local i spender might be less of a risk than going to the beach. well that's a news update at this hour up next our technology show chef takes a look at the world of deep fake videos don't forget you can always keep up to date with all the latest news on our web site at w dot com or you can follow us on twitter and instagram to actually do tell me your name is i'm directed been in berlin thanks so much for watching. in the light of climate change. conference today. what's in store. for the future. for the major cities to go to insight. into. how the virus spread. why do we panic and

when will all this. just through the text and come in and we couldn't read your show is called spectrum if you would like any information on the clone of arash or any other science topic you should really check out our podcast you can get it wherever you get your pod cast you can also find us at dot com. science. can you spot the fake deep fakes videos fake a heist in deceptively real and they're a serious threat to society in order to expose deep thanks to experts are embroiled in that battle in the name of science while one of them fakes the other on masts we met up with both of them defaced our topic today unsaid.

some defects are funny others are really mean it's fascinating how researchers can manipulate videos with the help of artificial intelligence i tried out what it was like to control someone else's face. take me out as far as johnson. personally. or i do you know. it's pretty impressive to see how easy that was with the right technology and in real time to i don't even want to imagine what could happen if a fake president declared war how can we stop fake news or deep fakes that's what scientists are asking themselves to we met the 2 most renowned the fake experts in the us their academic competition is advancing research on the issue. let's consider how lead to be the best d. think artist in the world the german computer scientist teaches at the university

of southern california and manipulates videos for science you know as hard as it sells we are also trying to advance it in order to be able to detect them right because if you don't understand how they're being generated if you don't have the capability of generating realistic you face there is no way to effectively detect those either his opponent honey fareed uncovers deep thanks a professor at the university of california berkeley he's a world leading expert in digital forensics. i think the biggest obstacle in detection is that the field is moving so quickly every 3 months you see new advances in the creation of say content so our adversary if i can call them that is not static they are changing and they're changing relief really fast so we start developing these new friends of technology and all the sudden there's a whole new way of creating deep fakes and now we have to worry about that as well it's crazy. to see how quickly this technology is advancing where you straight from

photos technically we're all able to alter them at home with basic software but the deals a few years ago moving images were proof that something had really happened deep fakes have taken this to another level there's growing fear that from information is circulating without anyone noticing especially when it comes to big events like the us presidential elections whether it's photos or videos taking is isn't new but it's getting better as both experts warn we've always had fake news from the day of the printing press being developed but huge the difference we this what's new here is not we can create fake news we can create fake images we can create free videos what's new here is the sophistication with which we can do it the democratization of access so almost anybody on the internet can do it but here's the real issue instant global publication the most scary thing is that it feels like you're

stealing someone's identity and you can put anything you one to their mouth and you can make them do whatever you want so far deep technology is used mainly for one purpose porn one study shows that this amounts to 96 percent of the dea fakes online today stars like emma watson and scarlett johansson were some of the 1st prominent victims their faces were super imposed on to pornographic videos with the help of ai the program enabling such fakes was 1st uploaded to the online forum reddit the anonymous user called the fakes was quickly banned but other users took over his software code and kept expanding it today similar software is still freely available online the name is a combination of deep as and deep learning technology and fake but how exactly do defects work how hard is it to make them look real the effect. expert how lee

explained it to me in los angeles. many deep fakes are based on so-called guns algorithms that are on supervised meaning they learn without human guidance instead they're trained with raw data real photos for example. you're not modeling anything you just collecting a bunch of pieces all you need just make sure there are lines like this and you crop the feast red and you're manually you would try to make sure that there's a balance of different lighting condition different expressions and it just learns that by itself this is what is so powerful about the steep learning based approaches. the novelty in this approach is that to algorithms work together what generates under the checks almost like how little and honey for me. it's like an arms race read you have a generator that tries to generate some images in the beginning you know it's just like some random pixels but then what you try to do is you tell instead of telling it where to improve like traditional supervised learning you just tell it no it's

not good enough so you will improve improve again until you would fall the discriminator so that discriminated can't tell any more what a difference is that in itself improves the algorithms work on their own without human assistance. to know why this one minute long video took 4 days to make it's a deep faith generated by how leader and his team. to create it they use face what technology and videos of comedians imitating us politicians. over the years but this is the fake the algorithms were fed 5000 images but there's no secret recipe for the perfect. sometimes we don't know why we add more data and it gets worse so there's a little bit of. black magic and you know cooking over there we've all got photos on social media right i think it's pretty scary to think that somebody could use my

pictures to make a fake video of me that's why methods to flecked the fakes quickly and reliably are becoming more and more important tech giants like facebook and microsoft have launched the deep fake detection challenge with a prize of nearly 1000000 euros here at the w. we've developed a very few cation program to the media which can help analyze their talk content there's also a number of specialist startups on the issue how different is an expert in digital forensics he uses how these defects to find his erf occasional programs and he's always on the lookout for the best way to use a i pray i approach is exactly what you think it is you get a bunch of real videos you get a bunch of fake videos and you teach them in machine learning system to tell the difference but that's a sensitive subject since deep fakes are also created with machine learning a detection tool could just as well be used to generate better fakes but honey for

read goes beyond the ordinary approach instead of just relying on ai he and his team analyze videos to identify people's unique facial features we try to understand how do people move how does the mouth move how does the expression change in part because we think that those are properties of the video that the machine learning can't currently airlines and that means we are less vulnerable to counterattack so any time you build forensic techniques you always have to ask yourself how will this be used to make better fakes this is grid visualizes aspects such as head movement line of vision and facial expressions in order to spot differences honey for reed compares a real video with a video that was possibly altered this method detects about 96 percent of faked footage. one possible use for this might be to integrate detection software into social networks algorithms in the future google has already provided scientists

with some 3000 manipulated videos to help improve the detectors. to make these $3000.00 videos google employed 20 at actors and then manipulated the images with algorithms so tech giants seem to be taking deep fakes faeries seriously and it's no surprise experts predict the damage caused by manipulated audio and video files could range as high as nearly a quarter 1000000000 barrels and 2020 and what's worse there's no perfect program to help us detect the fakes yet but every new technology has a good side to right the fakes also offer f.-u. exciting possibilities for the future i magine you can have a customized movie where when you download a movie you can say i want this actor or this actress to be in the movie and sort of who it is that sort of. this is a huge opportunity for the movie industry soon visual effects could be much cheaper to produce but so far deep technology cannot create images in the same high

resolution as movies. but it's good enough for you tube videos. which is so nicolas cage because the godfather. to fake technology could also bring celebrities back to life like the spanish painter salvador dali. and the dali museum in florida visitors can even pose for a picture with him. i think one of the most interesting ones is in hollywood studios for creating movies so if we have a movie that is filmed in german and we want to listen to it in english you create a lip sync deep fake where the person now it looks like they're speaking english but super cool like this awareness campaign a 9 languages leiria isn't just any. ever being said he said. being you know what is being. a

little different i'm going to do the right thing to make it the creators unadmitted 3 d. model of david beckham's face wow that's cool so you might be able to watch me house shift in hindi or kiswahili completely in sync and perhaps even life it's just a matter of time before we're able to manipulate 4 k. videos in real time it's just up to computers getting a little bit faster but back to more serious matters people around the world need to know about this new technology how leap presented his face what studio at the world economic forum in davos the most important thing is actually that people know that this is possible right if they wouldn't know because usually be fooled if they know that there are ways to actually create the to manipulation is extremely believable then you know we will have some we have protection but honestly i have to say that both experts hunted for it and how we have managed to call my worries

about defects and little fake news has always been a huge problem not just today but the perfect technology is not even necessary for it. the more people are aware of the facts and the possibilities to manipulate the views the better they can deal with misinformation is something true or false that's not a question of deep fakes what do you think are you scared of manipulated videos or is this all just a storm in a teacup let us know what you think you can do this on you tube too there you can find a video of me getting my fairy own 3 d. avatar and how nice lap and los angeles i even set in the same chair as will smith the hardest part was i had to make 26 different faces take it out on our youtube channel thanks for watching and see you next time.

motorsport is it still a man. no way. to compete in research. and their winning. the power of motorsport expert consensus fastest female race car drivers on my. let's. read. on w. . posters from nigeria you know does what really would stunt. during the. winter. grand successful beyond belief. this is the way we're doing.

it would. be in 45 minutes on d w. percent drama competition. marketing. fight intuition love money. trans fats and. old. live. we are living during the most extraordinary time in history. transport

will go fully electric. the rest of the auto and mobile.