone who who isn't machine learning and training you want to
have the clearest you want to have the most natural data possible our CTO says data must be like sushi as raw as possible and in the end any kind of blurring and traditional back black bars would only limit a company to really execute their mission on self-driving Vehicles even worse they they might come to a conclusion that they don't have a measure and they would decelerate a development or even better if they have that they can drive the innovation was quite complex because it's fisheye data it's it was quite a quite a different type of camera set and now basically this part is covered for them and it's it's only as as I said for the landscape it's only one example of of many is how to how to bring privacy actually into the real world with AI and for AI and so then you have those single Solutions um but what you really need and this is what I said before you need to somehow integrate it into into your processes and they are
also the first companies that are doing this to uh to example Siemens building up Sim skeptor and one dedicated stage is offline processing and data anonymization there you see an artificial license plate DXE a tier two from the US also building up their data pipelines directly with an anonymization solution built in in order then in the end to uh to allow the four data being processed in a compliant way and I think this is this is in the end one of the key targets very explicitly we have Innovation with big data on one hand and we're having privacy on the other hand I hope I could give you a sense why privacy is not just an accept button but really really matters and that data is important I don't need to say and that it's really relevant it's bringing those two worlds together with new type of Technologies in the end what what allows um not just basically European companies that are affected by the gdpr but worldwide companies that are collecting data all around
the world to move Society forward in terms of innovation and data privacy and without further Ado this is it for me if you're interested in more data also we have a white paper on worldwide data privacy regulation you scan this code you I can send this to you via email or my my colleagues will send it to you via email and I know I shouldn't have finished that much early but I wanted to leave some time for some questions and yeah thank you so much thank you very much Marion and so whilst you guys are thinking if you do have any questions um I had a question for you so we talk a lot about the impact of labeling on effectively the ethics or the results of the output of the model and so as a anonymization expert maybe you could help educate me and then the rest of the group here on different methodologies of anonymization and how that in fact then infers a different result or potentially does or does not infer a
different result and what's your view there yeah so first anonymization before talking about the Technologies are very very glimpsy anonymization is a tricky tricky word in fact there are some people including our head of data privacy in in in in Germany who believe there is no anonymization at all in some ways he's correct because um so um we're working with the Deutsche Bahn they're actually one of our investors um and they have 40 000 cameras and there was a there was a discussion at which point data is sufficiently anonymized and then the discussions were going on and on and there was the point where we were talking if you have a single point in an image everything else is black a single point of a of a station and you also have the plan of when persons from the staff is working there when they when the train station closes the single point becomes an identity because you have to plan you have the point you know what the person is doing of course labor
law if this person is sitting in a corner it doesn't do his work right so it was a rabbit hole of do we want to really talk about it because it becomes very absurd at some point about the ifs and whens so which the worker has completely the right to be anonymous so this dot doesn't shouldn't exist but the question is is it basically only the randomization or do the rest of the processes also need to stay in place for example shutting the cameras off at some point so the more relevant question when it comes to an organization and this is by best practice now the recent definition within the data collection scope is it still feasible to reconstruct an individual within time effort and resources and if this is not the case then by best practice it's anonymized and I think this is a very much better definition for it and then if you're asking about different types of of anonymization of course you can have basically an almost absolute immunization almost again eventually
not possible like blurring the entire body right this is one for for image data for text Data it would be removing the data completely or you can go and approach with synthetic data that we you're replacing effectively the original data with artificial data which still resembles the original data but it's not connectable to the original data and then Is it feasible within time resources and effort to reconstruct a certain person for example then actually depends on the domain and we're looking in in our use cases with automotive and hundreds of thousands of hours and it's not feasible to click through an image and try to find with an artificial face a person anymore but it's more being a safeguard also from Advanced analytics to find identities and track them so I answered your question a little bit longer but anonymization is quite tricky um but it's it's uh it's a fascinating field because it's the it's the ability to actually remove the most sensitive data and making companies compliant fantastic thank you very much for
that right um so with that said I think we can move on to the next we have a question here from Faye you have the picture of a 75 percent of the countries now also leverage a gdpr right I want to have one challenge question because gdpr started in Europe right if you of course the it's a success for gdpr that 75 of the country in the world leverage this regulation however if you compare the success of AI I think Europe is fine behind China and the US and since you you promote privacy I'm also really think it's very important how can you predict in two or three years whether the situation will change that the European will be one of the leading the top two in the world because now 70 is the success of gdpr but we are fine behind right what's what's your explanation and your prediction for the future yeah a good a good question so I think that not just within machine learning Europe is behind but also in other
disciplines so it's maybe not the gdpr that slows down Europe but it's other things that slow down Europe because there are Technologies to actually overcome this um the for you for your question worldwide perspective more and more nations are adopting to privacy regulation similar to the gdpr and if there's a general understanding similar to the UN that that certain data needs to be protected then everyone at some point will be in the boat of having basically data privacy to to comply with and then it's not Europe anymore and in the US it's already happening in Canada it's quite happening so it will not be um a a competitive disadvantage at some point anymore if this continues as it continues right now and if you continue the development to solve this problem then let's not talk about the problem but the solutions to bring this together does this make sense it's reasonable but I I still do not see the Solutions in order because gdpr is a regulation right as I mentioned this morning regulation is
not sufficient we needed a way to the success right yeah yeah and hopefully we can take that offline and find some solutions together but thank you Marion one more thing yeah so you saw Thomas here we anonymized Thomas with an artificial face and he's not found anymore online so you can sleep now calm he's protected as well fantastic thank you very much and I should say that from you thank you right so thank you for such an incredible talk and it would be impossible for me to stand here today in Germany talking about AI with not having one of our most important players in the ecosystem Our Nation Beloved uh sap and we're lucky to have fee join