Gadgetory

hello I'm Gary Sims from Andrew authority and some of the buzzwords at the moment are machine-learning ml and artificial intelligence AI and there's a whole question about the ethics of the decisions that are made by AI and m/l systems now we don't actually have any ants at the moment but there are certainly lots of talking points so if you want to find out what are the problems with ethics and AI please let me explain now of course the whole area of ethics and morality is in itself absolutely huge so we'll just put that to one side for the moment and let's just deal with how AI and ML deal with the area of ethics and morality for example in business you can have a business that is ethical or a business that is unethical now the problem is that not everything that is legal is ethical and not everything that is unethical is illegal so for example they might be strong lobby groups that Lobby a government to make the laws in a certain way or maybe a weak government doesn't actually address certain things and therefore a business can function and that business can itself be with inside the bounds of the law however its activities are unethical and certainly one thing we don't want in the future is an AI or a machine learning algorithm that is staying within the bounds of the law but it's pushing as hard as it can to see how far it can go but actually remain inside the law itself in fact when it finds a loophole it will go through that loophole and expose it because that will be its principal gain to fulfill its goal and if his goal is fulfilled through a loophole that's exactly what that system will do well my favorite examples of ethics is the idea of fairness now I have three children and I have raised them together with my wife and we have tried to give them an ethical and moral standing about what is right and what is wrong but one of the things we never needed to teach them was what is fairness they always knew when something was not fair and they were not receiving the fair portion now it was a different thing teach them to always be fair to others that was a whole different lesson but they knew intrinsically when something was not fair now will an AI or a machine learning system know that something is not fair because it might be the most efficient it might be the most cost-effective it might be the thing that reaches the goal quickest has been programmed with but it's completely unfair so how does the ml system know that let me give an example let's say we had an ml system a machine learning system that was controlling emergency services so we're talking about you know ambulances and paramedics and fire engines and maybe even you know sort of first-aid and maybe hospitals and emergency rooms okay and this machine learning system is there to try to optimize the use of all of those resources now that's a really good idea but the problem is that if you set certain goals then the machine learning algorithm will shortcut those goals it will cheat to fulfill those goals so for example if you said that the goal of this system is to treat as many people as you can well if that was the case then maybe the machine learning system would give more priority to people with ingrowing toenails or to people who've got cats stuck up a tree because very quickly the appropriate resources can be given the problem can be solved and then the next case can be taken so the throughput the number of people helped actually goes up quite high the guy who needed a ten-hour operation is put to one side because there are 10 people who can have a 45-minute operation is quick and cheaper and therefore 10 people get treated rather than one and of course this has no ethical dimension to it it's purely a numbers game and you can also get the opposite way around if you say well we're tweak the machine learning so that doesn't do that well actually then you can find that all it does is treat people that need ten-hour operations and then people that are in less risk situations never get treated cuz they're not going to die they're not in harm's way so why should they be treated at all so setting the goals is really really important and a machine learning algorithm will look at the data and look at its goals and it will find the quickest way to achieve that goal and of course made treatment and treating people is incredibly complicated because not only are you dealing with risk you're also dealing with a whole bunch of factors including age and the severity of an illness and you're looking at maybe other secondary issues that affect the primary issues like lifestyle maybe income diet you know the kind of education they've had the kind of way they treat their own body all these things come in to the factors that play in how you treat people now of course a machine learning algorithm could make the wrong decisions that are unethical but according to a dangerous being given it was the right answer and one of the problems is that all ml systems are based on data so the machine learning algorithm actually gains his experience from the data that has been given now if that data is itself biased that it will itself create biased rules and it will follow those rules even though they are biased and we're assuming therefore incorrect or one of the reasons the data set can be biased is because it's missing data so you may have a whole bunch of characteristics and factors that are listing in the statistics but actually there's something that is missing that would actually change the characteristic of the whole data set and of course there are a whole bunch of built-in intrinsics that we know about that actually are not expressed in data so for example the value of life or the value of another person and this brings us to the next question which is who is to blame when a machine learning system makes the wrong decision for example let's say you bring a friend around and you say to your smart speaker tell me a joke and it tells a joke that is not appropriate it tells a joke that is offensive who is to blame for that is it me as the owner of that smart speaker did I have something do I do something wrong to make it behave like that is it the producer of that you know is it the the manufacturer of that particular smart speaker is it the programmer do I have to kind of take a guy called Fred to court to say you know why did you let your smarts be could do that or is it the data set person or the people who provide the data said you they provide a data set that was not a ethical because I was talking about jokes but of course when you multiply this up to be indifferent knows education and health care and business practices self-driving automation all these things then these mistakes that get made can be much more costly than just someone being offended by a joke they didn't appreciate and then of course that leads to the flipside of the question if in the future we have machines that are you know fully autonomous and they are making decisions based on their programming if they commit a crime are they guilty can a robot be guilty of a crime do robots have any rights so these are all interesting areas that are to do with morality and ethics because the bottom line is data doesn't give us except thing that can't be gleaned from just a set of data points I'm going sinful and raw thority and I really hope you enjoyed this quick trip down this quite complicated path about ethics and morality and artificial intelligence if you enjoyed this video please do give it a thumbs up also please subscribe to the Android or thority youtube channel is really good if you hit that Bell icon up there so you get a notification every time we release a new video and last but not least please do go to Android Authority comm because we are your source all things