this is bbc news. the headlines: china is seeing a sharp increase of positive covid cases, just 10 days after suddenly easing its strict lockdown measures following a wave of protests with capital beijing being one of the worst affected areas. in shanghai, officials have ordered most schools to take classes online. the award—winning iranian film actress taraneh alidoosti has been arrested after publicly condemning the first execution of a protester in iran earlier this month. an iranian news agency says ms alidoosti was detained for publishing what it called false and distorted content and inciting chaos. leo varadkar has become

prime minister of ireland, returning to the top job as part of a power sharing deal between the country's two leading parties. he's taking over from micheal martin. the pact brought together the two rival parties that have historically dominated irish politics. now on bbc news, click. this week: computer, draw me a picture of a lizard drinking coffee. nice! yes! we're looking at the light and dark side of ai art. oh, dear. oh, that is peculiar. laughs. zoe's finding out why everyone�*s talking to a new ai chatbot. "are you alive?" please say yes,

just please say yes! alister gets his ultrasound on. and lara's getting ideas above her station. impressive display albeit a little intimidating to see myself that size! do you like a bit of art? i do like a bit of art. any idea what is art, and what isn't? that's a big question. do you like a bit of art? i do like a bit of art. any idea what is art, and what isn't? that's a big question. i guess for me, it's anything that aims to look good or even make you feel a certain way. and that could be paintings, photography, music, performance, even ribbons. yeah, ribbons is a work at the now gallery in london which invites guests to re—examine their views on textiles. did it work for you? ijust wanted to touch it. lovely. but for me, the most exciting piece of modern art that's been

created this year is this. 0h, what's it called? a cute corgi in a house made of sushi. right, and who's it by? now, that is the million—dollar question. in the last few months, all sorts of weird pictures and videos have been popping up online. some look like clever mash—ups of existing photos. others look like the work of pretty groovy digital artists. but they're not. each of these images has been created from scratch by a computer, based on nothing more than someone typing in a few words to describe what they want. now, i have to say, some of them are stunning. the pace of this has been quite staggering because, you know, a! is fast moving anyway, but even by those standards, this has been as fast as i've ever seen.

dr michael pound is an expert in deep learning, and he spent the last few months exploring how these systems work. what you could do is you could just come up with any prompt you like and we can just generate an image and see what it looks like. but, as we try and generate completely new images of... "cute rabbit wearing a parachutejumping out of a plane." it becomes obvious that these systems don't really understand what they're drawing. oh, dear. ooh, that is peculiar. all right, well, that is very believable. what's happened there? it's kind of like a cloud, but it's also furry like a rabbit. yeah, for every great picture you see online, there may very well be quite a few weird ones that don't get posted. but the good pictures are really good, and the fact that this has all happened so fast has blown my mind. so what on earth is going on? to put it simply, it's complicated. but if you give me 90 seconds, i will give

you my best explanation. for years, machine learning systems have been vacuuming up all the images they can find online and their accompanying text descriptions. over time, the systems have learned to associate certain words and phrases with certain types of pictures. and now they seem to know enough to be able to work in reverse — start with a description and get a picture. so if i type in, "an abandoned futuristic dystopian city overgrown with plants, bathed in sunlight", it will first come up with this.

now, it doesn't look much, but i think there are some clues in here. for example, does it know that the word "plants" means it needs to put in a lot of green, and that green needs to be low down in the picture? do all pictures of sunlight it has ever seen involve a single point of brightness somewhere in the sky? i think this is some kind of average of all plant pictures and all sunlight pictures. next it uses these broad brushstrokes to guide it, and it slowly adds more and more detail over and over again until, in 100 steps, as if emerging from a fog, you get this — which i think is pretty darn beautiful. and you don't have to start with a blank slate. give it a starting face and you too can grow your own mad scientist in a

steampunk laboratory. and then there are the teddy bears. in the same way you can specify a subject, you can also specify a style. these systems have seen enough monet to know what makes a monet, hence the most beautiful picture of teddies that monet never painted. oh, and here's van gogh�*s bears, picasso's bears, and even one from hp lovecraft. but here's the problem. what if you wanted a painting in the style of a living artist, one who earns money from their commissions? as an artist, my work, what i do, it is a part of me. it's a part of my identity. and these companies have allowed users to generate things that look like i've done it, that look like my peers have done it. and we're going to generate all kinds of things that you never agreed to,

that could either be against your wishes, that could be against your beliefs. karla ortiz is a concept artist, illustrator and fine artist who has designed for the marvel cinematic universe, amongst others, and she has discovered ai—generated work in her style and that of her peers online. i would even say that to me it feels almost like some kind of parasitical copying, you know? laughs. it's like, oh, we're copying from another artist specifically to look like that artist. oh, but it's slightly different, but it sure looks like that artist did it, huh? it's concerning. this is a live issue, but there may be hope for artists whose specific styles have been replicated. an artist wouldn't be able to claim an infringement for a style being copied, because copyright doesn't protect a style. but when artificial intelligence has been used to make something in the style

of someone else then their original works would've been copied by that al. so if an artist did have a problem, they could make a claim for their original works having been copied in the first place as part of data scraping and data mining. but there is that other issue surrounding ai, too. given that you make your living from your art, are you worried that you might lose income because suddenly it's possible to generate stuff that looks like yours? i'd be lying if i said no. i am worried. i have peers that are also very worried specifically — and this is a concern i share — for those that are entering the industry. and i've heard reports of illustrators whose projects were cancelled or — "oh, hey, these interns that we were thinking of hiring, maybe not really". and, like, there's whispers of that happening more and more within our industry.

and if you've ever taken a photo and sold it to a stock photography website, you'd better watch out too. most websites, when they write articles, they put photos on to go with the story. and the photos are usually not the actual story, they're sort of stock "people in an office" or something like this. now, we're not quite there. people, hands, faces, they're difficult to do for al. but if that gets better, you could imagine that you mightjust go off to an algorithm and say, "i need people looking really angry �*cause their printer doesn't work", and there we go. and you have solved that problem. that's the kind of thing i think it could find some use for. over the years, we have asked repeatedly what happens as a! takes over more and more of the repetitive mundane tasks? and the answer is always been, "well, at least we'll still have our creativity! we can all go off and be artists!" and al's like, "no, i'm

coming for that too". and as a! starts to play a part in pretty much most parts of our lives, it may be about to shape the conversations we're having, as us company openai has released a new type of chatbot, and it's impressing critics with its writing abilities. zoe kleinman has been giving it a try. you don't always know whether you are talking to a human or a computer. pretty obvious if you're hanging out with tesla's optimus here, but sometimes less so when you are relying on typing and texting. a number of tech firms are working on the future of chatbots — an a! platform trained to engage in conversation. and one in particular has gone viral. it is called chatgpt, from the firm openai, and more than a million people have had a play with it since it launched. it gets its wealth of information and styles of writing from the internet. i'm going to let it

show you what it can do all by itself. first question: "explain chatgpt in two sentences". "chatgpt is a chatbot trained using the gpt—3 language model. "it uses natural language processing to generate "responses to user input in a conversational manner". not so conversational so far. let's try another question: "are you alive?" "no, i'm a virtual assistant "created by a computer programme. "i do not have consciousness or the ability to feel "emotions like a human. "i exist only to provide information and assist "with tasks within the limitations of my programming". right, here we go. i've come to glasgow university to meet drjeff dalton, a senior lecturer in computing science, to see what he makes of it. i think that if i were to ask this on an exam, that is probably a low b. dr dalton spent a lot of time running questions through chatgpt.

chatgpt is very good at generating natural language. right now it can generate a paragraph, it can generate a poem. its limits are it can't generate an article. it can't write a phd thesis. it can't write a specialised report. it can't do those things because it's fundamentally limited in the fact that it doesn't understand what it's talking about. but is it growing too quickly? practitioners are surprised by how quickly these modern systems are developing, and some of them are starting to think we may have systems as smart as humans in as little as ten years, which is sooner than most people thought previously. as platforms like chatgpt develop, what more could they be capable of? maybe they'll replace internet search engines. maybe they'll generate content for websites. maybe they could even write reports like this — although i'm not quite sure where that leaves me...hmm. that was zoe.

now time for a look at this week's tech news. sam bankman—fried, boss of failed crypto company ftx has been denied bail in the bahamas after being arrested. us authorities charged mr bankman—fried with one of the biggest financial frauds in us history. scientists have announced they have overcome a major barrier in nuclear fusion. they can produce more energy from a fusion experiment and they put in. this has taken 60 years and thousands of people to reach this stage, although experts say much more work is needed before fusion is a viable answer to our energy needs. fusion is the energy source that powers the sun and the stars. the reason the sun shines is because of fusion reactions at the core of the star. it is the merging of light elements into heavy elements and releasing energy in the process. taking hydrogen, fusing it together with helium and releasing tremendous amounts of energy. europe's new weather satellite has launched from french guiana. the meteosat—12 uses next—gen

technologies to gather data from europe, the middle east and africa to help track violent storms. and iphone 14 users in france, germany and the uk can now use emergency sos, which lets them contact the emergency services via satellite if they can't connect via cellular or wi—fi. on the edge of glasgow i am at a company taking tech you probably know from healthcare and finding ways for it to be used in everyday life. ultrasound is the coolest imaging technique. the benefit of ultrasound is it is completely non—invasive. it's just sound waves that we're talking now, you can see inside an object without having to open it up. for novosound the process started out in a very industrial way. traditional ultrasound manufacturers would use machining to build an ultrasound probe by machining or breaking away material to make a probe. we do an additive manufacturing

process, that we call printing, that's just a layman's explanation of what we actually do. we use a plasma vapour deposition process, a very physics term for atom—by—atom building up an ultrasound sensor. and then we use some other additive manufacturing techniques we have borrowed from the semiconductor industry to build up a sensor, which we then put in an embodiment which is 3d printed. so there is three different steps of printing, as we call it. this is the novosound kelpie, it's a 64—element linear array, and typically it is used for inspection, corrosion mapping. because it is so flexible it can be wrapped around pipes and conform to complex geometry. so it can look inside of objects using ultrasound. what sort of things are you looking for in a pipe? typically we will be looking for corrosion or cracking, any defects. can i have a shot? yeah, go for it. getting from this very industrial use to something we can interact with every day is a big focus for the company. we have already proven

the ultrasound technology in the industrial sector and aviation, oil and gas. but most people know ultrasound from the hospital or medical imaging, so what we really want to do is be able to take medical imaging out of the hospital and put it into the homes for everyone. a lot of the wearables that are on the market at the moment only looks at the surface of the skin. they are optical and electrical. as soon as you use ultrasound you can get deep into the body and that opens up a huge amount of different parameters you can measure such as dehydration, blood pressure, tissue fatality, tissue movement. and what are you seeing on the tablet? we take an ultrasound picture, zap it over the wireless, put it onto the computer so we can extract information to go into machine learning algorithms. that allows you to measure stuff like, what we're showing here is dehydration, so how hydrated the tissue is. it is all through a completely wireless platform. novosound aren't the only ones working in this area, with engineers at mit in the us having developed a sticker ultrasound earlier this year. the key bit that puts us ahead of our competition in this area is the fact that i have

the ultrasound picture i've got on the screen in front of me, there are no wires, it is truly wireless ultrasound we have here. and it could be as soon as in the next year or two that the tech novosound are developing here is available in devices we can buy. now, every year billions of litres of water are lost due to leaky pipes. and as the planet heaps up and the risk of drought increases, it is an ever more troubling problem. yeah, so some companies are starting to turn to ai—powered sensors to help identify holes in their systems. and paul carter has been finding out more. scorched grass. receding water banks. and record—breaking temperatures. a result of this summer's heatwave, some companies even had to bring in a hosepipe ban. thames water, which serves 15 million people across london and the thames valley,

was one of those companies. it leaks around 24% of its water supply, and along with other water boards, it came underfire for restricting water use while its pipes leaked. we know it is unacceptable to lose large amounts of water through lea ks. that is why we are determined as a company to drive leakage down. thames water is trying to fix their leaky pipe problem with the new a! powered sensor. it aims to identify the location and size of a leak, just by analysing noise. i'm in surrey, an area hit badly by this summer's drought, with the founder of fido tech, the company that created the sensor. here today, we are looking for leaks. so we are in a typical residential area which we know has got some water loss. this is a little acoustic sensor that we fit onto the pipe network. it listens to leak noise and what we have done is a bit

of a! that looks for the noises the leaks make but also filters out the other noises that leaks do not make. neil and his team place these sensors under a drain cover overnight. it's now time to retrieve them. what's the next stage in the process? basically we use the app now, this little device is bluetooth, bluetooth to the phone, and the phone then talks to the cloud and all the analysis is done in the cloud. the graph tells you that consistent noise firstly, and then a! has said ok, i have taken that knowledge, recognised it, and it is a small leak. hissing noise. oh, wow! so that is the actual sound of a leak? right now on the system. so it is a live capture effectively so you can go ok, yeah, i agree with the ai, it is a leak. these are the audio signals we have recorded and it is the status of the leak we have detected. this part of the process is called cloud correlation. it's been developed by professor muhammad zeeshan shakir at the university of the west of scotland. when you have audio files stored in the cloud, they would extract this

features out of it, and try to co—ordinate it. you can say correlation is a process where you try to look at how they look mathematically and try to find similarities in time or efficacy domain. and then the cloud would return you the location of the leak. this system is underpinned by a massive archive of audio files, hundreds of thousands of recordings of leaky pipes from around the world. not only does this inform the cloud correlation, but it should let the ai identify any given leak. but how effective is it? previously using old techniques we have had a 50% success rate. so out of areas that we go into, 50% of the time we come out with the leak we are looking for. what fido has enabled us is to increase that by 20%, so that's up to 70% now in total. it is allowing us to be targeted and where we raise a leak, that we know it is a leak there. ai powered sensors like fido's have not completely replaced

the traditional way of identifying leaks — which is, believe it or not, engineers using listening sticks. but their role in global leak detection could come even more important. now, i wonder whether you remember this guy. i came across him in ireland last year when he was a big thing — well, at least his head was. this was the rather ambitious tourist attraction that its creators hope will one day adorn the skylines of cities around the world. the point being that anyone could be 3d—scanned and then projected onto the giant. it sounded bonkers, didn't it? it certainly felt somewhat of a "big" project. big indeed. but it turns out it really is moving forward, isn't it? yes, and recently it moved all the way to a big expo for the attractions industry, which was quite frankly, the quirkiest place i have

been to in a while. it certainly looks it. and you grew up by the funfairs of the british seaside, i thought you would be right at home there. it felt quite odd when the actual sea was missing. but it was here that the giant had started to come to life —well, a 1.5 metre prototype of the head at least, one—eighth the size of the finished giant's once. the difference between the old head on the new head is that we have got a matrix skin of leds which is linked to the volumetric scanning machine, which now allows you to not only project an image onto the statue, but also to talk. so now it's my turn to be scanned by the volumetric capture booth�*s 49 cameras. i'm going to have the head of a giant. the images are then broken down so the data from them can be used to create a 3d face. that's weird. my face gets flattened

so it can then be wrapped around the model. and even at that size, it is slightly unnerving to come face—to—face with. i am going to have the head of a giant! impressive display, albeit a little intimidating seeing myself that size. and i seem to have lost my hair. i'm going to have the head of a giant! it needs a wig, a giant wig, and it is a bit odd from the side. it has a nose but nowhere close to a nose like mine. absolutely no comment whatsoever. when i did always wonder how they cope with different face shapes? yeah, i know. from straight on i didn't find the generic one that odd, but you do lose the magic from other angles. hmm. now the focus of this project has always been sustainability. and you can see on both the head and the scaled—down version of the whole giant, where the solar panels will be embedded, aiming to fully power them with renewable energy.

but all that scanning that happened earlier has another purpose as well. really? it also creates an augmented reality avatar that anyone can use on a smartphone anywhere. if you don't happen to have your own giant handy. exactly — and you can make yourself whatever size you wish, even bigger than the giant. and back to normal size again, i think? you are, yes. that's it for us from the now gallery in london. thanks for watching, we will see you soon. bye— bye. hello. as this cold snap comes to an end, we've some really unpleasant weather to get through during the course

of sunday, at least for a time. some disruption due to freezing rain in places, really quite icy almost anywhere in the uk. some sleet and snow, but rain and wind too, all arriving with this quite threatening—looking area of cloud which is sweeping towards us as i speak. and within this area of storminess, you can see much milder air sweeping towards ireland and the south—west of england during the early hours of sunday. and it's on the leading edge of that transition between the cold and the mild that we have the weather front. now, ahead of it, it's still freezing cold in the morning. city centre temperatures — minus four, minus five in places, a lot lower in the countryside, and then that rain willjust override that frozen ground. in places it'll be sleet and snow across the hills, but freezing rain possible, and an amber warning in force for northern parts of england. so rain falling on frozen ground — that could lead to very icy conditions, ice accumulating on pavements, on cars, on power lines as well. really quite dangerous.

now, this is going to be around the peak district, the pennines, into yorkshire and possibly the cumbrian fells as well. but ice is possible, as i say, almost anywhere — not necessarily in the south—west and the west of the country. i think by the afternoon it will be a lot milder here. really quite blustery there in cornwall and devon, 12 degrees celsius. and then basically through the course of the evening into the night, the rain will come and go, the winds will be strong and it'll be turning milder and milder as we go into the early hours of monday morning. and then on monday, we've got this large area of low pressure over us still. lots of isobars, so strong winds. the rain really could be quite heavy, particularly around western parts of the uk, welsh hills, cumbrian fells, western parts of scotland. i think the least amount of rainfall the further east you are. but look at those temperatures — positively balmy compared to what we've been having over the last couple of weeks or so — up to about 14 degrees, an extraordinary rise in temperature injust

a short space of time. and as far as the week ahead is concerned, you can see it stays unsettled and generally on the mild side, with temperatures typically into double figures. take care.

this is bbc news, broadcasting to viewers in the uk and around the world. i'm rich preston. the headlines: will it be messi's moment, or a big day for mbappe, as argentina takes on france in the world cup final? excitement on two continents — we look at how fans in paris and buenos aires are getting ready for the big match. china sees a sharp increase in coronavirus cases days after beijing eased restrictions following a wave of protests. the iranian film actress taraneh alidoosti is arrested as anti—government protests continue in iran for a fourth month. singing. and, ukraine's 2023 entry

for eurovision: tvorchi chosen in a live show from a kyiv bomb shelter.