now we have discussed the german level and also the european level of regulation but the question is what's actually coming up what's in the pipeline um and fabian just told me that the next guest is also particularly interesting for me because it's what she will tell us is what will happen for all of us and especially also startups and people working in the field of ai so coming up next um is um somebody from a legal perspective so that's new in this track doctor ozil and fender schmidt she's a partner at laosen-rexhanvelte and she will focus on what's in the pipeline for iron regulation please give her a warm welcome very glad to have you here ah i think you might be muted or our can we hear her nope okay i can tell you people are running around trying to fix it but you can hear us that's good so at least one way it's working okay if anybody in the oh yeah now we can hear you amazing okay cool okay okay great
let's go okay let's go i have 29 minutes left so thanks for welcoming me uh thanks for having me um yeah uh i had um heard what uh damien said at least most of it um due to technical checks and of course what uh yes my aim would be to give you a little deeper perspective of what's really going to happen because of course um there's a lot of theory out there now and um lawyers would like to be a little more specific so what we are talking about um lawyers generally love definitions so if we hear artificial intelligence um we might have um all kinds of different understandings and therefore lawyers always start with definitions and this should not only be a lawyer's thing it's also very interesting for you because what's coming up could affect you or it could not so therefore it's really important to define who should um comply with new obligations and who's meant by new so here's what the european commission is currently working with artificial intelligence systems are
software um and possibly also hardware systems but most likely software is included in those systems so this is the start we're starting from software we're starting from some thing we already know specific software it's a software which is designed by humans so we are leaving out any system which is actually designed by artificial intelligence so we always look at where we started and we most likely started with an invention by a human being this human being is giving this software a complex goal which is meant to be achieved and the artificial intelligence is then actually acting in a physical or a digital dimension by perceiving its environment through data acquisition it's interpreting the collected structured or unstructured data reasoning on the knowledge it has collected processing information derived from this data and then to achieve the goal given so this is it's a definition suggested by the european commission's artificial intelligence high level expert group and i think this is definitely where european legislators will start with so we are not only talking about strong
ai we are also talking about very simple algorithms interpreting data and then deciding upon a very simple action like um advertisement being posted um to a targeted group so is that all some people that have spoken before with those it seemed like we had no regulation at all and now that politicians and lawyers are coming in um everything is going to be very unsexy and we are preventing things to be done um so is there actually no regulation at all well um the people who are dealing and producing products with artificial intelligence know probably very well that there is a lot of regulation already in place we have the general product safety directive of 2001 we have of course data protection um we have the consumer rights directive we have a lot of sectoral legislation like the machinery directly and of course overall we have the fundamental rights which have to be kept and secured with any product the fundamental rights are expressed in a couple of further directives like the race equality directive or
directive unequal treatment in employment and occupation so we already have a lot of things and there are even more rules damien also stated that there are already a lot of standards out there there are general standards from iso and iec for software and then there is sector specific standards like the iec 82304 one for health software for example there are iso standards for quality management for software for security standards and also for standards of use there are of course the technical and organizational measures you have to be aware of with the gdpr so also here we already have a lot of regulation uh which already applies to artificial intelligence products now why do we need more regulation and this is something that had been addressed by speakers before actually lawmakers but specifically also society sees gaps in existing law so they said we need enhanced security because artificial intelligence has always something to do with connectivity and openness and therefore this could lead the specific situation could lead to manipulation intruders as we have discussed
a lot today already there's data dependency which could lead to bias and also to false results this data is accumulated and curated that's how um quality is here for the state and then we have autonomy artificial intelligence is acting at least how society sees it could act autonomously of course every artificial intelligence has been programmed before but it's taking its own decisions based on a set of data which is then not again scrutinized by a human being not every step is watched by a human being and i mean that's purpose of ai there's a lot of fear uh german angst as uh was mentioned before about opacity so that um society doesn't know what's going on how those decisions are made what's the next autonomous decision the ai is taking there's a high degree of complexity of products and systems and the complexity of value chains so and therefore very often um product liability laws don't fit very well they don't fit to services at all um fault-based liability rules might not because um if
the artificial intelligence is acting autonomously there might be no one it might be no one's fault and i peelers don't fit either so what's actually in the pipeline and in this regard it's quite good that the conference had been delayed a little because in may there was not so much out when i first spoke to fabian the white paper by the european commission has only leaked um and nothing else really was was out there so as he already said it seems that the european commission is acting much faster right now than we were used to it because in formal times when a white paper was out on the table you could at least wait uh four to six to eight years until this was made into law that it could be a little different this time so my initiatives are on the table there is the european commission white paper on army and even more important and more interesting the report on ai iot and robotic safety reliability a lot of specific uh evaluations on
how liability law applies to ai right now and where are the gaps so this is really the european important that identified that something is missing is quite important because it doesn't really make any reference to effective property rights for the development of artificial intelligence so the european parliament put out a report on this international property by its aspect my only second october benefit um in parallel the uk commission is working on a large package the digital services act this digital services act is to replace the e-commerce although artificial can you hear me because we can't hear you again uh was there was some outages in between so we were always hoping for a bit more constant connection i don't know if it's your microphone now i see myself in various dimensions that's pretty cool um so maybe we can try to reconnect her i don't know if that's possible um or we because then uh because we just can you hear me now now now i can hear you okay that's better okay because i'm
actually not doing anything other than a technical check yeah there was it was on and off that was the issue uh so there are millions of people working on it but so maybe i don't know if it's your microphone you can take it out um and then we can just because we have a bit we can just jump in we have 15 minutes left so if you want to we can try it again can you still yeah what do you want me to do cool if we can just go back maybe one slide sure so this one the gaps i don't see i see myself not the slide let's see the one with the bubbles the one with the bubbles this one and and there's no bottle slides yet sorry i can't just i can't see it but maybe we we do this one now and then a bit more because i think now the part got really interesting um and then yeah so just interfere um if you can't hear me because uh of
course i want to i want to tell you all the important things and and it's now getting really important so um i was ready to take notes so please go ahead okay all right so um i i just started off with this with this slide where i said what initiatives on regulation are on the table right now and we had the white paper out already since february this year but other important suggestions are coming up as well as um there is the report on intellectual property rights for the development of artificial intelligence technologies of october 2nd so that's brand new then there is the digital services act the commission is working on right now is a larger package uh where we just had the consultation process um and where uh the the committees of the european par of the european parliament um put out their suggestions what to include in such digital services act i'm coming to this um later on because the digital services act is going to be really important for you and
really important to watch there are an international level there are conversations started by the wipo on intellectual property law and artificial intelligence and then as you heard already from dortheia bayer there are there's a national strategy for artificial intelligence the us of course had put out uh special artificial intelligence for the american people and there are a lot of further national efforts i totally agree with damien that of course this all only makes sense on an international level because you all want to sell your products to an international crowd so um no use uh for national strategies um without any any international um connection so how um is the scope of future regulation um how could this be done and and we heard a little of this before as well it could be sector-based um which means that um there is specific laws for single sectors and industries or there could be a risk-based regulation by evaluating certain risks for certain types of ai application this will also consider the sectors but it's more the
the focus is on the risk a certain ai is providing for society and for fundamental rights so eu strategy seems at the moment to be a little to the to the right so a little more risk-based but they will have sector um specific considerations as well the the european commission put out uh some central goals and there it gets a little cloudy so they they um have the overall aim to form an ecosystem of trust um they want to consider consumer protection privacy and data governance society societal and environmental well-being protection of fundamental rights accountability strengthen trustworthiness and ai and ensures general social acceptance of ai but can we be a little more specific because this would be really helpful for people who like you who are working on those products and i think yes uh we can get a little more specific so what are the plans eu legislator actually plans for a set of further obligations for products that include ai there will be very likely special requirements for high-risk applications like geometric
identification like of course in sectors where there is a significant risk that can be expected to occur like health care like transport like energy like parts of the public sector but that will also be taken into account um whether the ai application in the sector in question is in addition used in such manner that significant risks are likely to arise so the second criterion will level down where ai is used but no special risk is involved in the in the certain usage of this ai we heard that there might be external regulatory audits and explainability um there could be ex-ante which you put it in the market conformity assessment and labeling schemes there will probably be the obligation to inform competent authorities and also to inform users about interaction with ai there will probably be an obligation for human contact point and for redress systems also to obtain trust and reliability in the system should be technically robust and accurate this is all of you there needs to be quality origin and neutrality of training
data to achieve this technical robustness and robustness and accuracy of ai there will most probably be further rules on on data protection probably also data and record keeping to reduce obesity um accountability liability and burden of proof uh will certainly be um an area where new rules um will be set out and um we at at the moment the uh european um commission um makes it obligatory to have a human and control and also to provide a risk assessment of the ai that is used with the european parliament's resolution just now october 20th we get a first impression of what could already be included in the dsa package a first draft of the digital services act might already come beginning of december and we will then see whether artificial affiliate artificial intelligence regulation is already included the european parliament voted for suggestions to at least include information obligations the aim of technical robustness robustness and accuracy um the special protection um for data but also of course a check for the neutrality origin and quality
of training data and most likely there will also be new rules for accountability liability and burden of proof so who does have to care um at the moment um the plans are that address should be the actor who is best placed to address any potential risk so this is not only the developer and it is not only the deployer of artificial intelligence it could be anyone who is in the actual position to stop any kind of infringement we have seen this in other european legislation um like for example the the possibility to get injunctive orders against intermediaries used for copyright infringement in article 8 of the infostock directive so i think this will be an approach which is also interesting for you that that this obligation could hit anyone in the value chain now what are the consequences and takeaways uh from this new plans first of all of course it's annoying for you regulation always means higher expensive that's clear legal compliance expenses it's called but regulation could also save money because it means
clarity on responsibility and this makes it easier to calculate and especially to ensure risks so insurance companies watch out for liability risks if those liability risks are cloudy it's much more difficult to ensure those risks if there are clear rules you can comply with like accountability and transparency requirements or also conformity and labeling requirements you can comply with this and therefore reduce for example your insurance fees and also calculate with the risk you're setting out um also as all of this will come into place um probably sooner um than you might think one could also make up their minds whether ai driven business already might require specific contractual frameworks with future anticipating classes this is sometimes done when something something is in the pipeline it's not really clear um what the exact wording of a new law will be but clauses in contracts can already frame this for liberties also this can also be taken care of in contracts so there's a little more i wanted to tell you um i'm actually a copyright media
i'm an ip lawyer so my focus is also on ai nip and there are a couple of very interesting questions around ai and ip i will just use the last three minutes to go through um the first one is is ai protected by ip why is it important it's very important because it's all about money um because the question is do i have to license the ai i'm using and if i'm the developer of an ii can i eventually license the ai to others or is it not protected at all is it only my trade secret and i have the the only option i have is to keep it secret so um there are a couple of uh measures but nothing is really fitting it could be a patent yes but only if it has a technical effect otherwise just mathematical methods as algorithms are not patentable copyright for software yes could be but only if it's um a specific expression of humans own intellectual creation so there must be some kind of creativity with
it underlying ideas methods and principles are not protected neighboring for databases yes can be but then it has to be the database which is the thing that isn't infringed so it's not really fitting either um how about ip as food for ai ai needs a lot of data and a lot of this data is actually content protected by copyright could be blog posts newspaper articles scientific papers or photographs found in the internet um all of this is copyright protected there are exceptions to copyright um with the text and data mining exceptions but it might not actually apply it's quite narrow this exception so you really have to check before you're doing this data mining whether the content you're using is copyright protected and whether the things you're doing with it like reproducing it for your mai training is a copyright infringement otherwise you really run a higher risk of liability and need to agree to licenses with the right holders third question is is ai creating ip you certainly see that ai is capable of
producing art that at least looks like um art that is protected by copyright so right now uh copyright protection is not given to anything that is produced by ai because creator and holder of a copyright can only be a human being so at the moment there is discussion about how this can be helped with and whether ownership of copyright for example should be given to the person that legally prepared and published the work which is created with the help of ai and in the end ai might even protect help protect ip because as you might have heard there's a lot of discussion around article 17 um at the moment of the dsm which requires content identification for platforms like youtube um with this content identification ii can actually help to protect ip of course it will be interesting whether ai is any anytime capable to detect something like irony which is under copyright exceptions but that's even sometimes for humans hard to detect so as you can see they are already marking um their certain
irony so as you have seen and as you have hopefully heard and my tone is hopefully not off again the pipeline is long and the pipeline is packed so stay tuned thank you thanks a lot and we're very sorry for all the technical issues in between but we hurt you very very well until the end we don't have time for questions um thanks so much for all of this and i guess there will be a lot of people asking you no questions later that's fine get in my chat cool you