/filters:quality(75)/filters:no_upscale()/filters:strip_exif()/slv-lab/media/images/177664135053d/kate-crawford-seated-portrait-slv-approved.jpg)
As Artificial Intelligence (AI) systems pull data from across the internet, library databases and our online presence, it generates a significant change in our relationship with knowledge. But what do we really know about this pervasive technology? Do our societies have robust guardrails in place to protect people and democracies, and have we reckoned with AI’s environmental impact?
Join award-winning Professor Kate Crawford, one of the world’s foremost scholars on the social and political implications of AI, as she explores how AI is reshaping our societies, ecosystems and power structures with host Natasha Mitchell in State Library Victoria's 2024 For Future Reference lecture.
*The following transcript was machine generated and may contain errors.
Transcript
0:00:03 - 0:07:30
Natasha Mitchell
Good evening all. Isn't this such a beautiful space? Thank you to the library for hosting us. What a wonderful event. And thank you, Joel. And thank you, Collin Hunter, junior junior, for your welcome. I want to also acknowledge the fantastic work being done around the world, including here in Australia by First Nations leaders on artificial intelligence. Creating indigenous protocols around.I also using indigenous knowledge systems to really scrutinize our relationship with AI and to ask what sort of, relationship do we want? And how can I be used to empower communities, not just exploit us? How can I not be another colonial force? And, if you're interested in that, the Abundant Intelligences project is fantastic to look up.It's a real delight to introduce tonight's very special guest and welcome her back to Australia and back to Melbourne. She is one of the most important and influential, consequential and nuanced, intellects looking at AI and its consequences. And I've been invited to just give a few sort of opening thoughts about artificial intelligence and where we're at, I guess, Kate and I both, kind of emerged through a particular era.I did engineering, and, here we are today looking at this technology and its profound consequences because, of course, just a few years ago, I mean, for a long time, artificial intelligence was mainly the domain of military scientists, engineers, researchers, computer systems nerds, tech startups, to an extent, a handful of science journalists like myself, a few experimental artists, and most certainly humanities scholars who were interested in understanding the impacts of technology on our lives and minds.That has all changed. All that is still going on. But AI is fundamentally here. It's changed at breakneck speed. It's everywhere. It's hungry for your data, for my data, for all data, for any data. Because data, of course, is what teaches machine learning algorithms and artificial intelligence systems to do their many wonderful jobs. So the comments that you're posting to social media, that's data.And, it's being used to train up large language models like the one that drives ChatGPT, uni, who's a uni student in the room. Have you used ChatGPT to, write a couple of essays yet? Oh, well, there's a few nods, a bit of a blush in seconds. The photos that you're posting, you know, the old school Flickr, your Facebook account, they're being used to train up facial recognition, systems that are being used in surveillance and similar.You romcom viewing habits on Netflix. They're being used to train up recommendations for you, your online purchases in the middle of the night, those guilty, those guilty online purchases, your Google searches. It's all feeding, the information that you served up in terms of what you might binge or buy next. That's not where it ends. Banks, law enforcement authorities, intelligence agencies, governments, the state they're all mining your data and using it to make extremely consequential decisions about our lives, which is what we're going to hear about tonight.It's doing all sorts of great things as well. AI is being trained up on things like data sets on epilepsy so that doctors can better prescribed medications in a more targeted way, or where the data so that we can predict weather patterns more effectively. So there's a whole lot of great work happening with AI. Really, all it's there to do is help us crunch a whole lot of data and make sense of it.But as perhaps we'll hear tonight, it can be also, it can be useful, but it can be abusive, it can be benign, but it can also be biased. It can be sexist and racist and ablest and all those things that we are, because we're teaching it on the data that we create. And the other thing that I'd add is that I might be everywhere, and yet it's hidden from view behind firewalls and commercial in confidence increments and corporate black boxes.And so our data is being used to make a whole lot of corporate profits, which we hand over for free, but at what cost? So and here we are in a beautiful, wonderful institution of intelligence, a library. What will I mean for these institutions of intelligence? Our libraries, our universities, our schools? Perhaps we'll hear about that tonight. I can't think of a better thinker in the world, really, to help us make sense of our increasingly extractive relationship with this technology that we are creating.Kate Crawford, perhaps you, like me, first came across Kate Crawford's work as one half of that brilliant electronic music duo, beef tech, who remembers beef tech in this audience? Yes, hello. Very danceable as an engineering student, though, that music really spoke to me because it was exploring our relationship to technology and how it shapes our minds and our music and our cultures.So thank you, Kate, for hours of pleasure. And I've got to say, I sampled quite a bit of that music in my early radio documentary, My King. Professor Crawford has, of course, gone on to become a world leading scholar on the social, ethical, environmental, political impacts of artificial intelligence. She is a professor at the University of Southern California in Los Angeles, a senior principal researcher at Microsoft Research Lab.She was named in that very posh inaugural time 100 list of the most influential people in AI and her marvelous book. If you haven't read it, Atlas of AI. It is a fantastic introduction to the themes that you'll hear tonight. It's won multiple or international awards. It's been translated into 12 languages. And along the way, Kate has advised, advised, countless policy makers, including currently, on the AI Council of the president, Sanchez of Spain, but also the white House, the UN, and much more besides.And she's an accomplished artist. So if you ever get the chance to look at the website of The Anatomy of an AI system, it is a staggering piece of work that she did with, Vlad and Yola, which won the London Design Museum's award. It is a fantastic interrogation. You'll find out more about it. It's currently at MoMA.If you happen to go to New York and their latest collaboration, Calculating Empires A Genealogy of Power and Technology, won a big European Commission award as well. And we'll be showing at next year's the Venice Biennale. So we are extremely lucky to have Kate with us tonight, to present to you. She'll present, will have a chat, and we'll open it out to the floor for Slido.Questions, please welcome as loudly as you can, Professor Kate Crawford.0:07:30 - 0:39:29
Kate Crawford
What a beautiful introduction. Thank you, Natasha, and good evening, everybody. It's so lovely to see you over here. And I have to say, it's an honor to be invited to give the Helen McPherson Smith Trust lecture. Of course, we are gathered here tonight on Unceded Land, and I'd like to pay my respects to elders past, present and emerging.And I also want to thank the amazing Paul, Paula, Joel and Audrey who have done so much to make me feel at home and welcome in the library. It's been an amazing week. I've seen the treaty exhibition. I saw this extraordinary photography of Rennie Ellis, and I also got to go in the basement and look at the archives, which was probably my absolute favorite thing.So there are archives of famous magicians in Australia's past, and all of their secrets and tricks are in the archives. So you need to know that it's also personally special for me to be back here in Melbourne, because when I lived here in the early 2000, this library was a really important meeting place. I would catch up with friends on the steps, we'd discuss ideas.We'd sometimes even occasionally study, but we'd also protest and we would debate. And I think we need institutions like this to create spaces to address those issues that matter the most to us. And there's never been a more urgent time to discuss artificial intelligence and to understand its impact that it's having on our society, our politics and our planet.And it's so happens that we're gathered here on a particular anniversary almost exactly two years ago, ChatGPT was released, just a plain white input box with the words how can I help you today? And in a much smaller font, if you squint your eyes and look really, really closely at that gray text, it says ChatGPT, you can make mistakes.Well, indeed. And we are going to see what kind of mistakes have now become commonplace. But ChatGPT also set off a cut throat technological race. We saw tech companies all scrambling to release bigger and better AI models as quickly as possible. So now generative AI, as we heard from Joel, is built into your phones. It's in search, it's in email, it's in design tools, making text, images and video for millions of people every day.But this bland front end gives you no hints at all about what's going on in the background, the enormous infrastructure that is needed to keep this running because generative AI is, in fact, one of the most resource intensive planetary structures that our species has ever created. But AI is also surprisingly unreliable. Take this recent gem. So Google's AI overview, a people you know aware.Yes, a few people putting their hands up. We've all seen it was suggesting to several people that when you make pizza, you should always have a layer of glue to prevent the cheese from sliding off right? Glue. Super tasty. In fact, one very brave journalist actually followed the recipe and then ate the pizza. There she is there and it was just as disgusting as it sounds.It also suggested my new favorite cardio exercise running with scissors. Fantastic. Google assures me that it's also really good for my pores and it's going to give me strength. So just, you know, let's go and run with scissors after this. It'll be great fun. So obviously this is pretty funny, but some recommendations are much less amusing, like this one.When Gemini threatened a graduate student who was just doing some research out of the blue, the engine said, you are a burden on society and a blight on the landscape. Please die.So you might imagine that a threat like this is a major technical glitch, and would even cause a product to be shut down. Well, in the case of large language models, this is less of a bug and more of a feature in terms of how AI systems are trained. And this is true of all of the models, regardless of who's building them, because large language models right now are built using indiscriminate collections of data that we heard about from Natasha.And it's being curated very, very lightly, in some cases not at all. And that is resulting in this mess of really problematic data. And the result is that AI systems like ChatGPT and Gemini churn out these mistakes, also known as hallucinations, as often as 27% of the time. Now, this flood of dubious AI output is so common it's now been given its own name, and it's called slop.Now, slop is this shoddy, low quality AI content that is spreading across social media, search and the web. You've seen it. It's that hyper glossy almost, you know, too good to be true imagery, those eerily perfect AI generated faces. And of course, reams and reams of counterfeit books on Amazon. It's just everywhere. This kind of AI is a byproduct that is growing so quickly that platforms can't remove it fast enough.Slop is flooding the zone, so the internet, as we've known it for the last 30 years, is now changing dramatically. Now, Paul Virgilio, the philosopher of technology, I think put it best, when you invent the ship, you also invent the shipwreck. When you invent the plane, you also invent the plane crash. When you invent electricity, you invent electrocution.Every technology carries its own negativity, which is invented at the same time as technical progress. So I flop might be the most obvious form of negativity because we see it every day. It's visible. But what about those forms of negativity that are actually hidden from us? The ones that are not algorithmic but actually structural, literally built into the infrastructure itself?Those are much harder to see because I right now is affecting our environments, our cultures, and even how we define knowledge itself. So that is going to be the focus of my talk today, how to map the less visible impacts of AI so that we together can see the ship and the shipwreck. Now, some of the richest men on Earth, Elon Musk, Sam Altman and their ilk claim that the biggest threat of AI is that we will soon create superintelligent machines that will outsmart and ultimately destroy us all.This Silicon Valley obsession is also known as X risk or existential risk, and some people also call it superintelligence. But you might notice there's a slight paradox here, right? Because the same men who are warning us of a doomsday scenario are the very people who are trying to build and profit from AI as quickly as possible. So who do you suggest they are saying will save us from this apocalyptic future?Can you guess? Yes. It's them. What a surprise. And I'd like to suggest that this is more than just self-serving. It's actually a very dangerous ideology that is distracting us from AI's real and immediate harms. And meanwhile, tech billionaires are accruing unprecedented investment. And now in the US political power. So tonight I'm going to focus instead on what is actually happening in AI.And my talk is going to focus on two powerful interconnected forces. The first is the industry's extraction of data. It's being harvested from everything that's ever been put online every book, blog, essay, every time you wrote a little comment on Reddit, every photograph, every illustration, the demand for data has simply never been higher. And the second part of this equation is AI's environmental footprint.Now, when most people think of artificial intelligence, they think of zeros and ones or algorithms or code in the cloud. But it's actually profoundly material. And it's driving this exponential drawdown of energy, minerals and water to power generative AI. So what I want to show you tonight is that these two forces the appetite for more data and the demand for more material resources are intrinsically linked, with one driving the other ratcheting up in this sort of escalating cycle.This dynamic is at the very heart of how AI is built today. But to see it, to actually see how this works, we have to map it out. So my interest in mapping began about a decade ago, when I started by visualizing the full lifecycle of an AI device. This was the first one working with my collaborator Vlad Angela.We visualized all of the processes needed to make a single Amazon Echo. You know, those little cylinders that some people have sitting on their kitchen bench? Basically, we traced all of the minerals that are inside those components to places like West Africa. We traced where the data processing was being done in the United States predominantly. And then we looked at where these devices are thrown away, generally after only three years of use into e-waste tips in places like Ghana and Pakistan.So when you ask Alexa a question like, what's the weather today? You are invoking into being this truly planetary computational system, and vanishingly few companies can really build AI at scale like this. It is a profoundly concentrated industrial infrastructure. So let's begin with the first part of our story tonight. And it's data. There is an ad that Apple released earlier this year, which is an accidentally perfect metaphor for what is actually going on.It shows a giant hydraulic compactor that is slowly pulverizing musical instruments, video games, a typewriter, a camera, and the ad shows this kind of irreversible destruction, where centuries of human expression and creativity are being crushed into a kind of sticky paste. And I think in some ways, this is the most vivid representation of what is happening in AI, this kind of flattening that's going on today, this normalization of the forms of human writing and image making and music and all of these forms of creative expression to treat it as though it's all just the same substance, it's all just flattened out data.So this is the AI hydraulic more as I like to think of it. It's mass harvesting images, words, music, video all compressed into AI models to feed us back different versions of the same. But to understand this current explosion of AI data sets and gargantuan AI models, we have to go back to where this all began. This is IBM's computational speech recognition lab in the early 1970s.At this point in AI, expert systems were in fashion. That meant teaching computers grammatical principles and linguistic features. But that all changed in 1972, when Fred Jelinek was hired. He rejected the focus on human centered expert knowledge. Jelinek began using statistical methods to analyze how often words appeared next to one another. In fact, I used to have this joke where he'd say every time I fire a linguist, my model gets better.So he was not interested in a word's meaning, but in a words frequency in a sentence. So if you're going to try and make this work, you had to have a lot of data. But back then, large word collections were really hard to come by. So his team tried using IBM technical manuals, but surprise, surprise didn't sound like humans.I think we're not surprised by that one. And actually it was quite ironic. The thing that worked for them was a major antitrust lawsuit against IBM that went for 13 years, and they called a thousand witnesses, and that became that text corpus. Can you believe it? So interestingly, at that time, Robert Mercer, who was quite young, he was a young, dashing man.You can see him here. He was actually, you know, quite an influential figure in this lab. But you might have heard of his name now as the shadowy billionaire who funded Donald Trump and Cambridge Analytica. But back when he was young, working in this lab, he coined a phrase that would stay with the field. And he said, there's no data like more data.So we could think of this as the genesis of the statistical turn in AI. In the words of the historian Chao Cheng Li, that was the moment where we moved away from trying to get computers to understand us, to making them try to predict the next word in a sentence. And today, I think we're living the results of that moment with AI systems that can produce an answer to any question you might have.But with no understanding of what that means, or if that answer is correct. So we're now seeing a dramatic shift in the scale of data extraction. Let me give you an example. Back in 2015 I started studying this data set, which is called ImageNet. This is one of the most influential and widely cited data sets in the world.It was originally designed with the aim of, and I quote, mapping out the entire world of objects. Small goal. It had, 14 million images scraped from the web and hand labeled by crowd workers into 20,000 noun categories. This was a breakthrough for computer vision systems. But when I started looking into this as a researcher, I was absolutely shocked at some of the categories that it was using to put people in.There were categories for criminals for drug addicts, for Jezebel's and alcoholics, and many far worse terms that are not repeatable. Tonight, this was a regime of classifying people that played to the worst stereotypes of race and gender and class. And of course, let's just remember, none of the people whose images are included here had any idea that they were part of one of the most influential AI data sets that went on to create so many influential systems.So I think in some ways, ImageNet is a lesson. It's a lesson about what happens when people are treated like objects. And it was also the moment when consent was no longer seen as important. If it was on the internet, it was fair game for AI. So let's fast forward to today. Data sets now consists of tens of billions of images.For example, there's one called lion five Fivb. That's the data set that's behind AI systems like Stable Diffusion and Midjourney. I'm sure some of you've used them. So this dataset has 5 billion images and text captions. It's truly an internet scale training data set. But if you look inside it you'll actually notice something really striking. This is not a representative sample of the internet, nor is it a neutral mirror of human culture.Sometimes we hear this idea that AI is just reflecting back what humans do. It's actually not. It's very skewed. It actually prioritizes certain types of content. So with the Knowing Machines team, we published an investigation earlier this year showing that the largest source of images for lion is actually e-commerce sites. You know, primarily that number one was Shopify, and number two was eBay.What that means is that AI tools that are trained on this data are steeped in commercial esthetics. From the very beginning. And when you use them, you're not just making any image of the world you can imagine you're seeing through the eyes of a shopping site. So the inevitable result of massively scaling AI data sets is that your computational demands skyrocket.And that means more structures like this. This is a hyperscale data center, and we can think of them as the sort of nerve centers of AI. This is where the data is processed and the models are served up. Each one of these is large enough to house multiple football fields. They're truly massive. I've visited many of them. They're roughly 8000 of these world wide, with many more under construction, and they operate around the clock 24 seven.These are the factories of generative AI, the very expensive, and they use a lot of resources. Which brings us to chapter two, what this means for our ecologies. Because already we're starting to see the effects of large scale computation in our atmosphere, in the water table, and even in the Earth's crust. This is the Saleyard Atacama. It's the largest salt flat in Chile.And I had the great privilege of visiting this mine earlier this year with activists who are protesting the rapidly increasing lithium extraction in this absolutely stunning desert is the richest site of lithium in the world, and it's now the heart of an ongoing political struggle for one reason rechargeable batteries. Because lithium undergirds every single device and platform in the world, Elon Musk calls lithium the new oil and we can guess why.Because Tesla is the number one lithium battery consumer, estimated to be using half of the planet's total consumption of lithium. But we know that those reserves are actually under pressure. The International Energy Agency said that the world faces shortages of both lithium and cobalt by next year, and we're starting to see the global conflict over minerals for I really heat up.You might have seen this story yesterday when China announced that it will ban the export of several rare earth minerals to the US. These are the minerals that are essential for building the semiconductor chips that make all of those hyperscale data centers. This was, in fact, a retaliation against the US, which last month restricted the sale of AI chips to China, such as the chips made by Nvidia.So make no mistake, we are now in a supply chain war over artificial intelligence, and it is escalating rapidly. But if minerals are the backbone of AI, energy is its lifeblood. And as the data demands go up, the models get bigger. So does the demand for energy. So Sam Altman, who was addressing Davos earlier this year, admitted for the first time that Gen I could push the world to an energy crisis.And why is that? Well, because every generative query you might make. So let's say you ask ChatGPT something that is 15 times more energy intensive than a traditional web search. And in fact, those demands for energy are now pushing the limits of the electrical grid in the US. So what is the response from AI companies? Well, last month we heard that Google is now, firing up seven nuclear reactors in order to meet their demands.And in fact, some companies are installing nuclear reactors inside the AI data centers. What could possibly go wrong? And of course, we still don't have long term containment options for nuclear waste, which has a half life of millions of years. And it's not just energy, because generative AI also needs enormous amounts of fresh water to cool down all of those heat producing GPUs.This image is from Google's data center in The Dalles, Oregon. After a lawsuit, it was revealed that this center was using 29% of the city's total drinking water. And worse, this area of Oregon is already in a multi-year drought. And a paper by UC Riverside researchers has estimated that an exchange with ChatGPT. So let's say you have, you know, 20 back and forth is the equivalent of wasting a half liter bottle of water.And that is being reflected now in the environmental reports from Google and Microsoft. Both companies are reporting major spikes in their water use. So these are just some of the demands of generative AI. Now, the philosophers Michael Hart and Antonio Negri describe the information economy as premised on two things the dual operation of abstraction and extraction. In other words, tech companies abstract away these material conditions of production that we've been discussing while extracting ever more data and natural resources.But I think this equation needs to be expanded. I think what we're seeing in AI is a three part process of abstraction, extraction, and destruction because our attention is being constantly targeted, particularly by new AI models that are fine tuned to track our interests. A our passions and our pleasures, to feed us more of the things that keep us glued to our screens, be that TikTok dances or Instagram cat memes.My personal favorite this does, I think, make it much harder for us to focus as communities and as a much wider population on this bigger question of what is at stake for all of us. And now we're seeing the release of a new AI model basically every couple of months or so. Take this one. This is the new text to video model Sora, which was recently leaked online by artists who were very angry about being used for as kind of PR campaigners.And this was a video that they were using to promote this service, and it seems a little on the nose to me that they're generating with AI extinct megafauna that died out 4000 years ago because of climate change. Yeah. So this image haunts me, but probably not for the reason that the marketing department intended. So tonight, I've given you a very quick tour through how the rise of data heavy artificial intelligence is fueling an exponential demand for resources.In short, what is happening all around us is the creation of a planetary infrastructure that is directly competing with humans and animals for basic resources like water, energy, and land. And it's pushing the limits of a planet that's already in a climate crisis. So you might be thinking right now, where is all of this heading? Well, I want to conclude with three paths forward.And hopefully some some of these will give you some hope as well. I want to start here with libraries, because I think it's in places like this that we're going to see something very significant happen in the next few years, because these are going to be increasingly the only trusted institutions of knowledge and connection. So let's go back to this idea of AI slop, that general pollution of our information ecologies in some ways.I think the biggest risk that we're facing is not that people will believe things that are fake. I think it's that people will stop believing things that are true. And that means that news, information, knowledge itself will no longer be trusted, and it will always be greeted with suspicion. That is a kind of epistemic collapse. So as trust in online information erodes.Where will people turn for the knowledge that they need? Well, I think it's going to be more and more in places like this, because libraries have a multisensory history as custodians of knowledge. The library where we gather tonight was established in 1856 under the banner of the People's University. I particularly like this because it represents a legacy of intellectual stewardship for all.And I think this is really now urgently relevant in an era where the internet is increasingly weaponized and polluted with AI generated noise. Libraries remain a vital alternative. They're not just repositories of history, but they are essential tools for making sense of our present. As the internet declines, we're going to need to rely on libraries more than ever to anchor us in knowledge context, and understanding.I think a second important path is going to be regulation. In truth, governments around the world have struggled to regulate AI, but we are now seeing some moves in the right direction because this year the EU brought the AI act into effect. And this is in fact the first omnibus piece of legislation targeted to AI. And it focuses on important issues of transparency, assessment and accountability.Tragically, I have to say, having just got here from New York a couple of days ago, the US is going in the wrong direction with President Trump saying that he will reverse Biden's roadmap for regulating AI. But I'm very glad to say that when I landed, it was the same day that Australia's Senate Select committee issued their report on AI in Australia.And I have to say, I was really impressed to see that it was pushing for environmental and training data transparency. It's 220 pages, so I'm sure you'll all go home and read it cover to cover tonight. But it's it is actually really worth your time because I think right now Australia is facing such an important decision, which is how much protection do you want from these downsides of AI environmentally and culturally?This is the critical time for governments to play a far more significant role in restraining tech power. Perhaps the final thing that's been giving me hope is remembering the long histories of technology. We can recall that these kinds of transformative shifts have happened before and will happen again, from the Industrial Revolution to the Manhattan Project. And we have so much to learn from those histories in the AI explosion of the last two years.I think it's been very easy to lose sight of that long term view. And speaking personally, I think making a new kind of map during this generative AI moment. And it's called calculating empires. So this is the follow on from anatomy of an AI system and slow down. And I decided to actually shift away from thinking about space and AI to think about time.So we created this 24 meter mural to look at the history of technology over five centuries, and then to illustrate it. We start with the rise of capitalism in the 1500s, and then trace how global networks, both cultural and mercantile, were taking shape from the printing press through to long range shipping, including advances in weaponry. These were the technologies that ultimately enabled land annexation and genocide.New technologies of classification. We used to catalog and capture land, animals and plants. It was a multisensory agenda of extraction and capture that I think we're seeing play out again. But now it's in the empires of I. What I learned doing this kind of a long jury project is that we are seeing a dramatic concentration of power in the 21st century, just as we've seen in the past, from the Ottomans to the British Empire.This historical view, I think, reminds me that change is inevitable. Empires do fall, and collectively people have pushed and won when they moved together with vision. So if there's something that I'd like you to take from tonight, is that nothing about technology is inevitable. I think there's sometimes this sense that the expansion of AI cannot be stopped. All we can do is tweak the edges or figure out how to tame these systems, be it, you know, reduce hallucinations or make better privacy settings or write stronger regulation.But these are always going to be partial and incomplete responses. Instead, I'd like to suggest something different that we reverse this arrow and ask instead what kind of world do we want? And how can technology serve that vision and not drive it? Do we want our AI systems to be accelerating climate change, or do we want them to help us solve it?Do we want all of our data to be taken without consent, or should there be a different vision for our archives, our histories, and our stories? This is ultimately the conversation I think we need to have, and it's fundamentally a democratic one. Thanks.0:39:29 - 0:40:14
Natasha Mitchell
Kate Crawford.Fantastic presentation. Wasn't that absolutely excellent? And now you have an opportunity to engage with Kate. I'm going to kick off with some questions to, lubricate the conversation. And we'd love you to participate as well. Let's really take this opportunity to extract some of, Kate Crawford's brain to participate. You're going to use Slido, so you can probably see the QR code up there, or you just go to Slido, Kirkdale, Slido and enter that code.And we'd love to see your questions. And I'll say them come up here on my lap.0:40:14 - 0:40:18
Kate Crawford
Isn't that sometimes part of the problem? Isn't it just nice to kind of talk face to face? But alas, no, we need.0:40:19 - 0:40:48
Natasha Mitchell
We really just wanted a roving microphone, to be honest. But this is all. Look, lots of questions coming in already. Fantastic. And if you're on the live stream, we really encourage you to participate too. We'd love to hear from you via Slido. So thanks so much for tuning in online.Let's look at Big tech. And it's it's a sense of it's ethical. It's ethical compass. It's moral compass because what was Google's motto. Do no evil or.0:40:48 - 0:40:48
Kate Crawford
Don't be.0:40:48 - 0:40:53
Natasha Mitchell
Evil, don't be evil. That was their initial motto, wasn't it? The mission statement.0:40:53 - 0:40:57
Kate Crawford
They faced that out about five years ago. Truly.0:40:57 - 0:41:18
Natasha Mitchell
They did indeed. And not only that, but they famously let go of key members of their AI ethics team who were real trailblazers in the field alongside Kate. And so how do you view Big Tech's engagement with the ethical questions?0:41:18 - 0:42:29
Kate Crawford
I mean, this is the, I think this is the problem sometimes that we expect ethics to be the answer to. So many of the issues that, I and others who were indeed part of that team, have been researching for so long, I, I think ethics are really important when you know, individuals are trying to guide their decisions.If you're making technical systems, you know what I mean? You'll be making decisions every day. Having those ethical frameworks are incredibly helpful. But I think when we look at the sort of concentrated corporate power that we see now in the artificial intelligence sector, this goes well beyond a question of ethics. This is a question of regulation. And accountability.And that has always come from the outside. So I think in many ways asking tech companies to, you know, ethically, you know, regulate themselves is simply not going to be enough. We've already seen so much happen. Obviously with the with social media, I think we saw how bad things can get, if you let companies just run, and I really want to see hopefully that with artificial intelligence we've learned some of those lessons.But as I say, regulating AI is extremely difficult.0:42:29 - 0:42:59
Natasha Mitchell
Regulating AI is difficult. Regulating big tech is difficult. Why do you think that is? There's there's something about this cultural dynamic that surrounds tech companies, the sort of techno utopia vision, these kind of boys with their toys and lots of women too. But traditionally it was boys with their toys. There's some sort of or collective aura that they have.Why have they been so resistant to to the possibilities of regulation?0:42:59 - 0:44:19
Kate Crawford
And I think you're right. There is certainly a sense of, an equation that many people have in their heads that technology equals progress, and that it's always going to be for the best. But I think there's something deeper as well, which is that so many of these technologies so artificial intelligence, of course, was first invented in the 1950s.And of course, you know, this this was a very slow and gradual set of technologies that have changed over time. But it was always very theoretical. People were building models in small labs. They were sort of testing things out, but that didn't really seem like something that you needed to regulate. It seemed exciting. It seemed, you know, progressive and experimental.But what has happened, particularly in the last couple of decades, is we've seen computer science and engineering super scale to the point where new technical systems are being live, tested on billions of people every day. And that is the big shift that I think we're not used to thinking about these companies as being core to everyday life, as impacting so many things that we do from how we see ourselves, we see each other and understand our cultures.That means that we have to think differently about the role of how we control, regulate and introduce guardrails on AI systems.0:44:19 - 0:44:43
Natasha Mitchell
We might come back to regulation, but we are providing them with our free labor in the form of data. And there's been an interesting quest to develop a movement. They call it data dignity. They call it data as labor. It's like a a union movement. It's sort of an industrial frontier in a way. How do you view that effort for big Tech to start paying us for our data?0:44:44 - 0:46:04
Kate Crawford
Yes. No. This is, the data dignity movement was started by two close colleagues of mine, Jaron Lanier, who some of you might know has written several books, and Glenn Weil and both at Microsoft Research. And, you know, I think it's a really interesting of course, Glenn is an economist. And so as an economist, I think, you know, we've all been seeing this, this problem of merge and his response was create a market.If we can have a market, then people can at least be paid something for their data. But sometimes I think a market isn't always the best solution. Partly that's because you might remember, this was about ten years ago. Studies were done to try and price how much your Facebook data was worth. And you know that economists spent years figure out the model.And guess what? You know, your entire Facebook account is worth something like two bucks, 50. And it's like, oh, I don't know that you want to make that trade. So I think in some ways, in the same way that, you know, human organs are not part of, trading market that you can buy, we see that as something that's really important to human dignity.I think in some ways data is very similar. It's extremely intimate. Our data tells us so much about who we are and what we do and what we want in the world. I think just seeing that as another economic trade isn't quite enough.0:46:04 - 0:46:13
Natasha Mitchell
In some sense, our whole, relationship to our own data has changed. So fundamentally, we've let go of privacy in all sorts of ways. Some that we've been cognizant of.0:46:13 - 0:46:15
Kate Crawford
I haven't yet. Have you guys? Well, I'm I'm.0:46:15 - 0:46:38
Natasha Mitchell
Not I'm hanging on for dear life personally. But but let's face it, people have been posting kids their photos of their kids on Instagram and Facebook for years, and they still do. You know, sometimes they put little love hearts over their kids faces, etc. and in some sense that feels like almost like a benign activity. It's not, though, is it?0:46:38 - 0:47:14
Kate Crawford
Well, I mean, I think that's the big realization that we've had in the last decade is that all of these services that we thought were communication platforms that you were, you know, sharing messages with your family and friends and sharing photos, you did not realize that you were, in fact, training AI systems. Every single one of those platforms has now become an AI company.So that's true of Twitter X, it's true of Instagram, it's true of Facebook. These are AI companies. So I think that's been the real shift. I think people are slowly becoming aware of this. But the issue is what do you do about it?0:47:14 - 0:47:15
Natasha Mitchell
What do you do for.0:47:15 - 0:47:51
Kate Crawford
Many people, you know, you have to use these platforms. It's, you know, the only way they can stay in touch affordably with people. In some cases, workplaces force you to use these systems. So in these sorts of cases, I think we we're looking away from this idea of putting it on the individual. Like you have to figure it out, you know, don't use ChatGPT or don't use Instagram or whatever.I think this that's the wrong unit of of change. We have to think about collective change and collective action, because it's only by actually moving as groups of concerned citizens that we've ever really produced major change. And that's what we're going to need to do in this case.0:47:51 - 0:48:02
Natasha Mitchell
Other examples, of groups or individuals who are doing really interesting, radical, perhaps work to challenge this corporate extractive model of AI.0:48:02 - 0:48:03
Kate Crawford
I mean, there are many.0:48:03 - 0:48:07
Natasha Mitchell
Some of those who were sacked by Google are doing some really great frontier work. Now.0:48:07 - 0:49:26
Kate Crawford
There's in fact, many of us have, you know, set up research groups and organizations, both inside universities and outside. But I think perhaps the, the, the more radical and perhaps the most effective thing that we've seen has actually been the first ever AI strikes that were this year. Remember, it was the Hollywood writers and the Hollywood actors, and they all went on strike to push back against, again companies bringing in AI models that were going to model actors faces and then basically own an AI model.So if you're Hugh Jackman, you know you're never going to have to act again. And you just keep appearing in Avengers movies forever and ever and ever. And so I think, you know, actors saw that, and then writers saw that. They saw that large language models could be used to edit their work, in fact, to whole cloth, create new scripts.And they all went on strike at an enormous personal cost. And we're talking about months and months and months of no pay. People really struggling. You know, I met with several people, at that time in LA who were, you know, just terrified. Nobody knew how long it would go on. You know, the entertainment companies were holding out, but they said, this matters.And they won. They pushed back, and those contracts were changed. So we've learned something again, it is a question of collective action.0:49:26 - 0:49:27
Natasha Mitchell
Wonder how long they've won for.0:49:27 - 0:49:28
Kate Crawford
Well, they've won.0:49:28 - 0:49:30
Natasha Mitchell
The slippery slope phenomenon.0:49:30 - 0:49:34
Kate Crawford
They've won for now. And they've also learned something really powerful, which is that you can fight back.0:49:34 - 0:50:24
Natasha Mitchell
Yeah. Yeah, absolutely. Let's weave in some questions. And along the what? Josie asks, what role should libraries uniquely play with AI systems in education? Let's pause on that one for a moment, too, because, you know, this institution is this incredible repository of knowledge. It's an archive of of history and knowledge. It's a it's a creative commons.It's something that tells the story of Victorians back to Victorians. It's such a vital institution. And now that AI tools can, in a sense, extract this knowledge and then spit it out again in another form, that's a form of curation of knowledge, like a library would do. So what sort of challenge does AI present to a library like this one, or indeed any library in the world?0:50:24 - 0:50:37
Kate Crawford
Well, this is actually very timely because what we've seen in the last year or so is tech companies approach libraries and offer to digitize all of their collection, which, by the way, is very expensive.0:50:37 - 0:50:39
Natasha Mitchell
You mean steal their collection?0:50:39 - 0:50:40
Kate Crawford
Well, interestingly.0:50:40 - 0:50:42
Natasha Mitchell
They would say the word for it.0:50:42 - 0:51:14
Kate Crawford
They would say digitize to create an exclusive AI model, which means that no one else can use it. And I think this is very dangerous because I think libraries have a fundamentally public mission, and it's for everybody. And so this idea that a particular tech company could say will just do all of that digitizing. Thanks. And it's just for us, this is happening like there are a lot of institutions who are, you know, financially under pressure who will agree to these kinds of deals.And so I think it's really important that we talk about it and.0:51:14 - 0:51:16
Natasha Mitchell
That that signals the end of a library.0:51:16 - 0:52:45
Kate Crawford
Not at all. The library will continue. The issue is that that corpus, all of the things that are in those archives and now feeding into one company's model, not other companies models, that doesn't mean that goes away. The library will continue to have its digital assets. It will continue to have its physical assets. But this what the question is, what is it contributing to?Is it contributing to something that benefits everybody, or is it contributing to a small number of people in a company who will profit the most from that particular collection of data? So there's something really big happening right now, which is what the libraries do. Do libraries decide to make their own large language models, which they can own and control and then make publicly available and try and deal with all of these problems like hallucinations and errors, which of course, librarians hate for good reasons, because we don't want to be giving out false facts and misinformation.Do we say libraries should be sharing it with all tech companies? Because, you know, that's just another part of the public mission? Or do we say that libraries sort of act as a kind of cautionary bridge, which is looking slowly to see how things develop, rather than having to jump on these things quickly, I tend to fall more into into that category.I think some libraries will and are building their own large language models. I think that's interesting, but it's very experimental. It's really difficult if you cannot be sure that that model is going to produce good information. So I think it's a moment for really testing the waters carefully.0:52:45 - 0:52:56
Natasha Mitchell
This is a great question. Many, are ready to claim AI is poised for failure. Are they correct or is this hype cycle different?0:52:56 - 0:54:36
Kate Crawford
You know, this is a really good question because what some of you, if you've been following closely in the tech press, there's a big debate that has really just sprung up very recently that this whole approach of scaling and scaling and scaling to get, you know, improvements and improvements has hit a wall. Basically, this approach of the large language model of just, you know, predicting the next word in a sentence has really kind of topped out and what we're starting to see in Silicon Valley is companies now start to say, oh, you know what?Some of those expert systems approaches might actually have something going for them. We need to have actual understanding, different approaches to how we engineer these models. So that is happening a pace that is already going on. I absolutely think that we have seen, just possibly the most staggering amount of investment, according to Goldman Sachs, $1 trillion will be spent in less than a decade on AI infrastructure.That's on those data centers that I showed you on the AI chipsets. It is just a staggering amount of money. Think about the Manhattan Project. You know, that was 40 billion. That's just like a tiny drop in the water compared to what's happening in a handful of years. So when you see that much money being spent and I do tend to say follow the money, that's an infrastructure that is not going to go away anytime soon.So even if this wave of, you know, chat bots that hallucinate is, is going to, I think, die down. And I think we are going to see a shift to new architectures, that big infrastructure that is being built that's not going anywhere. So I think in some ways we need to look at the deeper story that's taking place.0:54:36 - 0:55:15
Natasha Mitchell
Okay. We've got gazillions of questions coming up. You have to be quicker. And so I'm going to ask this wonderful audience on the live stream and in the room here at the State Library to vote, and some of you are voting. So I'm going to use your votes to drive what questions I ask next. But this one is a good one, I think, because I think it's easy to catastrophize about AI, and rightly so.There's a lot going on that is profoundly concerning. But someone asks, are the concerns many, many are keen to articulate about AI. AI only related to its human dimension. What does AI about protein folding or weather prediction sit in the realm of these considerations? Where does sorry pick your pattern?0:55:15 - 0:56:38
Kate Crawford
No, and that's absolutely right. And certainly, you know, there's an enormous amount of marketing that's been done around, you know, AlphaGo, which is the, you know, protein, which was actually pre the protein folding that was the go model that was produced by DeepMind. They then moved into this, you know, specifically protein folding work, which I think is extraordinary.But again, these, what we might call, scientific systems that live outside of social systems. And I think that's actually where that sort of uses of AI are actually are best put, I think the uses in weather data, in astronomy, in protein folding. Yes. Interesting. But the minute you connect it to, say, a human health care system or a criminal justice system where that data is already skewed from the beginning, it's skewed by its collection, by who are the people who are pulled over by the fact that, again, in health care, we know that men have so commonly been seen as the, you know, the people who participate in studies.We know so much less about women's health. That means we are training AI systems in health care on data that isn't showing you a representative sample at all. So this is where I tend to sort of draw a line is to think about, particularly when we look at these sort of, you know, AI for science, the further away it is from social systems, the more comfortable and.0:56:38 - 0:57:05
Natasha Mitchell
Which is interesting. I think there's a levels to interrogate there too, but there is some amazing work going on. I remember, spending some time with astronomers who were busy, using massive data set that they had of the surface of Mars to and they were using AI to analyze, the surface of Mars. And that was feeding into work that they were doing to try and understand the origin of the entire universe.So there is a really incredible it's an incredible tool.0:57:05 - 0:57:48
Kate Crawford
Well, let's let's be clear, though, it isn't at all. And this is the problem with the term artificial intelligence, because what they're using, and particularly with the rovers on Mars, is actually a very early form of image recognition. And it's actually very unrelated to what, for example, tonight's talk was about, which is specifically generative AI. They're not using generative AI for Mars, because you don't really want to create weird images of what Mars might look like.Alien. I don't think we need that. You might get a feel strange things. Yes, you get many different kinds of Martians. I think, you know, this is the problem of the term artificial intelligence is that it's sort of used as a cover to represent anything that's got something to do with deep neural nets. And even more broadly than that, people just use it.Oh, computers. It's artificial intelligence.0:57:48 - 0:57:50
Natasha Mitchell
We need to get fine grained to.0:57:50 - 0:57:54
Kate Crawford
Be much more nuanced and careful about what type of AI and in what context.0:57:54 - 0:58:30
Natasha Mitchell
So I'm going to roll a few questions into a row here. So everyone is voting for Kate. Are you frightened of AI? So we'll come to that because you voted. It's democracy here okay. The NBA is also asked what does it mean if the US goes in the opposite direction to the rest of the world on AI regulation?Does the nature of training? I mean, everyone's AI experiences will be set by the lowest standards. That's interesting. And then related how beholden I way in Australia to decisions made in the US for rolling. So let very start with the US and then we'll come to the frightened one.0:58:30 - 0:58:31
Kate Crawford
Right. Yes.0:58:31 - 0:58:34
Natasha Mitchell
So both actually I think they're both frightening.0:58:35 - 0:58:36
Kate Crawford
I mean frankly the US is a big.0:58:36 - 0:58:37
Natasha Mitchell
Part.0:58:37 - 0:59:20
Kate Crawford
In asking me if I'm frightened of the United States. Okay. So, you know, we are looking very much at, you know, two major AI superpowers in the US and China. And the concern that we're that we're seeing is that the rest of the world is being treated like client states. You know, that you're just sort of the end users of these systems will extract data from you, but you don't get to have much of a say, and you're not benefiting from the enormous amounts of money that are sloshing around.For AI investment. I think that's a major problem. I actually think the EU is doing something quite interesting by saying we are going to lead in regulation, even though we may not have, you know, a very large AI industry.0:59:20 - 0:59:32
Natasha Mitchell
They always have. They've always taken the precautionary principle on all, all tech regulation, and they're often shunned for that by the rest of the world, including Australia. They're seen as being risk averse.0:59:32 - 1:00:01
Kate Crawford
And and I think that's short sighted. I think in some ways, you know, the EU is is playing the long game here, which is to say, I mean, honestly, if you read the AI act, it's not setting a very high bar. It's literally like you must publish where the data comes from. You must tell us, you know, how the models work, and then we have to assess them if it's going to be high risk.I mean, that's what I just call reasonable. I mean, you know, this is what we're hearing a lot in the US is sort of decrying of use a heavy handed regulation.1:00:01 - 1:00:03
Natasha Mitchell
It's just streaming innovation.1:00:03 - 1:01:09
Kate Crawford
Extremely light, folks. I mean, it is as light as it gets. And so I think in some ways, if anything, you know, that's a concern about EU, is that the EU didn't come in strong enough. And in fact, you know, we can get into the weeds. But what happened in the last two months before that act was passed is that it got watered down by several tech companies based in Europe.So, you know, we are looking at a problem right now, which is a real unevenness. And so I think, you know, regulation is going to be important. And countries do have a right to say we want to protect our citizens from, you know, what happens with this technology. I'll give you a personal example. When I was flying in a couple of days ago, I always opt out.You know, if you get those little screens when you sort of get on a plane and then you can take your photo, and I hope, you know, you can always opt out when you get that on the screen. Also, I thought, so I said, oh, you know, I'll opt out. And he said, oh, you're an Australian citizen.And I'm like, yes, yes, I am. And he's like, well, then you have no choice. And I'm like, well, I'm also an American citizen. He's like, well, if you're an American citizen, then you can opt out, but not if you're Australian. Wow.1:01:09 - 1:01:10
Natasha Mitchell
And guess what does that say about Australia?1:01:10 - 1:01:22
Kate Crawford
Well, it's because we didn't have enough people pushing back on facial recognition legislation and we have more to do. But we've got, you know, we're just like others who who are doing that work here. And I think we didn't.1:01:22 - 1:01:46
Natasha Mitchell
Push back about CCTV either. It just appeared unlike when the UK was actively pushing back against that tech arriving on every, every street corner. Are you frightened of I, I hate that question in a way, but I'm sorry to say that it's a bit judgy of me, isn't it, when it's your question. It's just that I know that, RAF on ABC Local Radio ask you the same question as well.So what's your answer?1:01:46 - 1:03:29
Kate Crawford
Well, let me let me ask you, put up your hand if you're frightened of. I.Oh, okay. Right. So I'd say that's about half the room. And for the, the second question, put up your hand. If you're all really excited about AI. And I love that it's almost the same people. Okay, so I think that tells you something immediately, which is that we're aware that there is something really interesting and exciting that could be done with this technology.But what we're seeing at the moment is the way that it's being deployed. That is what concerns me in many ways. It's really less about the technology and more about the corporate application. I'll give you an example. So, a study just came out this week showing that call center workers in the Philippines, around 85% of them have to use AI in their jobs.They're forced to use it. And initially they said it was. They were really excited. This is going to be easier. It's going to, you know, help with all sorts of issues. Until their bosses were saying, well, now you have to do ten times the amount of calls and ten times the amount of work because you've got this tool helping you.So I think what we're seeing here that's less about AI and more about the way that it's feeding into existing profit and corporate structures so that is what keeps me up at night. And it's it's also this question of the AI industry itself being so concentrated. You really have to go pretty far back in history to find a similarly concentrated industry.It's really basically the equivalent of big oil or even the railways in the 1800s. It's we're talking about vanishingly small numbers of companies really, three in the US who have full planetary cloud and two in China. That is a very small number of people with a hands on the levers.1:03:29 - 1:04:08
Natasha Mitchell
At the end of your presentation, you said, you know, it's not a fait accompli. You know, technology, we do have some sense of control. It's hard it's hard to realize how that might be possible. And there's a question here. With the speed at which AI is developing, what is the best way to educate people to have the skills that they need to meet this demand?That's a related question. But also, how do you envisage communities being involved in driving AI in shaping an AI influenced future, that that in a way that creates a kinder and fairer and more equitable AI?1:04:08 - 1:05:22
Kate Crawford
Yes. I sort of wonder. This is this is a really common sort of thought that the only way to beat AI is to build more. I just build a kind of a sweeter, gentler AI. Yeah. And I think in so many ways, you know, we're applying AI in places where it's really not that helpful. You know, this is that this is the phase when a technology is new and you've got a hammer and everything becomes a nail.It's like, well, just put some AI on it. It'll be great, and people will be excited and people will fund your company because you said AI three times, you know, that's the stage that we're in now. I think the solution to that is not to go and build more or different AI. It's to think about what are the problems that we're trying to solve.In many cases, it's doesn't actually require AI at all. So it's about having that type of analytic capacity around where we're at. We we are we are in an extremely difficult time. Historically. We're seeing the rise of fascism. We're in a climate catastrophe. We're going to be seeing large scale population migration. These are questions that in many cases, can be answered in ways that don't put technology first.Technology can be part of it, absolutely. But it's not the quick fix of all things. So I think changing that mentality is so important.1:05:22 - 1:05:36
Natasha Mitchell
But can I fix some of the intractable social problems that we're facing? For example, someone's asked here, how can we make I serve 8 billion people, not just eight billionaires. Quoting Rachel Caldicott.1:05:37 - 1:06:40
Kate Crawford
Yes, indeed. So in many ways, I think this is this is the big core question, which is the deep inequity around who is currently benefiting from artificial intelligence. So the question is how radical do we want to be? So of course, if we go back in history and because I've been, you know, doing these this sort of very deep history project, there are times where technologies were seen as so important and so core to the public interest that they were nationalized.It was like, you've built this, this is great, will recompense you. But this is a national this is like an important national infrastructure. Now, you could just imagine that happening in Silicon Valley right now. I mean, it's it's always impossible to wrap your head around, but we can think about these ideas of infrastructures, and we're already starting to see nation states start to say, we're going to start thinking about our own territorial AI boundaries.Now, that comes with good and bad. But I think this idea of how do we start to think about equity? That is going to be the key question for the next decade.1:06:40 - 1:06:42
Natasha Mitchell
What does that look like for you?1:06:42 - 1:07:33
Kate Crawford
Well, I think for me, it it, you know, I'm really interested in how we start to use collective action, use government pushback and specifically strong regulation. And then hopefully, I think, empower the next generation of people who are going to be building these systems and tools to think differently about who they're actually serving, the problems they're trying to solve.Because I do think that AI is going to be extremely useful in solving parts of issues like how do we start to deal with climate data in a shifting planet? Absolutely essential. But then who gets to use that data? I'll give you an example. So one of the things that has been threatened under Trump's next administration is that he is going to demolish NOAA, which collects all of the climate and atmospheric data.So if you don't have anyone collecting the data, do we even have a climate crisis?1:07:33 - 1:07:39
Natasha Mitchell
I can't actually what I know is that know what the influence and impact of that scientists at Noah is astounding.1:07:39 - 1:08:24
Kate Crawford
Astounding. Imagine losing that. So I mean, I think what we're going to start to see is such a stripping back of important public data assets and resources that that's where I want to see people stepping up. How do we start to protect those assets? How do we actually make that something that people can still use and access? And this is also where libraries, I think are going to be so important because libraries right now are under attack.I can't even tell you how bad it is right now in the United States. It's not just book banning. It's not just, you know, fights over gender neutral bathrooms. It's the right of the library to serve a diverse population that's under attack. And we have to fight back.1:08:24 - 1:08:46
Natasha Mitchell
The late Kate B asks. Thanks for your question, Kate. The late Columbia, David Columbia wrote extensively about the links between generative AI, white supremacy and the far right. Similarly, in your book, you draw current connections between 19th century race science, and the use of data sets in AI. How does I appeal to the interests of right wing tech billionaires?1:08:46 - 1:10:21
Kate Crawford
Well, we're certainly seeing it happen right now, aren't we? And this is, of course, something that, you know, David, who was an extraordinary scholar and he was a it was a very serious loss. But he has been writing about this issue for a very long time and looking at not just the far right wing use of these technologies, but that some of these technologies are actually set up for authoritarian control, which is to say that, you know, just a small number of people can decide, you know, what types of posts will proliferate on Facebook and X, what types of political content you will see.You know, what sorts of, you know, videos will come up on TikTok that will tell you which candidate to support. This is extremely powerful stuff. These platforms, I think, have had an enormous impact, particularly on the US election this year, and they will continue to do so. And I think what worries me is not just that it's the FA sort of we're starting to see enormous sort of, you know, far right traction in these platforms.It's that we are seeing these platforms go dark for researchers. So we've had over the last decade partial access to sites like Facebook and Twitter, researchers could, you know, ask for API access to see some of the messages, to see what people were, you know, being exposed to over an election that was incredibly important. Many researchers, many of whom are here in Australia, have done incredibly important work studying social media platforms because they could get API access that has been shut off almost in.1:10:21 - 1:10:26
Natasha Mitchell
Ways we can't even interrogate. We can't even understand what these technologies are doing to us.1:10:26 - 1:10:27
Kate Crawford
Precisely.1:10:27 - 1:11:01
Natasha Mitchell
Yeah. I'll roll these two into one. As you quote the legend Andre Lorde, Audre Lorde, baggy pants. Excuse me. The master's tools will never dismantle the master's house. If you could rebuild our AI systems, how would you go about it? And this one, what is the role of the human in the system? How can we ensure that there is always a human interface, preferably not an exploited human, to ensure the nuance between information and knowledge?Wow, these are questions. Questions I wish I could read them all. So superb.1:11:01 - 1:11:09
Kate Crawford
I think we all have to go out and sort of, you know, have a have a beer after this. And I keep having this discussion because you're asking exactly the right things.1:11:09 - 1:11:19
Natasha Mitchell
And on this, how can we give control, agency and value back to the individuals and citizens who provide AI's core data? Stephen Johnston Sportsbet always connected. These are.1:11:19 - 1:11:51
Kate Crawford
Fantastic questions. And what they're really pointing to is this question of agency. Who has it, how do we get it? And particularly when what we've seen is, I think, an erosion of our agency as we've handed over so much of our not just data, but something deeper. I think we're losing and in fact, outsourcing our discernment, how we're making decisions.I mean, I've seen many people, you know, really start to use, you know, ChatGPT and LMS really to help make really important life decisions.1:11:51 - 1:11:53
Natasha Mitchell
Lens being large language.1:11:53 - 1:12:32
Kate Crawford
Large language models. Yes. And so, I mean, I think we're really outsourcing our cognition and we don't know what that's going to look like. I mean, I think this is something where it's too early to say, but it certainly in this phase we have to acknowledge that we're seeing something very profound in terms of that shift. And so this is why I think that, you know, the Audre Lorde quote is so important.And so often cited, that the master's tools cannot dismantle the master's house, which is, you know, really taking us back to this idea that just building a kind of a kind of better AI is not going to solve the issue. We have to think at some of these core questions around agency and around discernment.1:12:32 - 1:12:58
Natasha Mitchell
I have to ask, was this funny? And have you read and seen Yanis Varoufakis? And he's techno feudalism paradigm. He's written a great book called Techno Feudalism. He's a former finance minister for for Grace, and he was a professor in economics at the University of Sydney for many years. As well. And he posits that, the current mode of operation with big tech and us handing over our data is that capitalism's dead.1:12:58 - 1:14:26
Kate Crawford
Yeah. I mean, it's really interesting. It's a fantastic read and it's a brilliant point. And he is someone who I think has been, again, as an economist at the forefront of saying, you know, we're reaching the end of how current economic models will even help us understand this. A question that, in fact, McKenzie Wark, wrote, you know, years ago, which I try that was a fantastic way of putting it.Is, is this capitalism or is it something worse? And I think that's, you know, I think that's the kind of question we're really facing now. And I sort of tend to think about it as almost like feral capitalism. I mean, it's still it's still capitalistic in terms of its structure, but it's it's feral in the sense that it's completely lost at any kind of, you know, protections, guardrails, restrictions that were more true, at least in some limited ways in the 20th century.I think that's just been blown away. We can go back, you know, historically to say why from Citizens United and beyond. But we are looking at a moment now where and it's still absolutely staggering to me to look at the difference in wealth inequality in just the last five years. I mean, we are talking about a divide which is increasing so rapidly where billionaires are just becoming so much richer, and the rest of the world is now seeing a massive erosion in their ability to even buy food and pay their rent.That to me that we're talking about what frightens us, that frightens me.1:14:26 - 1:14:46
Natasha Mitchell
I know you love nuance and shades of gray, but I'm going to get you to do yes. No answers to some of these, and then we'll wrap on regulations. David Johnson says, how can I, can I well, how can I believe I'm going to ask it like these? Can I be used to say that broken democracy. Oh, you cannot you cannot say that properly without a yes or no answer.Oh, that's too hard.1:14:46 - 1:15:15
Kate Crawford
Yes. Oh, baby, you're killing me. That's too hard. Can it be used? Look again, we're going to the technology to try and solve the problem without asking what the problem is. Like. The problems of our democracy right now are many and varied, but one of the things that we are saying is that the use of AI to manipulate populations and to restrict and shift what they can see is one of the biggest problems we face.So, I mean, I think I would look at that as a problem first before we decide on what the solution.1:15:15 - 1:15:24
Natasha Mitchell
This could be a yes no answer, but you can always be a bit loosey goosey. You're at it. Is I really intelligent? I know, or is it just reiterating something done before?1:15:24 - 1:15:27
Kate Crawford
I can't give you a hard no on that one.1:15:27 - 1:15:52
Natasha Mitchell
Oh, it looks so many great questions. Thank you so much for bringing your collective consciousness to this event. For us. It's been great. We'll have a squeeze at some of these afterwards, but, let's look at regulation to wrap what if there was one key thing that the Australian government could do? For example, although I think this has to be an international agreement, what would it be tomorrow?1:15:52 - 1:17:28
Kate Crawford
Well, you said two things that are really important. The first is this idea of global cooperation. Because honestly, at a certain point, this idea that each nation state can have its own AI regulation is going to fail because all that an AI company would have to do is move to the country that has the lowest regulatory hurdles, what is otherwise known as regulatory arbitrage.It happens and it will happen again. So we have to be thinking internationally. We have to be thinking about types of global cooperation or it's simply not going to work. And interestingly, there's a new, a new international body on developing shared international AI frameworks that's having its first conference in Paris with the, president Macron's AI action summit, which is in January.So, if people are interested, you know, that will again be public. And that debate is very active right now for Australia. Look, I'll be honest, I'm, I'm deeply connected to the environmental movement here. And to see what's happening, particularly in terms of the the extraordinary and terrifying changes in our climate and what happened with the fires in 2020 and what's happening with water.It to me, if we can at least start to have absolutely enforced transparency around how much of our resources are being used in AI models and also hard set limits, absolute limits on how much can be extracted in terms of water and energy. I think that that would be the first thing that I would do. Of course, there are many more, but let's start there.1:17:28 - 1:17:34
Natasha Mitchell
Look, I just got to squeeze this 1 in 2. I know we've got a rep 30s ago. Is Harvey here all night?1:17:34 - 1:17:35
Kate Crawford
Are we going to just keep going?1:17:35 - 1:17:46
Natasha Mitchell
Is there a risk that if we don't share our data, our ideas, our lives in a way that can be used for training AI, will we end up being the ones not represented by it in the future? That is an interesting question about diversity.1:17:46 - 1:19:45
Kate Crawford
Good one. And this is an extremely an extremely topical one. Because of course, many, language groups are having this debate right now, which is do we want to hand over all of our data to companies? Because it will be good for things like translation software. AI is great at translation. Speaking of things that it's really good at, or are we giving away our heritage?And I think this is also really important for First Nations languages, you know, is that something that you want to just be handing over to tech companies? So this idea that somehow you lose representation by not being, you know, in AI data sets, I think it's actually, a really complex double edged sword because we've heard it before.And I'll give you a really horrifying example. So the data sets that I was showing you tonight, one of the sort of very obvious problems that was discovered by, doctor Joy, Bill and Winnie and Timnit Gebru, who wrote a very well-known paper, which showed that facial recognition systems were actually not performing on people with darker skin tones because they were all trained on white faces.So that's created this kind of this surge in the tech industry to go and record as many faces of people with darker skin tones to get them into the training data sets. Well, first of all, a whole lot of really kind of problematic things happened there, including, you know, lots of people having no idea about why they were, you know, having their photos taken.And then secondly, of course, these are now feeding into more accurate facial recognition systems, which we know raise all sorts of very real civil society threats. So I think the idea of representation for whom and in what are the questions that we have to ask. Because I doesn't necessarily mean that your interests are going to be actually respected.The question is, what are you actually giving up and what are you asking to get back?1:19:45 - 1:20:47
Natasha Mitchell
And it's no coincidence that those two women that you mentioned, one is now running the Algorithmic Justice League, and the other, who was let go from Google for her brilliance, is now running the Distributed Artificial Intelligence Institute. So may the revolution continue. What a great honor it's been to share this time with Kate Crawford and with all of you.I want to say that we honor the Ellen Helen Macpherson Smith Trust for its support. It's valuable, vital support for an event like this tonight. But we want to thank the Auslan interpreters for their heavy labor tonight.The the incredible work of specialist Auslan interpreters bringing knowledge to all. And I want to thank you for joining us on the live stream tonight. And thank our wonderful hosts and the producers of this event. They've been wonderful to work with. But most of all, please thank Kate Crawford.