Mass surveillance, digital freedom and cyber security

Transcript

[Music] [Applause] anywhere I don't know I have okay have I don't another to okay check check okay everybody um hi I'm Daniel I will be moderating I'm GNA start um with some context and framing um I would say that surveillance and freedom whether it's

digital or analog historically has always been a bit of a cat and mouse game so there's always been while surveillance uh Technologies and systems have always been being developed there's always been a counter point to this um usually ahead of it that's just the nature of where that edge lies right uh I would say where we're at in the world today is that web 3 is a massive uh revolution around privacy uh obviously it's really the mouse getting out ahead and offering an alternative to our system um but to provide some context here today I think we're mostly talking

about the cat there's been other conversations around the mouse but we'll be really talking about the cat the system outside of web 3 where things sit uh an emphasis on uh China and Asia uh the SDS uh examples learned from there um and then we'll also open it up to questions because I want to see if there so be thinking about this uh this relationship between the cat and the mouse most most of you all are in the mouse World 100% so I think if we can try to have good questions that bridge between the two uh that can open up some richer discussions around implementation pitfalls Etc so just prompting you all

now be thinking about relevant important questions from Mouse to cat okay hope hopefully that's helpful and then I think we just open it up for um some quick introductions you know a couple sentences for each of you yeah good I would start Martin yeah thank you very much for having me here uh I'm an acad mic in Germany uh say in in the field in between of cultural science and computer science and did that presentation at 11:00 on the social credit system development in China so following up uh Pier's uh presentation yesterday and uh since we in some points agree and in

some disagree we would have to talk something about about mice and cats maybe thank you hi my name is yanim known as a Jew um so I'm I'm also an academic prin uh princial academic in sociology so my special list area is uh North Korean human rights especially women's rights so I've just published a book on North Korean women's uh human rights and their activism um so yeah I'm glad to be here thank you well hi and Pier thank you for uh M being all here it's a pleasure I'm also an academic although earlier in my career and I work mostly on China and uh

digitalization of the Chinese State and specifically the social credit System test hello everyone uh my name is Julio um I'm a researcher um at the University of Cambridge where I mainly study the use of AI um as a disinformation system and it's used in um influencing the way we see information and receive information great to be here okay so um I think I want to drop right into [Music] um maybe what we start right into is on the cat and mouse so um now and let's Orient our audience a little bit you you

all have told me that respectively web 3 crypto blockchain Etc is not your area of expertise no not expertise if I had to ask you what are the most interesting Technologies happening in web 3 right now at the Forefront of this issue you would say I'm not sure which ones I would say are exciting or how would you answer that you one yeah um yeah I would probably say I'm not sure uh there's definitely I think a lot of prospects uh like I'm not an expert in the field I do think that there's a lot of prospects for the use of block chain Technologies um in key characteristics I would say for example

the importance of decentralization um I think these are all features that will come to be more and more important especially for information systems um but I'm not familiar with specific Technologies at the Forefront yeah um yeah I mean for me beyond the sort of the technological sort of details inside obviously I cannot mention anything about that but when I look at what's happening uh in North Korea and how North Korea has been sort of exploiting this sort of um decentralized technological system in order to evade their sanctions and but also I think it's not just about the um

exploitation of the system but also how it is used it is not used for the general public or people if you look at actually while people are struggling with the starvation and and poverty they are using it to uh feed the El group just a very small minority of leaders for their luxurious of uh lifestyle whilst also continuing to use that money to develop Mass weapons development as you know so I think for me it depends on you know which which hands actually get hold of this kind of technology and how it is used I mean there is definitely potentiality

uh but in in practice I think we have to be very careful about the implications of this kind of Technology the potentials are pretty dire Martin I mean uh you said to me that uh you think democracy itself is at stake can you unpack that just a little bit more yeah so if if we observe the State of Affairs then we can see that say for instance in neoliberal countries like the US private companies grab everything so the data wise uh in the in the very East and for instance in the People's Republic of China it's the party who gets everything uh so this choice is not easy to do is

it is should it should it be the state or should it be uh private companies or should it be like in the in the EU where there are lots and lots of regulations so we are at a sort of Tipping Point I have the impression uh about u h how we could integrate uh communication technology into whatever we want our lives to be like politically whether it should be a parliamentary democracy or an autocrat or whatever but this is this is this is going wild at the moment and there is uh there are attempts which are say not very

convincing say from the EU to to confine this but then there's still the the stuff we give for for free or for for very for very little service if you just want want to know whether there are discounts around around the places where I go I I would I would open up my my my GPS data just for say discount for almost for nothing so there is an everyday aspect to it and I could think that you that you meant this also this everyday aspect that is not uh say a direct uh consequence of uh high

technology but of everyday life say of culture and um I think we should discuss uh how how we want our societies to look like who should who should own the data and then the high technology like like this what we what we uh listened to uh during that conference about privacy and partial revealing of information and uh is very important that that we could implement it but on in everyday life we give our stuff away for almost nothing so this this is the main phenomenon I see it's not a technological one but I say cultural and

uh economical and so sociological phenomenon that we face at the moment so one way out of the current situation is policy Frameworks a social discussion that that results in material changes on a a regulatory uh basis um and probably some kind of algorithmic accountability or something around all this stuff the um but we have democratized information Warfare at this point right and the power of various actors to interrupt that dialogue right to degrade the commons uh

around uh conversations understanding the basic principles of of democracy uh from uh um you know good faith communication uh an ability to even have communication um maybe uh Julio this is a good point for you to drop in around AI to give your perspectives on AI where that goes the problems but also maybe way ways out as it intersects both with policy and in general yeah absolutely thank you um and I mean first of all I want to agree with Martin's Point on Democracy um I think we are really facing a dilemma where um I think when

we think of data and our data we do need transparency we need accountability um we need democratization in a way and I think that all the actors involved broadly can hardly be trusted in this tasks uh we don't want governments to regulate how information circulates we also don't want companies to drive how information circulates um and we do need some level of transparency and decentralization there um and maybe look into new patterns or new ways of handling data and thinking about data that doesn't involve uh control whether from States uh or corporations um and I think with AI uh definitely there is a lot there to think

about um in terms uh I mean like if we start from the the point of data there's a lot of discussion at the moment uh about how our data is used to train Ai and that is a very Hot Topic at the moment uh because a lot of people are saying we never gave consent for a data to be just thrown into uh large language models uh for example um and that that is something that is it is a very significant privacy issue even though it usually doesn't lead to any type of like recognizability like you could never know whether it's your data or my data but it's still an invasion of privacy when your data is an knowingly used maybe from someone something you wrote on Reddit or on Twitter is thrown into

model for training um and then I think the other very important point about privacy um is the very high level of personalization that we have through AI at the moment from for example through recommender systems or recommendation algorithms which are tools that knowingly or knowingly for us uh use our data uh to predict uh and rank what we are likely to enjoy more as content which doesn't sound as daunting I would say because ultimately we like to see things that we like in a way so it's not necessarily A an evil task uh but they come with very significant privacy risks uh for example if the system are

manipulated or uh for example if the owner of an algorithm which could be a social media company decides to push some content more than other types of content in a say non-natural way uh these are all very significant risks uh for our information systems um and yeah and I could go on to talk about that but uh yeah what to hear from everyone else as well if you have any comments and points on that yeah any other thoughts on AI relative your work also I'd be interested Martin to hear uh about just information uh or uh perception of um what is true what is real what is culturally valuable uh and the breakdown

of that potentially from your side or yeah so this is a very big problem because those AI systems that are now fashionable are built in a way that they have to hallucinate it is it is a built-in feature you can't get rid of it if you build it in that way you do uh so I'm very skeptic uh what good applications of this Technologies could be because if you really want to know whether it's in an oldfashioned way true or false you should not ask them because they are they are built to uh to to dream in between facts and to EXT extrapolate uh and I'm really concerned

to see that that this is uh thought to be a good thing to use in education I'm not sure about this I'm not sure it should be it should be part of uh education in skepticism very very fiercely so so to speak so in in education I would use that as a at something that could save time from time to time but which you would have to look at very very uh skeptically on the out uh on the outputs and um we say when when when I when I do a glance to the US I'm really shocked about the uh the way how facts are ignored obvious

facts and I have more more the suspicion that these systems fire it more and more and this is really really very nasty because they are built to to not know what is true and what is not well if if I can sort of uh object a little bit to to to that point um I think I do see it is a bit more complicated in the way that I don't necessarily distrust uh AI or like llms or chatbots in particular I think they do a lot of things well uh I would be favorable in using them in classrooms um for proactivity tools for many other

tasks um I do think that there are two main caveat that we should be aware about and one as Martin mentioned is hallucinations like these models do hallucinate and make up things that don't exist at times and whoever uses that these models should be aware of that characteristic and they're not meant to be used as a Wikipedia uh and that should be kept in mind um and the other thing is I don't think I think one of the risks for knowledge is that they do collapse knowledge even when they say the right things like the knowledge that we acquire is sort of like collapsed towards the middle of the distribution because if you train any model or a large amount of data they will sped out the most likely outcome that they have

learn from the data uh so if there is a variety of opinions in a society n llm will probably converge that knowledge towards the middle because that's just what's more likely for um for a machine learning algorithm to Output um and I think that is a risk in in many ways because I think Divergence of opinion is a very important part of how societies work and how knowledge uh grows and progresses and I think that is a very real risk of llms in education like we we don't want to only learn the center of the distribution we also need to learn about the the sides um and AI sometimes doesn't allow that but I am very positive about the use of AI given these two caveats uh telling people to be careful about these things best case scenarios for AI in

addressing this Julio what sorry best case scenarios positive AI application in the issue of disinformation of the commons and in privacy and fre I think the best case scenario is that we get it right in terms of Regulation um I think in the EU there's a lot being done with the EU AI act where um the law like even though it's not going to be very easily applicable there's a lot of really important principles for example that the data of an llm should be fair should be transparent should be representative um so this all principles that would help um make AI safer um and I think the best case scenario is that we develop

very uh hard guard rails on how AI can be used and we regulate regulate that in a way that allows progress allow us to use these systems which are very powerful and very helpful in many ways um in the best possible way uh without creating un necessary risks so I think regulation will be very important uh um and I think for information we do need technological progress and that's where uh blockchain or web 3 could in a way uh enter into this because we do need better ways of validating what's human and what isn't uh which is something that we are very lagging behind at the moment um and there has been a lot of things like proof of humanness in the

past uh but I think in general the fact that at the moment we're not reliably able to tell whether something is a generated is a pretty scary Prospect um and humans are very good still at detecting um when something is not human generated um it sounds unlikely but when you you put enough people in a room to detect whether a text or an image or an audio is not created by humans we are pretty good at that um so I think in an optimistic way I do think we still have the capacity of uh making AI safe um but we do need to find Technologies and systems that allow us to do that and then we need policy that will make this

go as safe as possible but I think in a best case scenario AI would be a great tool for productivity uh to learn um you you could you could learn a lot of um things you could learn to code there's like so many things that it's good at if you are aware of what the limits are right yeah I would say in a world that where trust is fundamentally breaking down whether on a social or political cultural level um we're royally screwed unless there are trustless Frameworks right I mean or I think to Martin's Point like where are the Havens where maybe trust still exists in some semblance uh politically or culturally uh I can tell you in the United States

that is quite unfortunately going down the drain right now um so um well and I think this can allow us to transition right now to you know if we think about the world right now as going through a massive Industrial Revolution and that Revolution is essentially the production of tokens of of of pieces of information that can be utilized agnostically by anyone and and uh obtained these mass production of tokens of information in the World At Large then China is the most relevant example right now of them trying to have a repository at all

levels of governance uh and jurisdictions of a centralized information system that could be drawn upon at any levels um and so I think it potentially be interesting to go into that territory as an example uh with both Martin and Pier um for a moment and uh I'm particularly interested to start with the question around the social credit system again back in this cat and mouse of where it's not working what is what are the breakdowns of this I maybe we start with the intentions initially and then we talk about the breakdowns and where it really falls apart in

practice either one of you can start with the intentions so I think we would agree on that the intentions are very very very broad it is say generalization of Finance Financial Credit security to social credit security and building up of trustworthiness so the scope is very very broad explicitly and there are lots of documents that we both found and just just did a publication on this where could read the original material uh so this the the the scope and the intent is very broad if you look at what came out of it uh it is in some ways obscure it's hard to know what came

out out of it it is a very very big task you said in your presentation yesterday put up data about what's what's going on where so it's still a project at least but what I also found was that there is actual uh propaganda and and advertising in the public that say your social credit score is your second identity card as the the sonologist photographed in changa Airport so it's it's not it's not a way but it was intercepted by Co and there were much more effective Technologies to to hinder people to take on their rights

with the health app which is done by chance red and not green if you want to go to that place to protest against loss of your fortune uh so there are much more efficient means but there is still something in the background and that will not go away and evaporate without Trace I think I'm very sure about that oh I think we have in Broad agreement uh first of all it's not going to go away and I would probably just add on that the what I found personally amusing in different way is that the authorities themselves don't really know what to do with it that's how I would argue it in in some field work I did if you talk to people

in the financial system they would tell you what interest means social credit system is credit rating it's Financial Integrity but other people in the state Council or different part of the administration the bureaucracy would say no we would have uh we could use it for broader regulatory purposes so depending to who you talk to and after 25 years nobody really know what to do with it still so that is a reason why in the presentation I was really making the point that what really matters the most for now is this infrastructure that exist and as you said and I second this a lot on what you said is a more efficient means I lived in China for the most of the covid period so I had this health kit that we had to go around and

this is so this was interesting because it was both a cut and mouse thing so it was working broadly but then interestingly each province and each City a different health kit so when I would go from Beijing to uh chonga whatever other City my Beijing code would not be recognized there so I would have to apply for a different code and in the end it's not sure how much the information would actually be shared between different system although and this is one of the points that what you also made before on the rule of big corporations most of these health kits will be by Alibaba Alibaba was providing the same service to many different cities around the country but then for some reason it was not clear how much the data was actually circulating and so

when weing like we're breaking down a bit this monolithic IDE of China we see that different administrations have different views about certain systems we are facing in China the same issue that we have in the west with predatory companies that would be harvesting as much as they can information on people again on social credit for example to continue There is this famous seasom score that made the headlines years ago that was made by alipe um and the Central Bank wanted their hand on this data because it was like Mass financial transaction data on the whole country uh on the population and halip has been fighting a tug of war with the governments for at least 5 years refusing to end the data and been saying telling the Central Bank you don't even

have enough people to deal with this data you don't have the skills we know how to do this because we've been doing this for 10 years you guys don't know to do it why would I give you my data so breaking the idea of the monolitic China is important to kind of get like the infighting that there could be a different part of the state there's a small anecdote from Shanghai there there were those traffic signs where people were shown up if they if they cross the road on red light this is this is put down now I I had a look and I now I know why because it is a typical computer problem there was advertising on buses with a big with a face on it say business woman and she got all those bad points because she was on the adver and always crossed the road

and red lights for the pedestrians so this that this shows that it is far from perfect and working and so these are say youth problems of a of a overly complex thing yeah well I mean this is where AI comes in right there's the infrastructure there's the data it's overly complex in current system uh human system breakdowns but accelerating technologies will sooner than later solve some of these breakdowns um I think is safe to assume right AI is going to enable different interest groups whether it's the bank or other groups to obtain what they need to

obtain uh uh quite easily and maybe even uh implement or take actions on that um so I'm curious where it also breaks down more from you mentioned a example of like a local gangster who had the worst credit thing you maybe you just say that because I think that's an important breakdown that's non- digital right non-technical absolutely this is also a question I would ask everywhere around this table is like well maybe we are more familiar in the West but there is also I mean in China there's for sure a industrial Lobby of surveillance that makes millions of hundreds of millions if not billions of just selling AI whatever enabled softwares for for rec for facial recognitions and here is my where my expert my personal expertise

stops but when you read a little bit it says that it's not clear AR much facial recognition and I en works that well and well in a way some companies have become extremely good at selling new production to local governments who are going to pay hundreds of millions and again this is in where I would hear everybody else opinion because I'm well there is no argument that surveillance in China works like like very well and to a scary level like I this is not a discussion but how much the new technologies that are being branded actually add up to the system that already exist or if it's also kind of inflated bubble because we need to have ai product at the moment so here I would be curious to hear others on what they say if this new technology really are an enabler or sometime they

both are enabling more efficient surveillance or also just some kind of cosmetic new product that we can sell yeah I me like I think I I broadly agree with what you're saying like the Technologies for surveilance were already there for a long time um and maybe social media was all along the biggest surveillance technology in terms of like enabling Tech surveillance um I think things like face recognitions have improved a lot in the past years uh thanks to stting the improvements in AI like I don't know if you all saw like for example how an AI can now solve captures very easily so these are things that are relatively new um so some improvements have come for surveillance

for example just improvements in facial recognition but I do agree that the technology has been there for a long time and if a government wanted to surveil a population like you don't need AI you don't need anything that is there at the moment uh it was probably easy before social media as well but with social media messaging apps is just extremely easy at the moment um and and it's all very shady as well like even within um within Europe and within the EU like we don't have that much information about how data is being used like even after gdpr and DSA and the AI act now uh there's still there's still a lot that is not known um and something

that I'm very passionate about we still have very little access to social media for research uh which is something that we used to have for many years uh and now even though the DSA actually mandates that uh platforms that are designated at online as online large platforms have to give researchers access uh that's still basically not done and most platforms will still re refuse access if you ask um so there's still a lot that that we don't know and it's not improving but I do agree that AI is at best like a small fraction for now um but then like this could change like AI like adds a lot to our like predictive capabilities uh which again

is something that existed for a long time but we can predict a lot better now for example your preferences that we could before uh just using your data so these things are improving and they could get better at personalizing um could get better at scaling um AI can make surveillance at scale less resource intensive so that's something that could change could make it cheaper so there's a lot there that could change but yeah the technology was there already and probably the int the intention the intention to do it was there too thank you can I can I add yeah I think in terms of um sort of the role of technology I mean you mention um uh about the uh the sort of surveillance

already um done well before the technology which is if I use the examples countries like North Korea which is very true because they have uh the system where the through which you know all the human beings manage to actually survey people through the planting the fear into people's head so it's you know the whole idea of panopticon I don't know how much you are familiar with the the idea of panopticon where the you know this uh Bam's ideas about this prison cell where the the people are kind of believe to be watched all the time when you don't actually know where the the God is watching you or not so this kind of psychological

kind of uh Warfare is already there and you know people people cannot trust anybody however with the Ed the sort of a uh technology it can have a huge implications for people's uh human rights and their ability because at the moment uh for countries like North Korea people can with the quite a lot of struggle and risking their lives they can escape their countries but with the fasal recognitions and things like that you know I believe the technology at the moment obviously we we've just started having the whole range of discussions around ethical implications of AI and all of

this but the speed of development has been really fast and I expect to be fast and the technology will improve hugely significantly and with that people's chances of being able to exercise that freedom freedom of movement freedom of expression and freedom of Liberty will be affected quite substantially and that's a real worry because the government ability to be able to actually monitor and Survey people uh will be better and better and it's going to be really difficult so I think we we have to be very cautious and you know very worried about the the huge implications uh for this kind of

Technology on individuals really and and today like uh whether in Korea or the weaker uh population in China do are examples of how uh any of surveillance is being used maliciously that you could point to absolutely I mean uh this is a thing that like there should be always said clearly uh uh as much as I spent a lot of time deconstructing some ideas about China looking at what happened in Shong and the mass concentration of wiers population that happened mostly between 2018 and 2020 2021 uh was absolutely uh an example of how a state

can become repressive and interestingly it was both done by an incredibly large expansion of digital surveillance uh with density of camera and gateways and mandatory phone scanning at different checkpoints around in the in the cities uh but also the as you mentioned and I think that North Korea would be an interesting example it's also human surveillance and the eyes of the others and denunciations or whatever and when you combine both this is where you see how terrifying it can be I would like to bring back the the uh characteristics of large language models on this kind of AI

to this topic if we imagine that these kinds of AIS would be involved in Mass surveillance then their outputs would be very much convincing it would actually really look that I broke in a jeweler shop because that guy looked like me actually and it would fit to my habits and it would be so convincing because these hallucinating things this this this uh the storytelling without without any grounds factual grounds is built into into those systems uh this is the absence of reason for that that there is it is so convincing but it could be wrong and if you think that there is

mass surveillance with lots and lots of extremely convincing evidence that is plainly wrong because it's by statistical means say 20% is extremely convincing but also wrong what what could we do with that so I'm really I'm really scared for this because there is no no reason forun to to uh to uh to check it and it will be and this is the the very thing these these machines do extremely convincing especially because they offered you a bargain deal at the kbab shop right next to the jewelry store and you went there and your GPS data you know points to Martin as the he got a

kbab and some jewelry and so he's gone for now J do you want to talk about some other human rights intersections and your concerns or views yeah I mean I mean beyond the to be honest the the examples of North Korea I mean you know the the list of human rights violations are you know quite endless so I don't really have to go into details but in terms of the implications of AI and the technology one of the things in relation to what Martin just mentioned is actually it it will come to uh the crisis of identity at an individual level but also Humanity level because we are starting to having questions around

what what what it mean to be human you know and we are raising those questions and with the development of AI and with the uh the sort of the speed of development there will be points where what is the sort of um the the the line between differentiat between humans and the Machine but also when it comes to you know AI has been used on a much more regular basis and I'm confused with this uh my replica of my version that really raises questions around who am I and who are we so I think you know there are whole range of ethical but also humanist questions we really have to think very

carefully about you know how we are going to use this technology and what are the kind of consequences quite often something uh unforeseen because I think you know whatever kind of a witnessed is people seem to Once get hold of something there are always this minority people who are very much focused on the power and then they are enjoying that exercise of power and that can have a really negative kind of dangerous outcomes so that's something yeah worries me okay okay we're going to open it up for questions shortly I just wanted to maybe lay out another framework which is if door one is the centralized

acceleration of Power by companies and governments that leads to a sort of uh authoritarian dystopia right door one awesome door two is um sort of a Libertarian quote unquote democratization all information right uh and all of the security uh goes out the out the door right so whether that's bioengineering we know weapons of mass destruction that's now available at chrisper to available at anyone with a garage and the basic chemistry set or any number of terrible things that people can do that you have to try to protect your population as a government

against right you have to control for these really radical Extreme Measures that are much more than the than just guns right that can happen harms that can happen to public that's door two right you don't control for that right that's sort of a c catastrophe too right that's not dystopia and it's catastrophe is door two so we're really looking for okay door three what's door three here right where's what are or four or or five um and I don't think we have the answers to what door three or four or five is but uh but it's uh quite serious I I continue to try to elevate the level of the seriousness of the conversation for the developers and the builders and

the people in this community because in some ways I don't think they're hearing the contextual the the alarm bells that how critical some of this technology may actually be I would add add just one more tension as a summary before we go to question answer which is the accelerating pace of technological development versus Democratic institutions a speed at which they can regulate and keep up with information is also a very difficult uh Bridge uh and the other tension being information uh disinformation uh common the commons of democracies being able to make sense to steer through this uh accelerating confusing complex world that we're in so um not to end it on a bummer note but those are the tensions I guess I'm

seeing any other major tensions that you all feel like or you're wrestling with or see when you look at this landscape like anyone yeah no I mean the way you you summarized it is I think pretty um compelling uh again I have only a limited expertise mostly on one area but I think it's also interest like what I always find amusing and it's probably the same here is um how would such question play out in a country that is by definition authoritarian when the counterpower doesn't exist and like so it's only internal B checks and balances between different part of the administration if you look at the whole

uh AI regulation discussion in China which I'm only very farly like familiar with uh there is interesting debates between uh the more academic or maybe scientific communities that would want to engage and will engage politically with the rest of the world and more maybe more part of the bureaucracy that are uh viewing pretty much everything from National Security lens um another examples like this that is very brief uh well last year or two years ago China implemented a basically control check or like a capital like a data check that couldn't be exported freely for companies it was a huge issue let's say you a company that have R&D in China and you were INF Fury not supposed to

transfer uh much information that you do in your R&D to maybe your aars in Europe in the US of course this created a gigantic backlash because foreign companies were like what or even Chinese companies were like what is going on interesting what came out after a year is that they completely relaxed these controls realizing under the pressure of the the business environment but also realizing that the administration in charge of this simply didn't have the number of people this is there's this is also maybe I don't know how much a this is again I don't know how much AI will automate or not certain task but when it comes to the regulator even if you're the National Security oriented regulator and you want to check everything maybe you're just not going to have enough people to do with this cuz in the end

you still have you still need bureaucrats you still need civil servants or engineers and there will be only a certain amount of them so maybe that's going to be just a little side yeah yeah I think I I I agree and I think it it connects to to the point on how quickly regulation moves compared to technology um and I think that will be a very relevant problem in the future because technology is just increasing in Pace and regulation just can't keep up um there is a monetary problem where the people you would need to get to uh to work on these Technologies from a government perspective are often too expensive for a government um like imagine from a large language model perspective like if you're an EU regulator you're given access to I don't

know the insides of gp4 like what can you do about it like they they wouldn't have the expertise to start looking into it um so it's just like regulation moves slowly and often doesn't have the means to actually look into these things um and I think that will be an important Trend uh for the future okay we can open it up I I would have one more suggestion these systems look int these AIS look intelligent because intelligent people interact with them if you let them interact one with another it's it very quickly is shown that this is that there's much lot of nonsense coming out of it so there is also this uh this

two-part these two parts in in conversation and it would be really necessary to flag everything that comes out of those systems which which are which are really important and really handy and and a big surprise for many people also for me if you would flag everything as something that came out of those types of machines because then we could learn how to how to deal with them so we learned how to deal with other digital stuff also which uh mimics reality but this is a new a new kind of of stuff and if I knew that some came out of of those things without reason without F then I could could think of it and deal with it in the in

the in the in the proper way then it could be handy and I could be skeptic so this would be a wish of Regulation flag everything as being coming out of that MH explicitly okay any any questions so yeah you've talked a little bit about um regulation and I I want to kind of move that around because we're developers and um the problem with regulation that you pointed out is that

it's slow but that's probably good because it's also incompetent um you know if you look at the farmer regulation it hasn't protected us from fake drugs it hasn't protected us from side effects it hasn't protected like just hasn't worked so I'm glad that Regulators are slow but also we're confused between regulation and legislation right like it used to be that doors would never fall off of airplanes because there was regulation like suing people after the fact is not regulation that's enforcement and so what would you suggest to us as developers as we're developing um our Technologies in this

age of mass surveillance in this age of technology which has more unintended consequences than intended consequences if we actually wanted to self-regulate as an industry or as developers because we know we can't count on those authorities what would you suggest that we be doing yeah I think I mean first of all I I do get the point I think it regulation is always a very tricky balance between like stifling Innovation and actually making societies safe um and Regulators should be competent uh there's no excuse not to be and we hope that uh regulation should be competent in what they do um

I'm mainly talking about how quick it is for highrisk scenarios which there's a lot of Technologies particularly do dual use technologies that can have very large scale High magnitude risks um and these are the ones that should be regulated quickly and with expertise um I think my advice to developers which I a study that I hear a lot from my AI ethics colleagues is just develop Technologies thinking about what some of the downstream implications could be uh which is something that surprisingly does not happen a lot especially in AI um a lot of computer scientists don't think about the implications of some of the products that uh they have veloped particularly

in AI particularly larger Labs um and I think so my only advice could be to think about what could go wrong and have that kind of mentality um where I do really see a divide between people that think about building and people that think what could go wrong about that technology and it's a fair division but I do think that it would be helpful if developers had some thoughts about potential impacts of releasing any technology into the world as well anyone else wants to answer the question any other question hello and thank you for the

talk uh my question will be uh what could be your definition of digital freedom and in what it will be in front of the evolution of AI in the next years digital freedom I would insist that it's not different from the other from the other Freedom we have like I can take it um I think my definition of digital freedom is that we know what our data is used for and we are in charge and control of our data um I think today that doesn't happen a lot there's a lot of implications that we're

not aware of um and we see for example with data used in AI for very responsible purposes which could be detecting military targets through AI predictions there's a lot that is been done it shouldn't been done um and that's not U the purpose that the data had originally um and I think digital freedom for me is really about knowing what data is used for and being charge of the data that I produce and knowing that one day will not be used to know where I live to for like War purposes or for targeting purposes I think that would be a good definition of for me that freedom is awesome I think we are at I'm unfortunately lots of interesting things

to think about but thank you everyone for joining and thank [Applause] you can you go did you did you know beforehand that we would walk on this on this carpet [Music] you