the town of paradise. their bodies were found in burnt—out cars. tens of thousands of people have been forced to evacuate their homes. yemeni forces, backed by the saudi—led coalition which is supported by the us and the uk, have launched a major offensive to take full control of yemen's port city of hodeidah. aid agencies say it could trigger a famine. britain's prime minister theresa may and the french president, emmanuel macron, have laid a wreath at the thiepval memorial in france as part of ceremonies marking one hundred years since the end of the first world war. the of the first world war. two leaders honoured more 72,000 the two leaders honoured more than 72,000 men from britain and the commonwealth who died at the somme. those are the papers. —— headlines. in about ten minutes, we'll have this week's edition

of newswatch, but first on bbc news, click. the news. we used to trust it. if it was on screen or in print, we believed it. but a few years ago, the lies online started to look really realistic. phoney news websites with convincing—looking stories, shared by your friends on social media. the fake news era was born. and once we'd all become aware that fake news was a thing, the term spread and

the meaning blurred. one of the most chilling examples of how convincing fake news could become is deep fakes. that's the term given to artificial intelligence techniques that swap people's faces in videos — seemingly seamlessly putting them into situations that never happened. we first covered the phenomenon non earlier in the year when user—friendly deep fakes app made the technology freely available online. we all share the same home. it seems the ai genie is out of the bottle. for a lot of kids the doors that have been opened to me aren't open to them. we are yet to see deep fakes make an impact in real news, but you can imagine the implications

if we can't tell what's real. the new tools developed by darpa's media forensics project claim to be able to automatically spot ai forgeries. one big giveaway? they really blink. another potential giveaway is to look for signs of life — literally. subtle changes in skin tone invisible to the human eye that can reveal a human pulse. currently deep fakes algorithms can't consistently replicate these subtleties. but as the technique develops, how long will that last? so the future of this tech definitely has some murky possibilities, but we are starting to see some genuinely useful applications of it too. here's lara. yes, there is definitely a dark side to fakery, but there are also some exciting possibilities. we have put a new algorithm to the test here at the bbc, with newsreader matthew amroliwala. today he is presenting, well, his own news, in more languages than he can actually speak. i am second—generation british. my parents originally came to the uk in 1959.

ijoined him to watch the magic unfold. what did you think watching that? ijoined him to watch the magic unfold. what did you think watching that? it's incredible, actually, and unsettling. because i know i can't do that and you see they've made me look as though i can. for this to work the lighting and the camera angle need to be just right. i can't speak any foreign languages... this isn't video editing. the footage is broken down into data, with neural networks tracking his lip movements as well as those of voice actors who are speaking the same words — in this case in hindi, mandarin, and spanish. now comes the trick, because once the system has mapped out and understands how the mouths of both matthew and the voice actors move, the software can switch these over, manipulating matthew's lips

to mouth of the translated words. this is the brainchild of london based start—up synthesia, a company dreaming of making affordable hollywood—type special effects available to the masses. although we tested it here in the new setting. it currently takes a whole day to create a digital model of a person. of course, the aim is for this dubbing to be possible on any video, regardless of how it's being filmed. although in a world where that becomes simpler even the company itself can recognise the implications. so regards to trust and videos and photos, and what's going to happen in this sort of space, i think photoshop was released in 1990. and since then it has become very easy to edit images, you can remove objects, you can edit the background, all these things that is done to most of the images that you view on the internet or in

magazines today, right. it is going to happen to video, i am convinced. i think humans will adapt to the fact that, just like we don't take photos at face value, we can't take videos at face value, necessarily. so while the possibilities are exciting, we mayjust grow a little bit more suspicious of everything we watch. i'm using some technology that we first saw a few months ago. a speech synthesis system by lyrebird ai. just by listening to a few audio samples of someone talking it can reproduce their voice digitally. like this. i am donald trump and i think that my digital voice is quite impressive. the lyrebird ai has been trained on many, many voices. and it's taught itself what makes each voice different. now that means that you don't have to record every phoneme,

every single sound that your voice can make. amazingly, it has found a more efficient way of sounding like you. the kind of algorithms we are using, it's something called deep learning or neural networks. something that makes these kind of algorithms special is that you don't need to give them specific things to look for. and so this dna of the voice, we know that you are able to synthesise new voices based on this and they will sound like the original voices, but we don't really know what is inside of them. so it's a beautiful black box. version one was trained on american voices, which is why it's synthesised worse as a slight american twang. but now i'm using a prototype of version two, which has been trained using spanish voices. and this is the result.

is this notjust the same as taking what i am saying, turning it into text, and then putting it through an online translation tool and then getting the resulting text and putting it through lyrebird? so not exactly. for instance, there are some words in spanish, strong ‘r', that are not common in some other different languages. so we could make you pronounce that sound, in your voice, even if you were not able to pronounce it before. why have you added this translation? imagine if you could do movie dubbing automatically. if you have a course on mexican cooking you could have it spoken in english or in italian or many other different languages on the same time as they are being released. lyrebird has already used its tech for good, banking the voices of those with motor neurone disease

so they can still use their voice after they lose their ability to speak. what motivates you? i want to make my best effort into preventing that this kind of technology is misused or is used for other or some people or it's used to steal the identities of people or create political instability. do you think it may be possible to deploy deep learning in your network's artificial intelligence to spot the fakes? definitely. this is something we are working internally as one of the potential preventative measures of this technology. however, i think that the problem with this is that, long—term, is that the generated and the real will be basically impossible to distinguish one from each other. and so that's why i believe that the solution, the ultimate

solution to this kind of issue is educating the public and letting them be suspicious of the media. it's 100 years since the end of the first world war, which makes this remembrance sunday an even more poignant and special day. one of the many commemorations taking place around commonwealth nations is being tested here, at the royal chelsea hospital, a retirement home for british army veterans. it's called nothing to be written. it's an immersive vr experience based around sending and receiving of the so—called field postcards that soldiers wrote during the great war. it has, sort of, parallels to text messages that people can send now. that thing where there is not a lot

of message that you can put into it, you can't tell the stories, but it is just a connection to say i'm ok, i'm thinking of you. i thought that was really beautiful. a century after the supposed at ‘war to end all wars‘, this is a highly emotive insight into what it was like both at home and in the trenches. there are no visuals here of the brutality of war. these guys have already seen it too much of that. absolutely incredible. and it feels cold. the sun has come out, right? this is absolutely amazing. one of the things about this session is that i was looking up at the sky and the colours of the sky. i thought that was brilliant. i take my headset off? that was so realistic.

you were there in the trenches. what they did, what they wanted to do, and that is to protect the country. many of them lost their lives. and i hope it has gone to people now these days what it was all about. very emotional. because i was with them. hello and welcome to newswatch with me, samira ahmed. brexit campaigner aaron banks under investigation by the national crime agency, so should he have been a guest on last week's andrew marr show? and did a united states obsessed bbc get carried away by this week's midterm elections? first, president trump's relationship with the media has always been fractious, but occasionally it breaks out into outright confrontation. a news conference in washington on wednesday was one such occasion. i think you should let me run

the country, you run cnn and if you did it well your ratings would be much better. let me ask, if i may... that's enough. if i may ask one other question are you worried... that's enough. pardon me, ma'am... excuse me, that's enough. mr president, one other question if i may ask on the russian investigation, are you concerned that you may have... i'm not concerned about anything with the russian investigation, because it's a hoax. that's enough. put down the mic. mr president, are you worried indictments coming down in this investigation? mr president... i tell you what, cnn should be ashamed of itself, having you working for them.

you are a rude, terrible person, you shouldn't be working for cnn. jim acosta, the victim there of a presidential tongue lashing, and we heard a lot about donald trump this week and about tuesday night's midterm elections in the united states, not all of it was welcome to everyone. including steven blakemore. whilst i do understand the importance of this week's midterm us elections they have rather eclipsed other news, particularly from europe. for example, by comparison how much coverage of the devastating tomorrow for example, by comparison how much coverage of the devastating storms and floods in italy? time and time again, the bbc will focus on what's happening across the atlantic, and follow stories coming