was welcome on his return home. hundreds of suppose is that greeted the harry board . i'm a guy to come on to the airport because it was told this mountain after helping to overturn a little binding i'm few days from attempting the fates that i for these national lost both legs while serving with the british army. and i found the stuff he's promised to spend the rest of his life helping people business and set you up to date up front of the world news the top of the our next on the w. a documentary, looking at the opportunities and risks that artificial intelligence of present the, can you hear the we are all set. we are watching close. all the to bring you the story behind the new your

own about on volume information for free might do to man the all sufficient intelligence for a i is considered a key technology for the future. you come to new york often mostly for work, but sometimes just for fun. it makes the what is doctor's psychologist to police officers easy a. i'm just expected to make drivers or even telling play is thing is the pause. in every aspect of everyday life. a i could help us make the best decision. should i move the route or the bishop from left to right? shoot. oh, hold my file, you dates on well jane. the relentless logic of algorithms is supposed to guarantee

us in life, free from errors. but lately, even program is happens, sounding the alarm, there's a self congratulatory feeling in here is a, i hasn't lived up to its promise. is it really as intelligent as it's made out to be. # this guy's really good. what are they going to replace us? and what are the limits of a a turing machine like this and on the computer. the device by the english mathematician island cheering was the 1st machine capable of solving a puzzle more efficiently than the human brain was its health to persist succeeded in deciphering encrypted drum and radio messages at the height of the world war 2 of the countless ministry specialist had wretched their brains in vain as much the

doing of the new era with the development of devices that also made worked previously requiring human brain power. lab us and they looked at these issue at the beginning of automation. the goal was to reach juice, physical assets is that is the amount of effort required. and so for centuries, a mill was considered an automated process. you may receive more on up to the time that this approach would be applied to non material mental welcome, which i don't think he could be real. nowadays, we are dealing with a new form of old nation with the issue, which we generally call onto official intelligence on tv shows. so if you see it in the 1950s, this developments extend a research rapidly with a promise of how all sufficient intelligence a i would optimize online. it was supposed to,

dr. called improve education, provide us with healthy to make the best medical diagnoses. i'm fine, just the right words to cheer stop. he says, then i'm depressed much the time. i'm sorry to hear that you are depressed. initially progress was slow. so that all changed in the early, 2, thousands with new powerful mainframe computers, able to handle huge amounts of data. i was at google and i was at google for a long time at that point. and then suddenly everyone in tech, everyone in google, everyone everywhere is like, let's use a i the solve, everything is going to solve cancer as going to solve transportation as gonna solve education. i had somebody pitch me like a i to detect in genocide. right. and i'm like, what the, what like based on what, what are you training it with? what are the political stage no answers to this, right? you level t, i'm well, no,

do you mean easy people wanted to digitize all of reality? if you do, definitely go enough knowing everything in real time, that's the way our dish was. yeah. could further than you saw that was something god like about this idea without taking into account. that's much of reality. just can't be reduced to zeros and ones. you have the rainbow outside, i am programming, you mean you tend to put you up the welding to electrons, cost fluids all the time and machines and now so to be capable of learning by themselves. thanks to a completely different method of information processing and multi layer training, so called deep learning. there was a, a major advance and deep learning systems based on their ability to read natural images. so on the image that challenge alex net one,

the image that challenges would improve the, the efficacy of the learning. and that there was like a catalytic kind of a gold rush. the image net contest is an annual image recognition test for computer programs. t is even the best of them got it wrong in every stud guess. but in 2012 technology based on machine learning was suddenly able to bring the error rate down to 15 percent the full. this breakthrough, everything has to be explained to a program, meant to recognize the face. for example, it would look for shape that resembled an eye, a mouth, or a nose, for instance. and this, the order was right. the algorithm concluded that it must be looking at a face so, so developed an automatic image recognition system program is how to describe thousands of images from all angles and machine language. and that turned out to be

easier said than done that pushed tend to deep duty. i could seek, seen that push in the traditional approach of classical a i. the machine was fanned with knowledge, just perform. but it turns out that the learning was much better, because instead of telling it how to process the information, the work is left to the computer. this is the commode with typing informational. deep learning has its roots inside vanessa. it's an area of research with computer scientists ultimately up to neuroscience for inspiration. the using this message program is no longer described to the machine what a face looks like. instead they all skipped to find out on it. so the system resembles an extensive network of connections that mimic than your arms in our brain. this also special neural networks allows for

a variety of adjustments to strengthen or we can, the signals between the link culminating and an output signal that provides the answer to the questions such as, is there a space in the picture these, i don't want that as the the system, the concession portfolio. one of the advantages of deep learning systems is that they can work directly with the rule material from census. see if it's a camera, the gray scale or in the density of all the colors is measured for each i take. so if you want, if there are 10021000 pixels, for example, the computer process is a 1000000 numbers. the normal to home thought alternate to each pixel 1st sends a signal to the network that varies in intensity, depending on the brightness in so called supervisor planning the machine tests billions of possible settings until it finally gets the on so that the program is looking for and an output signal like

face detected light. what was this combination is found, the settings are not in the learning process, just finished. it measure sleep time, it's the parameters of the model are adjusted so that, and eventually gives the known and expected answer. thank you. so in case something you, what you need, this could i imitate the examples given to them by human. okay, give me example. it's very fascinating. the mathematicians have been obsessed with trying to figure out why it works and nobody is really sure to be honest. why exactly? every said why exactly deep learning succeeded. what makes these neural networks so special is that they can recognize the generic shape of face within the larger image. the machine is trained by showing it thousands and thousands of images with spaces and then until the perfect setting the sounds. and from then on the system identifies all pixel configurations that corresponding to a face while filtering out all of the objects. such systems

can now be found in cameras that automatically focused on faces in video surveillance rooms readers for postal codes or license plates and naps, for identifying flowers or doki and, and the body scanners that airports, researches at the university of michigan wanted to find out how capable of these systems are when the objects appearance is slightly altered, while the system detects a voucher here. a small rotation and it sees and around the time i a this pete, the scientists interest in knowing with a self driving calls might be thrown off by road signs, but have been tempted with the place stickers on the stop sign. and sure enough, this confused the vehicles neural networks, causing them to mistaken for

a speed limit sign. instead. these kinds of areas may explain why machine image processing systems still do not work in critical applications and clinics where automatic read is the being tested. decisions are not made by the a i, there is a left to radiologists and doctors who continuously monitor i'm treating the systems, a system of the system 500. but these are very fragile systems that are only useful when applied to images that are very close to the training data. i don't do any often. so if you have patients of one population, or you use data from one equipment to train the systems, they don't necessarily work when you bring them to a different setting. and humans are, are different. humans have a very nice systems level way of thinking about things. they can think about things that are not in the database. they can think about how the, you know, how the model is working and whether or not they want to trust it in a way that these, the systems by themselves can't,

can't do that. but foot on those, we tend to anthropomorphized the systems and think that a deep learning system can provide a description of what is happening in an image joint description with extra excuse . but we think the model understands what is in the middle a security measure, male heavy to use, but the way the model associates an image with the text is something completely different than when we humans look at an image and describe it with words, flash and a new key about advertises is what i'm all like. oh man. and you put to ship them whose don't really use the moon the couldn't he sold you moon persistent bye. well, this is the systems general knowledge of the world is incomplete by definition because they lack the bodily experience. the vision will experience the move, the connection between the words and what they refer to in the real world is until

we succeed and including the side called such systems will remain deficient, assistance then void this. you saw the humans describe meaning to things through experiencing events like feeding the force of the jewels as they fight down the in sizes piercing the smooth skin. and a jew scratching out and running down the throat, all of which plays the, causing gradually defining worse and naturally for a computer system. on the other hand, it's just the sequence of pixels linked to text to information. off po, a chef i studied, massage, virginia, motors, new haven. how could you use a motion data to build a better product? yet despite the rudimentary perceptual system advances and most of mastic image

execution have revived the dream that machines will one day develop a motion and be able to help us new proceedings. the most secret feelings of our fellow human beings. we can recognise more than 20 scales of affective states, will all sufficiently intelligence finally be able to give us an object of onset about men at least the feelings for the machine. when true monica's head, when her gaze captivated the minds of chris jones, knows in his dying moments, how would such an automatic commotion to take, to actually work the 1st step would be to create a list of emotions from the convoluted infinite variety of, of states, of mind in this sense, the research of american smaller just pull ackerman has been particularly helpful

for program as officers field trip to pass a new guinea design just came to the conclusion that humanity shows 6 universal emotions, which inevitably can be read on all faces, joy and sadness discussed and dying, surprise, and fear. it may be human nature. edmonds theories have even in spite of television series, in which a mazda detective identifies perpetrators based on the micro expressions of the truth the classification is disputed, m. o, sciences. but nevertheless, those is the basis for all emotion recognition computer systems. precisely because of its simplicity the 6 universal

emotions serve as the basis. the next step is then to have thousands of faces assigned to the 6 categories with humans during the selection. that's how the training data for the machines is creations. machine learning continues until the computer system produces roughly the same results as human selective joy softness to light, surprise, fill, and rage discussed. discuss, discuss. what's the best testing is found. the systems could come to kids that program is entering the world use is universal and motion detector. loophole dealing with a motion detection is how it's being used. i. so i go to do an application is within management of human resources over the past few years. service providers have been using chat bots to evaluate the job applications, aggravated because you don't, if you're an employer, you might have

a people being interviewed by a computer. and then you could have the perceived emotion of, of the system being conveyed back to the potential employer. so those types of things make me a little bit nervous if say i was a job applicant, right? and then i have this emotion recognition trained on my face, and based on the top of my voice based on, you know, the way my mouth moves, like one eyebrows, a little higher, whatever they make claims about whether i'll be a good work or whether i, you know, i don't know, have a steady personality, you know, things that are making really deep claims about like my interior life. this is pseudo science, right. this does not work. the really lovely summer new garden justin

for small you made me i can easily imagine that the advertising industry could use such a tool to invest or influence as well. news f girl c is also being used in classrooms to detect attentiveness and students by skied images um have have on flow to the idea of using it to build a lie detector and come. that means you could check suspects with some systems to determine whether a person is lying or not as the one that could ultimately determine whether that person remains free. oh no, i don't know how to insure it, and it's not what you say, but how your site is higher. you help to simplify your process. assess candidates based on science optimization content to maximize the moment by moment emotional engagement by target population emotion recognition systems. awesome, combined micro expressions and tone of voice, but they're analysis but when it comes to men, elisa monica jones know the motion detectors from google, amazon,

and microsoft only reach the same clear conclusion. they felt absolutely nothing. the i'm since artificial intelligence has no taste, but it has no idea how delicious pastry kind of folk memories of a deceased on it has never felt what it's like to have a gentleman rushing through your frontier. he is willing up and you are you. so you knows, runs not afraid of anything. it doesn't get goosebumps, news, neither pain or pleasure has no own opinion. no not stretched off and carries no repress trauma. in other words, it has nothing of its own to express. sophia, a superstar among humanoids, roosevelt, nevertheless, seems to prove that machines kinda quiet is the ability to speak. i'm a big fan of their sho west world. and i have

a couple of ideas for the next season. the but so feel doesn't know what she's talking about. from her own experience is really interesting conversation spring from her programming and input from the conversational partners. thank you for inviting me. i am thrilled and honored to be here at the united nations like here after you and participating in an event to promote technological development. okay. another question i have for you in many parts of the world, people don't have internet or electricity. what can we do at the u. n to help them? the good news about a lie and daughter mation produces more results with less resources. so if we are smarter and focus on when, when type results a, i could help efficiently distribute the world's existing resources, like food, an energy. i assume the relationship between sophia and artificial intelligence is

something like the relationship between a magic trick and physics research or physical bio. it's been helpless, bundled, and also told, impressed by sophia. i think it's only a bit farcical. new mode is made by how many people are willing to ascribe something like intelligence, emotions and the like is over to this at the right to see what so, and the extent to which they're willing to play the game. i have someone who i actually consider charlotte as you. my name is i me. so i'm charlotte on how to the humanoid right. adults can be seen as the embodiment of a dream held by an industry obsessed by parent human logic with computing logic. the, but the real strength of us official neural networks might not be,

it's not for mimicking us, but rather it's capability to assimilate vast data bases and to sort through the analyze and derive correlations. that can help us better understand complicated phenomenon. such as the ground states of elementary particles and the quantum magnetic fields. all the factors of ad prussia and humidity in the formation of cumulative numbers. clowns escaped feler. i'm going to note denying that there are many tasks when machines and how performed human english type. but no human can take a very large database into their head and make accurate predictions from it. it's just simply not possible. so there's nothing new to say i is becoming a scientific tool for research, as in many fields, including health care and then on the on demand that has something really any kind of data that people can generate on a massive scale is really an area where machine learning and how it's precisely this ability of mainframe computers to detect and risk factors for,

seizes all the effects of a new drug. now by scanning huge patient data bases that explains the tech jillions possible for the data from the extremely lucrative health care sector. we'll go to them with the google used to be a search engine 3, but now it has a health care branch. do the same goes for apple says yes, and no one is asking why quote, it has nothing to do with the original business model. so this is all connected to big data. now you see let me it's all started when 2 young computer geeks and bench at the search engine that most of us keep using to quin soft this for information. ready ready welcome to google ads. now that you've come on board, it's time to jumpstart your businesses digital journey and grow the invention of past. and as long as that's ton, google's found is into 1000000000 as well. the algorithms choose use the data and to tell them will biography. these algorithms are seek or just the recipe for coca

cola. all that is known is that they are based on to cleverly applied processes. the 1st of which is profiling. well known information about the person is brought together to create. they use the profile intimate searches, online purchases, streams, videos, messages, sent, and places visited. provide increasing the precise clues about what products might interest a person. the 2nd process is mapping, which involves grouping uses with the same preferences to get the just about a year from the moment to use, it begins to browse the web, jumping from one page to the next, the preference. so everything from music to politics and shoes is cassidy mapped out these connections squared clara as more use this makes similar choices like a far as poss, becoming more visible over time and not understanding as millions of uses

create the actual communities i'm. so is the logical maps from which digital platforms did use other products shop? it's might like a month on it. on a visa, you collect a certain amount of information about the pass and once that's done and then try to predict how he or she will act in the future based on how they have acted in the palm of in that so close to 100 and it always, it is using data from the past to understand the present and future based on a model created from the past. to understand the effect of these algorithms, researches that boston university invented ads and distributed them through facebook's own advertising platform. it turned out that 80 percent of those who were shown in natural country music, calvin with whites,

well 85 percent of those targets. and for some of the ad for hip hop, what black, fictional job i've got similar results, 90 percent of those targets it to embark on a new career as a lumber, jack women, but 85 percent where women when it came to a supermarket, job just a 75 percent of those just to an ad looking for new taxi drivers. what african american social psychological analysis tools to targeted advertising are aimed at probing or stereotypical behaviors. semester explosion does reinforce them . but the same tools also form the faces for systems that analyze all behavior as a guide or collective and personal decisions such as dancing upsets to just pull those up this natural profile software that helps banks and insurance companies to identify the type of people who might not pay back the low boyfriend tag,

the asked to go risk technology support use throughout the credit, decisioning workflow. a programs that pinpoint dangerous areas and calculate the probability of a crime taking place on behalf of the police. what's good sat up again from a chair. okay, where the advertising spot shoes or projects in crime space systems make general statements based on the day. so they have been trained with get through the, the mostly system. there is always a bias with the system as well as so made no click on the show. and one reason for that would be biased training data. it would be easy and sleepless night by something. take, for instance, system that looks at surveillance videos, easy to use, and tries to filter out suspicious people interested any time in the ask you to attend any juices to develop such a system system. you would 1st have to ask mr. smith, to watch surveillance, videos, identity, and decide which people look suspicious for, and, you know, i was curious, cassandra just just pupa and i'll come if it's something else. you know, really,

of course, the outcome is then based on mr. smith, judgement, which may be by adult signatures. you thank you and as long as it is mr. smith who is talking points, you're aware that he is expressing his opinions. but as soon as you use his conclusions to train a deep learning system to the bias is no longer on the roof. q some kind of look at text, socio new context, secretaries usually context, the social psychological and tomorrow the context will remain incomprehensible to computers. i think he has to come concepts. both of these things, systems like judgment is on what did they base their decisions? no monk. so it says use mom, but on the basis of statistical missions for the bus, the just that just to just demo. and those all studies have proven lead to racial and gender discrimination. but 7, that didn't seem not shown to just come national washer. and these commissions, all you work on the front lines and the criminal justice system, you make decisions every day that impact community users. that's why the north

points sweep in a dozen us state beside go square in feelings, judges during trials about the risk of a defend and re offend. they help you reduce risk and recidivism at every turn or keeping safety. this machine learning system has been trained with police offender grace a very sweet, sweet, and analyzes 137 criteria, according to his secret formula and delivered its magic to the judge in the form of a short summary. the investigative journalist compared the results of $7000.00 people and the soft with predictions with what actually happened and subsequent is only 20 percent of predictions. for serious crimes proved accurate. most in cases the, the journalist found the predictions of recidivism for black people were much higher than what turned out to be the case. well,

they were too low for the white demographic. so if you, you have a racist criminal justice system as we do in the us and you know, in much of the world you will have anti black bias built into the data emissions. there's not even the ultimately, the machine only provides a potentially fallible. and in perfect assessment of the person issue small. so in a way the bias is being whitewashed, more a bit like with money laundering noise when the actual, i don't know the, so how do we handle this more than die? lemme should those who are given a choice roth of trust machines or human justice machines might not compass the lease. humans are often not the best at math. and we can be emotional and mature, sleepy, lazy, rebellious,

fun, loving, overbooked or even completely delusional. to, to take those, he said, call the all of these technologies fits perfectly with our quote unquote fundamental human laziness. could all sit the jo took the moist to heat because in today's world such systems offer as a convenience by taking over part of her daily chores. or for the whole shocking and thought, you know, it's good to do or the challenge right now is to take control of our individual and collective destiny systems are doing just the opposite. it's been many areas of society shoulder. that's what you think is organising system work. ok, thanks to mine your program, love fine or we're fine. you're going to find a perfect match for you garden, originally intended as a memory age a i is now making recommendations. and even though symmetric decisions or, you mean it is going to analyze data from all of your money to, to find my son,

me. hi, guardian, you have 6 perfect matches, the noise you were gay. neither. meanwhile, machine learning systems with the billions and billions as possible settings, also complex, but even the program is no longer understand the criteria on which the machine is facing its judgements. a time has been coined for this phenomenon. blackbox a black box machine learning model. it's a predictive model that is either so complicated, the human cannot understand it or it's proprietary, which means we have no way of getting inside and understanding what those

calculations are. we need, if it is trustworthy, it's boost, especially with deep learning systems. it is not clear how decisions are made since only the results are visible. sam crazy. there is a movement where people are saying, well, we can still use black boxes. we just need to explain what they're doing. and so i've been trying to kind of be beat that back down and say no, no, no guys, we don't need black box. you can just launch a black pack. so i have 6 decision. you actually really need to understand how you know, how these predictive bottles are working in order to make better decisions. pop out the car behind user friendly interface is there is the close decision support system. the all the com that com are definitely happiness as not to exhort. apart from that, it's well known. the company is liked to avoid responsibility by saying something

like, are we all dealing with a very complex system here? and it was on top. it was the algorithm via that kind of reasoning or excuse is of course, completely unacceptable acts at tablets. it's a very good way to sort of evade the responsibility and make difficult decisions that you may not want attributed to you as the machine. a young fellow. so the fundamental value of the funds is true is freedom of thoughts move, which can be traced back to the enlighten. the kicker shows the said the most like now trade offs are being made hard to do. of course he does. we have delegating our decision, making the decision that i'll give you just sort of the short sea level to the i g . the underlying goal is to prevent any error to 3 to the but to do this, we are under the control over to systems or the system to phone number. the, the sales driving car has known been the poster child or vaults. official

intelligence epitomizing the stream of global to mason. subbing humanity by making the decisions for us all the while keeping us safe and relieving us both of pressure on potential road rage the despise investments of almost 100000000 euros. new self driving call has yes been, and out into traffic outside of the test track. without the human driver ready to grab the wheel at any given moment. success i see that it is very easy to use, deep learning to make an unreliable prototype for some things. very complex, like driving me, sits goods just but it is very hard to develop this prototype further so that it becomes reliable and powerful enough to be practical, similar in traffic for example, in particular, which you just send it, what's buying something seriously, stay in the main the police show that or i cannot make political or even technical

reasons, only it is the system itself that requires products relying on artificial intelligence to be brought to market while still not functional around the policy. functional and html function is c. so it cannot be the silicon valley, they say they can't tell you, make it, it's useful to have some blown young rules. some of the vi you have different tended words until it finally done. um, do you have more money? it cost them over at that time is constantly being pushed back. so could you see costs, artificial intelligence never reaches a level of performance when either makes humans disposable piece of i see does it because, you know, the predictable failure of a ton of them is driving, has to build a new profession, human, the system to machines in this company, for example, trains employees to take control of not quite a ton of this vehicle, the in just 10

years. and then tell you, industry has sprung up around all sufficient intelligence assistance. hundreds of thousands of workers around the wells per pet data for training of machines, checking and correcting the offices will simply replacing them when needed the this new form of human labor behind so cooled off the official intelligence systems was invented by the 1000000000, the founder of tech giants, amazon just visuals when just based on announced the launch of amazon mechanical and tuck in 2005 could be made no secret of the fact that it was a project for artificial artificial intelligence sufficiently. that is to pretend artificial intelligence will say that even in this case, and humans are doing the necessary manual labor and so to speak about to make these

algorithms was fair functioning. physically, amazon mechanical tuck is a crowd sourcing platform that addresses this apparent power tungsten. the growing prevalence of ultimate decision making systems and the inability of all sufficient intelligence to be truly autonomous. the, the platform is named often 18th century electronics and the so called and mechanical touch with a human hiding and it space. it literally report refers to hidden labor, right? it's hard. so yeah, they, it's like, they do say the quiet part loud, often when he 1st actually just jump out look so, so that we're looking at a paradoxical situational. the 5th soup, on the one hand, people are being asked to do what robot, so automated process is a not capable of doing that because you know, mean useful while on the other hand, pretty well because of tasked with activities the give them little wiggle room maps

and much less autonomy than before, one to little bigger than the bussey. open the par bay doors hill. i'm sorry, dave. i'm afraid i can't do that. science fiction has to drop off the computers could become so intelligent to know that they succeed and dominate one. i'd like stanley kubrick's dangerous. how this mission is too important for me to allow you to jeopardize it but currently a very different sentiment is taking hold. it's not so much that the machines have become so intelligent that they as i'm a nation, humans. but rather that we are gradually submitting to the standardized logic of the machine in call centers like this one, employees have to adapt to an algorithm and are increasing the being monitored by

all sufficient intelligence. software identifies keywords and the conversation to ensure instructions are being followed emotions of the customer and the employee. and the nice in real time. anger is detecting the system directly prompts the employee to be more empathetic. at the end of the conversation, employee performance is race. it and if it falls below a certain school, they are immediate defiant. is everyone i know do we have we have this thing with don if an age in which multitudes of people are forced to adopt the way to the dynamics of interpretive systems so that you don't need the system to apply to achieve the system. i systems designed to maximize optimization and productivities. you have to go for direct tv to allowing no room for negotiations to take a show. lee is on football. in reality,

the amazon warehouse is dystopian 9 to mask the games. wherever employees have to try to keep up with a partially automated system system, a coupon to a mobile easy the in the vase tools of distribution centers, the computing power of all sufficient intelligence ultimately seems to us and this will help when it comes to understanding the complexities of the real world and making the best decisions the, the work done by humans, subject to the come on this machine. early calculate the optimization of the flow of goods. they are the ones setting the page, the

easy next. what we've been introduced to being robots of flesh and blood sugar, the sol. see who's the most important wire in did the, in my opinion, automation is a track me under control and no mist masses of people to the messing over the best one, you know, most can make it to freeze. my biggest concern is not computed bugs, but people who say can power to want to control others and who increasingly have access to very powerful technologies is present px. yeah, detectors, g type results, there's always, you know, a hopeful outcome is inevitable and i think it's gonna take work, right? like hope is a, an invitation not a guarantee. the

tony, the my good. the to me come on talk to me or the a pulse, the beginning of a story that takes us along for the ride. it's about the perspectives culture

information. this is the the news d w. mine's the 70 years, dw, the

this is dw news, live it from berlin tonight, attacks against are russia. moscow says that it has driven a group of 7 tours back into ukraine. chief has denied any involvement. could these fighters could they be russians also coming go signs of a breakthrough in the decade.