yeah. ai got human — or at least it felt like it did. companies like google and openai started showing off stunning photorealistic images like these, all created by ai from short text descriptions. and then ai started having conversations with us. they were starting to generate stuff that felt human. this field of generative ai seems to have exploded so quickly. chatgpt is the single fastest—growing application in human history. and it keeps getting better. the latest version, gptli, even seems to be able to look at a picture and work out what would happen next. and just look at what the latest ai image generators can do. notjust still pictures, but remarkably good videos as well. this short film was created by one user simply by typing carefully worded text descriptions into his phone. i think the reason many people are now paying attention to ai is that it's finally behaving like the ai we were promised in the movies — computers that we can chat to and that are doing humanlike things. and that's why it has created a really emotive response in a way that none of the ai built into the devices all around us ever managed to. and that's where the danger lies, because if it behaves like a human, it's reasonable to assume that it thinks likea human. but it doesn't. you know those predictive text functions on your phone? well, these try to guess the most likely next word in the sentence based on what you have typed so far. and in really simple terms, that is what these chatbots are doing. they have read millions and billions of sentences online, and they have learned what a good sentence looks like — that's why they sound so human, the sentence structure is really good. but there is no guarantee that they will get the facts right — because they don't understand what they are saying. and image generators don't understand what they are drawing. for example, microsoft's bing app now uses the dall—e image generator. i asked to draw me my initial made of liquid metal and it made this. pretty decent letter 5. it then said would you like me to add some sparks to it, and i said why not, and it turned it into this. i mean... where did the 5 go? the reason is it doesn't know what a letter 5 is. it doesn't think like a human, it doesn't understand anything. but ai generators like midjourney can do wonderful and weird things. and that's the main weakness here and why we can't trust it. if ai can create anything, then how do we know what's real? i don't think the pope ever went out dressed like this. but if we use it wisely, there is immense possibility. it can crunch data like no human can, and never has that been more important than in healthcare, as marc cieslak has been finding out. june works as a healthcare assistant. she knows how important breast cancer screening is. i see you've had previous surgery before? yeah, that's clear. today, june is having a low level x—ray, or mammogram, performed. it's part of a local breast—screening programme called gemini. so we are running this ai as well to see whether it is able to pick up cancers as well as humans can, you we see these little white dots that the ai is slightly suspicious. we would want to do a biopsy on that, recommend it, especially because you have had a previous history of it. a biopsy will be performed, removing a small sample of body tissue and sending it for further tests. here at aberdeen royal infirmary, june's scan has been reviewed by ai software as well as human clinicians. dr gerald lipp demonstrates the process using anonymised scans. so what we are seeing here now, we have a lady who has mammograms on her left side and right side, you are looking for differences. there is a lesion in the left breast here, and of course this is something you would expect a human being — you can just tell there is something different in the pictures there, and if you click on this ai button, it circles an area to check. but the main area of most concern is this area circled here, where the cancer is on the left side. in screening, you want to pick up things that are small before they become big. programmes like this one identify breast cancer in roughly 6 in 1000 women. radiologists, known as readers, examine patient scans for signs of cancer. on average, these human readers scrutinise 5,000 mammograms a year. 250—300 patients will be called back, and 30—a0 of those will require closer attention. and there is a chance that with that number you could miss cancers. within the rules that the national screening council have given us, we are not allowed to use the ai automatically as part of the process as yet, so we are using the ai as an extra check at the end of our reading process. in 2016, a private company, keyron medical technologies, ——kheiron medical technologies, began training an ai model called nia using hundreds of thousands of medical scans. itsjob — to identify breast cancer. until now, this ai has been intended as an assistive tool for use by two human radiologists. it has become the foundation of the technology being used in aberdeen. the health service is experiencing massive staff shortages. experimentation with al could eventually relieve some of that pressure. the first clinical evaluation of its kind is being carried out here. 0rdinarily, two human radiologists would examine every scan, looking for conditions like breast cancer. but could the technology that's being tested here one day replace one of those human medical staff with an ai? i think the goal of this evaluation is to see what's the best way we can work with al, where there is replacing one of the radiologists, where there is part reading some of the normal mammograms, or where there is to improve our cancer detection as a safety net. this project is a collaboration between the nhs, the university of aberdeen and private companies — microsoft providing cloud computing, and the ai model developed by kheiron. the next steps of the partnership is that we are taking our ai across the uk to over 30 nhs trusts, to over one million women to gain access to the ai screening. it has been a few weeks sincejune's biopsy, i caught up with her via videocall. i wondered how she felt about an ai assisting in such sensitive work. your images are on—screen and people are looking at them. whereas when it's an artificial intelligence, that feeling that somebody is looking in on has gone. the biopsy showed that i do have an early—stage cancer, we've certainly caught it at an earlier stage this time, but because i have had previous history with it, i'm going to go in and have a mastectomy — it's not the treatment i want to have, but at the same time it is reassuring that it has been caught. screening programmes are crucial for improving patient outcomes. for now, medical staff are still the first line of defence in protecting against breast cancer, but ai is likely to play a significant role in future life—saving efforts. that was marc showing us how ai in healthcare can be really useful. but on the flipside, when it comes to ai being used to generate things like art, it can be problematic. yeah, one of the big issues being copyright. i mean, who should own the images that al creates? it's something that ben derico has been investigating. ai art has taken a massive leap recently. i mean this one sold for over $400,000 at auction at christie's in 2018. with image generators like dall—e and stable diffusion, almost anyone can create a new art in a matter of seconds. but the models that makes this art don'tjust do it out of thin air. they have learnt to mimic styles, even specific artists, through a process called training — where the models injest millions, sometimes billions of images, scraped from websites all around the web. combined with text describing the images, they now have a data set that lets them create almost any type of image from a simple text prompt. it produces some interesting stuff, but the problem is many artists never gave their consent for their art to be used in an image generator like this. so what should artists do? so we have seen art theft before. but we have never seen it at this level. this is karla 0rtiz, she's a concept artist in san francisco. a concept artist is a person who provides the first initial visuals to what something could be in something in a movie. she has designed art for magic: the gathering, and even in marvel�*s doctor strange movies. last year, she discovered that her art had been scraped into an ai image data set. especially my fine art work, and that to me felt really invasive, because i had never given anyone my permission to do that. 0n midjourney, another popular generator, it's incredibly easy to find posts using karla's name to generate work that looks incredibly similar to hers, and the same is true for dozens of other artists online. so earlier this year karla and a group of other artists filed a class action lawsuit against stability ai and a group of other ai image generators. in the meantime, karla made the decision to take her work off the internet wherever she could. she figured it was the only way to avoid a computer scraping her work into an image data set without her consent. but what if she could still show her work online and keep it from being used to help generate new ai art? honestly, we just never had any idea it was such an impactful problem. this is professor ben chow, from the university of chicago. he and his lab say they have developed a solution. they call it glaze. at its core, glaze uses the fact that there is this ginourmous gap, difference between the way humans see visual images and how learning models see visual images. because we see things differently, glaze can make changes that are almost imperceptible to the human eye, but that dramatically alter how a machine sees it. so, if you are an artist, you glaze your art, post it online, you can rest comfortable in knowing that a model that is trying to steal your stuff from that piece will learn a very different style that is incorrect, and when it is trying to mimic you it willjust fail, and halt these attacks early. as you can see, the ai artwork generated from a piece with glaze is similar in content but not really in style. to learn more, we asked the team to show us glaze in action. you can see some changes already on the texture. the right side one is the original one, this is the glazed one. you can see the face has some added changes here on the painting. right, so the brushstrokes are a little blotchier, it looks like there are these kind of yellow patches of pixels in there? exactly. if you had laid these out in front of me and said "tell me what is different, iwould be " ha rd—pressed to tell you." the promise of glaze is exciting for artists, but critics say the ai art generators are taking inspiration the same way a human does — by studying other pieces and learning from them. crucially, they say, these aren't copies. that's led the companies being sued to ask for the case against them to be dismissed. karla says, though, that's not a good comparison. i don't see one image, let alone billions of imagery, and instantly like archive it in my mind, and then i'm able to generate exact copies, or similar copies in the blink of an eye. some artists have said they would be willing to use their work with al image generators, but they say the process should be opt in, not opt out. stability ai says their new generators will be opt out going forward, and adobe says its new image generator, firefly, has only been trained on images from its stock library. but even there, adobe contributors say this type of usage was never explicit in their agreement. in the meantime, the internet being the internet, people are already trying to break glaze and get around it. i am under no impression that it will last forever and protect against every form of attack, but we are hopeful that it will buy artists some time, and in the meantime, i'm hopeful that glaze will not be the only tool of its kind. for karla, that's the point — to buy artists time for regulation and the public to catch up. when peoplejump in on these and say, "oh, this is, wow!", they need to recognise that it is "wow" because of the work that is taken, and all of that work was taken without their consent to use, to train these models so that they can generate that stuff that makes people go, "wow". ai art is likely here to stay — so pressure from regulators, input from artists and an informed public will be crucial to make sure these revolutionary tools are built alongside the people who helped make them possible. that was ben, and this is this week's tech news. microsoft will have to pay $20 million to us federal regulators after it was found the company illegally collected data on children who started xbox accounts. the federal trade commission said the technology company failed to inform parents about the data it was collecting. apple has unveiled its first major hardware in almost a decade — apple vision pro. the long—rumoured headset combines virtual reality and augmented reality and is controlled with eyes, hands and voices. the headset allows you to watch movies, write documents in a virtual world, and even immerse yourself in your own photography. it helps with the connection that you still have with people, so you are fully immersed but you are not cut off by your reality. binance has been hit with another lawsuit by us financial regulators. the securities and exchange commission said the crypto giant ignored rules that are meant to protect investors in the us. binance said it would defend the platform vigorously. and researchers at the university of cambridge have created a robot that can prepare eight salad recipes. the robot watched 16 videos of a human making a salad, after which it was able to copy them and prepare it itself. bon appetit. just because technology can do something, doesn't mean that we should let it. the ethical issues raised by ai are fundamental to how we regulate it and how it becomes part of all our lives. it is time to rethink our interactions, look at all the possibilities and the risks. and somebody who grapples with this a lot of the time is nina schick, who has written books on deepfakes. nina, we can't even possibly begin to know what's real or not now — how on earth do we deal with this? that is an existential question for society, because i think this is the last moment, if you will, in the internet's history where the majority of data and information and content we see online is not generated or created by artificial intelligence. because we are seeing this new field of artificial intelligence, so—called generative ai, that can create content and information in every single digital format. and the use cases of generative ai are so profound, increasingly we will start to be engaging with ai—made content, it is going to become ubiquitous. that seems like a pretty unsolvable issue. you have to take a cybersecurity approach because there is no silver bullet that will fix it, but you have to kind of start building layers of resilience around society to navigate this kind of new era of ai. i mean, you can imagine a world where people are fooled by ai—generated images, but i can also imagine a world where, if something is true, people just won't believe it. and so someone who that image affects, maybe a politician or a leader, canjust say, "well, that is fake news", and even though it's genuine, because there is so much doubt cast throughout society. you've hit the nail on the head. that is a phenomenon known as the liar's dividend. because it is not only that every piece of content or text can now be generated with al so you can "synthesise" or fake anything, it is also the understanding now that everything can be created by ai that undermines the integrity of everything that is authentic. but should there be one sort of international way that things are done, is that even possible? well, the eu ai act will be the first international regulatory piece on al which will probably set standards for a lot of the rest of the world. as for an international body, ijust don't think that is likely. when it comes to jobs, it's obviously going to change the nature of what jobs are available. some jobs may no longer exist, other newjobs will emerge. but of course many people are trained up to do the jobs that exist in current society, so how do we deal with that? i think the impact is going to be bigger than the industrial revolution. i think the labour market will be fundamentally different. now there is this debate going on about, is this going to augment us or is it going to automate us? ultimately i think it will become a very political issue, because it will be both. and the really interesting thing is that this is true now for the first time for white—collar work. do you think there actually is a worry that we are giving more control over to automated systems because they are just faster and better than us — and then every so often something goes wrong, a bit like the financial crisis. sure, i think that's a legitimate concern. it's really important to recognise that these systems are fallible. they are not purveyors of the truth, they are not omnipotent, they are not gods, right? they are just as good as the training data and as it turns out, you can already see with chatgpt, they have this propensity to lie, to hallucinate, to make some stuff up. they are not infallible. by giving these machines undue sentience or capabilities, i think we are actually stripping away our own agency, and that is that these systems are still very much in the control of organisations and people. and over the next few years, months and years, we have a chance to think clearly and strategically about how they are going to be integrated into society. thanks so much, nina. a lot to think about. and now for a creative look at what could happen if ai did start to take over. yeah, we have been to the misalignment museum in san francisco, which aims to educate people about al and it features. . . paperclips. lots of paperclips. the paperclip maximiser problem is a thought experiment that posits if an al was developed to create paperclips, it would eventually use all the resources in the world, possibly destroy all of humanity just to create more paperclips. the concept of the museum is it is a post—apocalyptic world where ai has destroyed most of humanity, and then it realised that was bad and so it created this other type of memorial. i created the church of gpt for this museum. the idea is you can pray for confess your sins to this vengeful, spiteful but also incredibly wise ai god of the future. yes, unfortunately, i have been using them for gaming instead of deep learning. it is a great sin, i know. i'm very sorry. so this piece is called spambot. they are collectively typing up an ai—generated version of brave new world. this artist was trying to draw attention to the increased proliferation of spam as ai is able to create more and more content. music plays this is an ai—generated music composition. it was programmed to write music in response to bacteria growing in a petri dish. it also raises a lot of questions about ip and copyright ownership in the advent of this type of powerful technology. the museum right now is a pilot, but we are actively using this time to try and develop a founding donor base to get a permanent location. my ideal is to have ten different rooms, it will be an immersive experience that tells part of the story of the development of ai and the potential futures we could have, both positive and potentially scary. i think it is up to us to make the future that we want. what a great visualisation of what ai might do, if it all goes wrong. yeah, let's try and make sure that doesn't happen, people, all right? good, we're all agreed on that. thanks for watching and we will you soon. bye. hello there. i think it's fair to say that the weather doesn't look and feel much like summer at the moment. these were some pictures taken on friday. it was particularly windy in the southwest of england and we actually had more rain in cornwall on friday than fell during the whole of last month, and we had rain far and wide across the uk as well. that rain has been working its way northwards on that weather front there, which is wrapped around an area of low pressure, and that will dominate the weather through the rest of the weekend. now, the rain, by saturday morning, is in the far north of scotland. temperatures 12—14 degrees. some showers already arriving and we'll see more of those as the winds pick up, particularly across england and wales, with the strongest of the winds in the south. a0 mph gusts, quite widely, maybe a bit stronger, especially around some coastal areas. and it could bring some damage and some disruption. especially as those downpours arrive. and we'll see these showers breaking out more widely through the day. some wetter weather, particularly for wales and western parts of england. thunderstorms almost anywhere. some heavy showers arriving in scotland and northern ireland where it's not going to be quite so windy here. but temperatures are still a bit disappointing, really, for the time of year. we're likely to find 18—19 widely, a touch warmer in the southeast where there shouldn't be as many showers in the afternoon. the low pressure itself is continuing to push northwards. it may take away the worst of the weather for sunday. although having said that, it could be a bit windier than saturday for scotland and northern ireland. and we've got some showers here, some of them heavy, maybe some longer spells of rain. but not quite so windy on sunday for england and wales. there may be a bit more sunshine around, but there's still the chance of some showers, too. even though there aren't as many showers around, we've still got those temperatures peaking at only 20—21 celsius in the afternoon. now, looking ahead to next week and some changes on the way. it's not going to be quite as windy next week. there may be a bit more sunshine around and fewer showers, but we're not going to get any of the heat that's affecting southern parts of europe. you may be pleased about that. 22—23 degrees the top temperature, probably, over next week. it's going to be much hotter across southern parts of europe. temperatures not quite so high around coastal areas, but it will be especially hot as you head inland. live from washington, live from washington, this is bbc news. the largest strike in nearly six decades is under way in hollywood, as actors this is bbc news. join the picket lines. guatemala's presidential election thrown into turmoil after an attempt to after an attempt to oust a top candidate. oust a top candidate. and india makes its next big and india makes its next big step on its way back step on its way back to the moon. hello, i'm carl nasman.