Robots behaving badly

the best of we the other

thing is thank you thank

you very much of this is the 3rd time that have come to or again and have enjoyed it

quite a lot of our I've become completely addicted to to to I my title was roads behaving badly but I I keep thinking that probably should change it to this I for 1 welcome our new robotic and robotic overlords now every time I % I have a 2 different there are are things that I like to do 1st is an apology and

then a story 1st the apology and the apology is about them I've presented now for 2 years have been researching on the right side of artificial intelligence They whether copyright and what with the role should have copyright and whether we should allocate any sort of ownership to artificial intelligence this is the 1st and then looking at the other side which is the responsibility the liability of IntelliCorp pair of a fair artificial intelligence and robots and this story

is that I became very interested in this area particularly because of book by a Châu Straus and it's called a halting state it so it's a sequel to Caltech status with called rule to 34 now rule 34 refers to an Internet mean that it states that everything in the world can be subject to a Internet pornography but the book a specific the talks about the rope rubble to robe agent that has gone crazy and started killing people with Internet of Things devices now this is completely independent and just be because it decided that the people where his enemy's this specific people were Internet trolls and the set of these people are trolls night in my objective in life is to destroy trolls so I'm actually going to do that to kill them he organized some very interesting ways for them to be killed and that could be thinking into various things about the responsibility that is allocated to artificial-intelligence and 1 of the problems and 1 of the the the 1st issues I

encountered is that we do not they understand quite well and what we mean by internet the artificial intelligence

so that gets to my actual apology my apologies that if you were here they invited us in in any way by algorithms watch I apologize because I don't have a lot

to say about algorithms yet my research is ongoing and the people that are working quite interestingly in algorithms and particularly some of my colleagues alien Edwards' Michavila shout to them people were were looking at the regulation of algorithms so I won't be covering that

specifically I also want be covering 1 aspect of the self-driving cars I will talk a little bit about that I can come to Germany and not to use not not use david Hasselhoff against and grow with with this but 1 of the things I I wasn't there I don't want to cover is the regulation itself of self-driving cars now my idea and I'm perfect if it finds people disagree with this my own idea on a self-driving cars on the regulation is that it's going to happen it's it's not a matter of of of if it's a matter of when and I think it's really quite possible that and the regulation is going to be quite quick because of the commercial interests that are surrounding it so the regulation itself is not a problem I think is it what is going to be a problem also other

things that means that I won't be covering the trolley problem which is quite famous that when people are talking about regulation of self-driving cars and also

the 3 laws the laws of robotics air very nice analogy very interesting thought experiment but they're probably very useless when

we're talking about in artificial intelligence and also of course something

that is not and I knew I'd miss spending a lot of time telling you all the things that I'm not going to tell you which trying to give you an idea of the things that I'm probably going to tell you because this areas are not as

interesting the they get a lot of the coverage in the press they get a call a lot of interest and people document artificial intelligence but they tend to be a little bit lists interesting if I may say so like that now it all something else that I won't be covering is a cure drone so the idea of drones and robots are going to kill and and and automated ages that are going to kill people 1 of the

problems that I think we have been encountering so far is the trying to find a good definition of artificial intelligence and is this 1 of my favorite so the the the of start in the amazing book a Galicia back AI is whatever hasn't been done yet and sort of police into the idea of advantage of the it so is the future the

future of things that are not yet in existence things that are not possible this we we tend to imbue Artificial Intelligence at least in in in a lot of of the coverage in the press but also in in academic writing in sort of mythical almost godlike powers to artificial intelligence would tend to give them that now diet may become something along those lines but I guess that is probably going to

be something like this when we're talking bird artificial-intelligence we're talking more of a about various specific this devices robots that that do vacuum cleaning and roles that put together cars and all sorts of things so much cleaner more interested in the side of artificial intelligence has seen that this is where the legal issues are going to be more important in the mundane in the every day and whether your own by this Roman very kills the cat goes crazy 1 day and I think that's probably going to be the most the most relevant aspect of artificial intelligence

so 1st we have to start by defining what is artificial intelligence I like I told you that the definition by a Douglas house that the but that this is a more recent 1 in a very good book on artificial intelligence is a collection of essays air by Russell and Norvig it sent to else on the Russell but anyway AI is a study of agents the received precepts from the environment and perform actions and I prefer definitions like this because this types of the Phoenicians are actually leaving the intelligent parked outside of that I think it yesterday in in molecules we heard a very nice term about natural stupidity and are supposed to to artificial intelligence and I I think that if we leave this smartness the intelligence the relative intelligence the of the agents that is going to be more useful when we are finally dealing with with the legal consequences of this let's forget about the magical and the superhuman artificial intelligence that I that we see the point to common destroy your source and harvest us for as batteries early to mine Bitcoin or these all their idea of absolutely all-powerful almighty all purposeful artificial intelligence

so I want to their provide 3 scenarios for you or sorry 5 actions and prof you of the things that they are really taking place that part of artificial intelligence so

back in March all this year this was covered a quite a lot on the news I'm over there and car in Arizona crashed they into a person that I don't know if that's correct term but killed it this cyclists in in Arizona the police have made the details available and the video of the ways is quite terrific but prompted quite a lot of discussion of whether whether who was responsible for this was who were responsible was the person they it was supposed to be a person in the car that was going to stop the car if something bad happened and the the the in cameras and in in the car actually gave us a good idea that this person was like on the phone and wasn't even paying attention so this prompted for a lot of debate of what's going to happen on something like this this is going to happen this is going to continue to happen and everyone that even the most strong advocates of a self-driving cars are telling you this is eventually going to happen at some point a car is going to kill people now we are not going to get at any tests in the courts because over the in Wybe it was that everyone was expecting them to do the they immediately settled a out of court with the family a for an undisclosed amount of money were pro it talking about millions and millions and millions something more interesting

perhaps and this is something that is going to be more mundane and that's 1 of the reasons why am more interest in the mundane a elements of the of the artificial intelligence are things like smart contracts at now these are contracts that they're written blockchain by the way anyone playing and Republica in being you're getting everything to me you're going blockchain you're getting a algorithms were getting an artificial-intelligence air but anyway and have been very interested in this because this self-executing programs they're called smart contracts there it in the block chain so They're inferior some of these contracts cannot be changed and when they're executing you there is some mistake in the in the cold air that mistake is going to be for ever so is going to be embedded like legislation inside the blockchain because as on a mutable now other interesting case happened in November 2017 there is called the power at the wall at this is a contract that was designed specifically for Maltais signatures and there was a ball the in the in the security execution all the birth of the contract that meant that you could not access the the wallets at at some point 5 there is a condition of the time this happened about 500 million US dollars between 300 a muon depend on the the price of the chain is quite a lot offers of money was a lot and it's still a lot and this is just the computer 6 no the computer this is a contract there is this completely autonomous dynasties actually refusing to hand out the money and in the wallet now this only going to happen more and more in smart contracts and it is again 1 of the things that interest me the most because these are the implementations narrative of artificial intelligence that we're going to see every day also

examples of data mining for a for music things like a machine learning these interests me a little bit more because this is actually what prompted a lot of my research and copyright we're talking about a mention in the machine learning algorithms are going to go out and to try to learn from a large numbers of the data and

I could give you several examples but there is 1 deal for example that has looked at 23 thousand approximately 23 thousand irish folk songs and it is now producing music on its own now the interest for a copyright law you know you're like me is whether or not it any of the composers of the 1st 23 thousand songs could actually come back and sue the makers of the program because they're infringing copyright if you're training your Artificial Intelligence where you're you're you're the machine learning with any type of a large corpus of data and that's see text of musical poetry or movies or pictures or whatever what are you and cooperate and this is a slightly tricky so there's another 1 I love talking about this this was like this this is the case was presented at the public I think in 2 thousand 15 this is in 2014 Swiss group called median group Mitnick I'm probably mispronouncing that horribly they they created our shopping what called the dark net Web shopper that went to the dark matter and purchase things using Bitcoin and was just completely random whatever fit the price of the time that was OK you would buy it and it purchased then

yellow ecstasy pills and that our pictured here the that the box was taken into custody where good then later released and because of leaders no rare and was considered to be a an art project that ATR etc. what is interesting is that this is the older it think that is going to in happened I think not only with the with the a smart contracts but also in a lot of a smart bots and a smart

contracts that are are are going to go out and do actions for you is increasingly possible with artificial intelligence so

what is the legal discussion here generally the response whenever something a new technology arises there is going to be as follows you the bandit the technology which is quite good it

it is quite common and in in so a certain types of technology but Purdue biotechnology and you can start to regulate that you can try to self-regulate so whoever is involved in the technology the com the companies are going to be involved and quality are going to self-regulate or do nothing and sometimes both this 1 and the same the you probably get a strong feeling of what I

think about self-regulation and you can do something that my my colleague at christmas and also the University of Sussex go and Brown from from Oxford called coregulation which is actually and it almost like that a regulation between them public and the private parties are going to be at work symbiotically to try to to reach a good results you could try applying existing law because changing the law can be difficult and of you so you could draft new legislation to meet the technological challenges and that this is usually the solutions that are offered in most new technologies now with all sorts of phase of soul the again I'm going to talk a specifically about artificial intelligence in the sense of limited but also not from the regulation itself bought from their legal response and responsibility of something bad happens so let's go

back to the example of the self-driving car what happens if a self-driving car kills a person crashes or DOS runs into a wall and kills them that the person that is driving what happens to those circumstances then I actually I probably unique in order not to make it completely but I'm 1 of the few people actually think this is not a big deal and I'll explain why there generally when we're talking about liability were talking of 3 types of liability when it comes to to to perfect like this we have the product itself the car so the person that could be liable if there is something wrong could be the manufacturer of the car itself then we have a service the provider of the service the the software the cell for manufacture the developer that maintains a software the I is the there is some form of intimate connection and the they're liable for the for the crash and the USA it could be possible that some circumstance a and the user could be neglectful as they forgot much something they they Montaigne the car properly and side to the software's complete l dated anything that there's a road when there is no role don't think think that of all the vise versa so that actually could also tell us that there is liability there and possibly the liability therefore is going to be very dependent on the facts is the manufacture fault is a software provider at fault or is the user at fault or it could be neither of those as well there is a very strong indication in the case of the war were crushed and the police state even if though you have a person driving that car fully awakening and fully that's a really good reflexes probably they would steal in this this person would still would have died because it was just too fast even for the computer to operate now that is that could be another

possibility when it comes to smart contracts there is a very similar analysis you know the product service and user blocked we have an added level of complexity because a lot of the times this smart contracts are going to be completed distributed so if there is a fault sometimes you don't have any want to sue or or or someone is copying and pasting code to this happens a lot in smart contracts that people are lazy and they are just copying of the code from someone else and this may be very difficult to identify someone there to sue their who cause the bug can they be identified as the recourse the it can you actually get your money back in in the case for example disparity wallet and and the answer to that is we don't know yet to this is probably an area that is ripe for regulation in the

example data mining that I was telling you about and we have a very different type of liability we haven't liability that is going to be copyright and copper liability arises when someone has copied that war war do work done 1 all the acts that is exclusive to corporate donors for example use a substantial part of work without permission and but there has to be a Council connection between the work that is there the original that is being copied onto copy itself and sometimes I think that with some field that is particularly with their data-mining I don't think that is going to happen if you systems like the United States MOS derivative works are going to be considered for use in is going to be fine so is going to consider a transformative use and that and and the resulting song they is going to trust Ford's so much that is going to be fine but in some instances particularly in some jurisdictions like in Europe in general and the UK specifically remember Brexit that may be a different answer

doing a new system and this is probably 1 of the biggest questions that we have right now I then to and I'll be happy if people disagree this is that the very early stages of research that existing metadata and the negatives worse sorry lost particularly particularly an our fit-for-purpose out see taught in common losses stances the at a extra contractual liability sure in Germany you have very similar systems in the work to law I I the UK now so for example we have a concept and taught of of negligence as you call approximate count the cost where an event that is sufficiently related to an injury that the courts being the event to be caused by that injury so there is no connection and off-gas coastal collection connection to can yes that event was caused by the negligence in this circumstance that is going to be a potential liability under the to automated as a criminal liability in cases where is potentially criminal liability is very difficult to locate any liability because obviously in case you wrong by kills you there is no mens rea chances are there's bird bird there's break difficult and that the the roma wasn't intending to kill you for example so it is very difficult to prove criminal liability that little the negligence has to be criminal level and that's usually very high we may need to overhaul contract law to accommodate things like smart contracts maybe this is something sometimes it what it was actually filtered there are some idea ages and ages ago and when we stopped at 1st seeing autonomous agents in e-commerce there a couple of researchers have proposed that we should reduce the maybe the littler roman laws slavery and said was none of it at the sun and the slaves do not have personality and there is an enormous corpus of the roman law that is it designed to deal with people who have no personality but I actually have autonomous actions a concluding there are things that

we know we don't know the things that we don't know that we is to generally 1 of the spirit of my guiding principles in life I I don't think that we know enough at about some of the challenges posed by Artificial Intelligence these agents are already out there that's 1 of the reasons why I prefer the term autonomous agents in in many ways the agents are already among goes and we can already see some of the potential liabilities are going to arise from this but maybe there are new challenges that we haven't thought of so that's why I would ask you to sort of just keep your mind to your mind and your eyes open to new challenges and thank you very much show

just quality and defined it in person we have 5 minutes left for questions kind of singing there really on the right side yes hi honey as I you know works is to be built me on the UK lines getting that right here at the low dose not contracts for example may call for an international law because you have to be answered by a UK long yet I think smart

contracts themselves are are very very interesting there are deployed already you can you can find them they're everywhere on the blockchain particularly in view of Theorem blockchain this view is now thousands of thousands of these contracts and yet we may need a new legislation I think that that even looking at this from the UK law perspective I think that we have that they pose a few problems that that that we haven't answered in legislation so specific is my contract we may need to revisit at international law and we don't have an international law of contract such the the principles out there that we have at any e-commerce directive that actually allows it that the existence of this contracts because we we last fall the formalities we conducted electronically but yet they're at the aspects here that

the the fact that the law of the decisions of the execution of the contract may not reflect the we'll all the parties but is something that you cannot actually an action in a court of law probably yeah is going to necessitate at a revisiting of of what we understand that is the 1 area that I would say probably our most in need new revisiting again

at I completely agree even on and 4 but I am originally from a civil law jurisdictions and and it's even going back to my Google still codon in in Kasereka which is I think that you would you would be challenging for for that year I hope that the answers things when the piston and another question you mentioned that a lot of problems that can be covered by current legislation already color but isn't there a tendency that so there will be a different balance between the legal problems topics will given example of that sorry for self-driving cars but that year it it will probably apply to other as the OS Bulgaria right no and if you if you've got a crash between 2 cars or a car and a pedestrian and you typically have a situation where you have to individuals who have a conflict and then you have of light to that of a if you if you cover that to say a self driving car you when you have a manufacturer there now and you have a product liability case and you have a kind of imbalance of power because of the manufacturable proved that and and and the Standard case the system behaves properly so we assume that the system did something right so maybe the pedestrian something wrong and a a lot of cases less related to solving the theoretical legal problem but more like who can prove that the that that really the claim was correct or was right and I absolutely agree and and I

completed take your point point that there may be a a big imbalance between the parties particularly the large manufacturer and the small pedestrian I there's is a given a black and a mere episode of that that the self-driving pizza corridor run so little that a guy in this city in at that such etc. at the I think that most well I tend to agree with that most cases I think you're going to be solved by inches so you would take your assurance and as the driver you take your assurance that the company has its insurance and is going to be a battle of insurers like most precious nowadays not yet search yeah yeah and no no I I

agree so in cases of of a

pedestrian is the same situation right now and your a pedestrian bodies on short and your fighting against insurance company of the whole of all the car driver so it's it's already that imbalances there I think that is not so much of an aspect of changing the law but more of that eating steak access to justice a very basic elements having clinics and for people that do not have enough funds to have access to legal to legal system it's already of a problem because I don't think it's a problem of changing the Robert actually locating proper resources to it would you should go to protect individuals with did you have this kind as for yourself and your inputs and and I'm sure of is that you have a little more

time to continue talking with him after having an after its and we would just take some minutes to change the stage and then reboot continue talking about robots see in a minute the if

we the