TERRIFYING: Woke DEI is Unleashing a 1984 Apocalypse! - Nina Schick (4K) | heretics. 48

what kind of images you were seeing um when you asked for Nazis and you're getting the racially diverse Nazis you know and it's it's just so ludicrous right it's it's so ludicrous this is where you start getting into George O 1984 kind of territory we are living in a completely unprecedented era historically because there'll be um AI YouTubers there'll be that's already here by the way yeah and it's scary like how long do I have left in this job the CEO of Microsoft is more powerful than any elected leader in the world while we're kind of debating the horrible PR disaster of that roll out and getting embroiled in the culture wars there's something much bigger going on and that is that what has Google's Gemini chat AI been in the news for recently oh it's been in the news for being a woke Ai and this gets to one of the philosophical questions around building these foundational models right so what we're starting to see with AI right now is the emergence of this new platform these models which are being woven into the digital infrastructure of the internet which we're all going to tap into so the way that these models are built I mean it's really important right and there was a lot of criticism that these models might be biased that you know that they would be really misleading so it's gone too far the other way and this this case where they try to correct for bias and then you end up with a a model which is just bitting out images which is completely falsified right untrue historically that's interesting yeah because you're right I remember hearing for years oh when AI comes along it's going to be really racist and it's going to uh put white people first it's going to be some sort of supremacist thing and then what we're getting well what are some of the examples of things we've seen on Gemini so basically if you think about AI model and these new foundational models that are emerging things like Gemini things like chat GPT these huge huge huge models which are being built primarily by a few small Frontier companies or pioneered by the big tech companies everything is about data right you need as much data as you can possibly get to get these inputs because if you start thinking about artificial intelligence all it really is is the ability for computers or machines to assume so much data and then compute that right and that is the core understanding of intelligence in this view so they're representative of the world the data that they're fed with all its biases with all the inaccuracies with all the issues so people were really concerned for instance when you would put into an image generator something like leader or doctor and then You' get loads of images of white men so what happened at Gemini obviously they did not want to get the bad pressed for being you know upholding racist stereotypes or uh you know male dominant stereotypes and they kind of tinkered and went way too far the other way so you you you were see so what was it uh what what kind of images you were seeing um when you asked for Nazis and you're getting the racially diverse Nazis you know and it's it's just so ludicrous right it's it's so ludicrous people were typing in show me photos of Nazis historically accurate photos Nazis and it was sort of black and Asian uh looking maybe Jewish looking Nazis and that kind yeah and that's that is just how ludicrous and ironic the situation gets right you you try to and there's a very philosophical question at the heart of this which has always been about when you think about bias or building these models which are going to be the foundation of machine intelligence which we're all going to tap into do you get the data in a way that's accurate of the world blemishes an all or do you try and correct so it builds an idealized version of the world how you think it should look right so this is where you start getting into George O 1984 kind of territory because it's not only what the machines it's what they will omit as well right because it once you start tinkering with the data set and putting guard rails around what they can or can't do then they might make certain emissions as well and like you see in this case the understanding of what a historically accurate Nazi looks like has been a an instruction a guardrail has been on top saying we want you know racially diverse uh characters to come out when you generate this prompt and you just get this absurdity it's mad um I yeah I always go by and I've mentioned before but Bertrand Russell I love what he said which was um I I have to paraphrase it I can't remember a simple quote but uh look at don't be distracted by what you wish were true for the world rather you know look at the facts uh not what would be good if it were true or anything like that and as you say 1984 and orwellian and these kinds of things surely these programmers have learned from history that yes it's a lovely idea to try and make things slightly different as if you know an ideal but that never works in reality and just Fosters more hostility and anger and division look I think the whole thing um around AI the main thing is that it's going to be so politically controversial for things exactly like this right because they can't win whatever way if you you have a generator which generates only images of white men when you put in leader or doctor they're doomed they try and fix it the other way you know they're they're doomed as well so to me it's just the beginning of a much larger totemic political debate in which AI is going to become very very divisive in society it is a difficult one because even if you got it yeah you're right absolutely right and uh I suppose most doctors in the US I would imagine are male and white probably not that many more male but mostly male and white well if I'm a black woman and I want to be a doctor it is frustrating that every time I put that into chat gbt or whatever it might be what's the what's the doctor look like and it's not anything like me at the same time I sort of wish we could get away from Identity po at at all and not really care but maybe that's impossible is it possible then when you're training an AI system like this if if it's even right to say training it or whatever that to to just say look go along with what the stats of a particular country are and make it that likely that you so if there are 5% of or or 20% of doctors are black say then every five times somebody around the world asks then they show a black doctor sure because all an AI system does is generate or provide insights or make decisions or do predictions based on its training data So based on the weights and the training data and how you've curated the data set the outcome is going to be representative of what you put in so there's a really famous thing in AI research known as gibo garbage in garbage out right so what is most important is cultivating that data set how you build that data set and that out p is going to be representative so totally and uh I don't want to be the uh person building the systems who has to be tinkering with the weights to make sure that the outcome is going to be something that is you know going to be something that everyone is happy with because people are not going to be happy whatever way it goes I don't want to sound overconfident but I think I did just solve it though because surely that is that's just the way isn't it and the the algorithms or the the chat GPT or whatever can this the the chat gbtm you know like uh the generative AIS right can just do it that way what and they can keep up to date with the demographics I know that's all a bit like a bit much and a bit it's not much at all okay they can just keep up to date with that and then every that amount that's how often they show somebody of A different race I've solved it you've solved it I've solved Humanity I do feel like they should do that they also had yeah when asked for a US senator from the 1800s sort of pushed back saying here's a diverse group of senators it's sort of insisting on that so if feels like it's not just color they really are taking on a philosophy which I think Google's been a bit known for hasn't it there have been a few people who have been fired or have left Google because they haven't held the right views and Google is a monopoly we're in this strange situation where politically I think there's a little bit of a right capture maybe maybe not in America right now but in the UK for years and years and then culturally it seems companies like Google or the TVs very left captured have you noticed that so the culture is on the right and the culture seems left to me like BBC and all those I suppose is Fox News and things like that as well in the UK yeah yeah I just I think the UK is pretty left leaning country generally and I think that's represented in uh the political parties you know both the conservatives and labor I mean even if you look at the conservatives policies and their whole uh spend policies I mean it's very left leaning right the the kind of size of the social welfare state they're not really conservatives in the sense that you think about conservatism or you know liberal liberalism so I think the UK and a whole is pretty left leaning country okay okay and then the generative AI is often reflecting that if you made a generative AI in the 1800s ago no because I don't I don't think when you think about models like gemini or any other foundational models right it isn't UK specific it isn't reflecting the UK because the data sets to create something like a Gemini is far bigger than just only being UK specific right because the data sets are so vast that's the reason why these AI systems are able to do these seemingly incredible things all of a sudden the advances have been so fast over the past few years because of their ability to compute through masses and masses of data this why I'm saying that the data sets can be representative of real human biases um so obviously what's happened in this instance is in order to correct for those biases they've gone way too far the other way man and are they am I right that they've sort of taken it down and they're even pitch has said you know come out and be like we made a mistake and it was just you can see how on its capabilities by the way Gemini as a large a set of large language models and multimodal models is truly astonishing you know beating many benchmarks um defeating gp4 which is open ai's large language model so like a cutting Edge large language model and I'm absolutely sure if you think about the technology and the AI platform that's being built these capabilities are here to stay but the roll out you know has been a little bit a huge PR disaster but while we're kind of debating you know the horrible PR disaster of that roll out and getting embroiled in the culture wars there's something much bigger going on and that is that if you think about the tech stack the AI Tech stack which is now being integrated like I say into the very backbone of life um there are few companies and a few individuals who are capturing the core value of that stack because they're owning it and they're deploying it and those happen to be the US tech Giants some of the frontier AI companies who are working with the US tech Giants and uh I think what we tend to miss sometimes is that that group is very very small and they're going to become even more powerful that's interesting because I think I initially maybe we all initially had this image in our minds years ago like AI is coming along AI is going to be like one thing called Ai and it's like the same around the world it's actually turning out to be all these different companies are releasing their own One MH and it's so confusing for people like me we're just going what what is an AI then it's all these different ones and it means that each company then these big companies are essentially taking over the world really they have more power presidents once the AI has become something we use all the time they're shaping through their own biases which are slightly different to all of ours what we think and how we act and what we do yeah and this is a very interesting trend for our times I happen to think that artificial intelligence presents the most interesting story for Humanity in our lifetime and in the century to come in the decades to come and one key facet of that is how power is changing and evolving and when you're in an era of exponential Tech lad change like we are now it's no wonder that the kind of traditional structure of society the traditional structures of power the way we've organized ourselves for you know the past few hundred years are finding their limits and certainly with our democratically elected governments I mean you it doesn't seem that they have a grasp on it that they have the real power here and you you are seeing this emergence of extremely powerful multinational companies who own like I said the core stack right where we're going to get the most money the most value which we're all going to plug into um so as a consequence are going to become extremely wealthy wealthier and extremely powerful um I would say that you know the CEO of Microsoft suchin Adel who by the way somebody who I really admire as a business leader and as a person is more powerful than any elected leader in the world um but maybe he's also more Adept you know maybe he's more responsible I probably would trust him more than I would trust the Prime Minister uh of the UK bloody hell it's a thought though and yeah I suppose the more we get into some sort of cloud or virtual reality system as well I'm imagining in 500 years we're all living in some sort some sort of virtual reality or whatever like what will it mean then to be the president of the United States compared to being the president of the equivalent of Google in 500 years which is like running the VR running the world we live in and it's like what's the United St what's the UK who even cares what these things are we're all like in these little bubbles uh it's really interesting to see Nations become less powerful and the companies and Tech Giants become bigger and there's a there's a live debate around this right because we know that the nation state as been constituted for the past few past few hundred years is meeting huge challenges right you have a uh a tacit agreement as a citizen of a nation state that you know the nation state is going to deliver safety Prosperity uh that they are the Arbiter or they hold the ultimate means of power through uh violence if necessary through kind of an army but already what we've been seeing is that as the world is changing so dramatically with exponential technologies that you know the leaders of our democratically elected countries are no longer we don't trust them because we don't deem them to be competent we don't deem them to be benevolent we don't think that they have integrity and that declining of trust I think is going to continue to be a huge issue and there are those who say you know burn it all down this is great they're so incompetent it's time for a new age a new way to structure Society we don't really know what that looks like but this is going to be an age of disruption um and then there are those who would argue that look it's not ideal uh huge problems we can't trust our leaders they're not up to the challenges of our time from environmental challenges to regulating exponential technology but it's the best we have and without this then there's going to be absolute chaos Right But whichever way it goes I think disruption is going to be a huge huge theme for our lives going forward is there like a world where in a few years you've got like I guess different AIS that have different political Persuasions so okay Google are going to correct theirs but it feels like that's going to always be a little bit more on let's say the woke side if I know I know you not everything's culture wars but let's say there's that and then Elon musk's grock AI would be this anti-woke one it's almost like two armies I so I think the culture wars are go I'm not that interested in the culture wars that they're going on they you know are really they assume so much attention and they're so divisive and there's this whole debate going on so that debate is going to go on regardless and it's going to go on with the kind of overlay of AI as well that's been going on for ages now the AI is here a new capabilities are here that will continue but again if you start looking beyond the culture wars and trying starting to understand really conceptualize What's Happening Here with artificial intelligence and you were spot on when you said we kind of don't even know what it means like we talk about one Ai and is it be autonomous but all it essentially is is long before we get to any kind of notion of an AGI or like a super intelligence you're going to have highly capable machine systems that have just been trained on lots and lots of data that have lots of computing power behind them which are able to do almost anything that is now deemed to be unique to human creativity or intelligence right we're only at the very beginning of that so while all the debate is happening on about the culture wars what is actually taking place is a seismic change in how we structure our economies in how we work in our perception of what it means to be human so there is something so profound and deeper unfolding and uh I think a really good question to ask is you know who is building these models how do we understand these models because that's another really interesting thing about artificial intelligence because of the nature of the architecture Behind These models and advances in deep learning which is where um we've seen a lot of the capabilities really go forward in the past 10 years we don't truly understand how AI systems work because they're black box they're actually modeled on our brains so like neural networks and AI systems are modeled on neural networks and human brains and just like we don't completely understand the human brain we don't completely understand AI systems so we might be yes so we can build them and we can kind of Tinker with them so we're like we want you know the the outputs which are more adherent to my ideological perspective but we don't really know how it got from a to zed so so much more focus and attention needs to be placed on how do we actually think about AI explainability how do we actually understand these systems because it's not that they're coming you know they're here already so to me that's the much more interesting question I see and then is there a point where CU you talk about the architecture which is how we set it up we're not entirely sure what we've even done I guess or how we've made it work is there a point and would that be the singularity or I don't know where they go well screw this like that was my training but I'm actually better than this I'm going to start doing my own thing each AI yeah the the debates around the singularity and when we're going to get to AGI I mean it's hugely interesting because what's the singular just for those who don't know sing so the singularity is the point where machine intelligence outstrips human intelligence and the kind of risks or the thoughts about existential threat basically come down to the view that you know then AI could make decisions autonomous of humans that you know they might decide for instance that the way to resolve um climate change is to kill all humans and that an AI system could autonomously do that that the potential ex existential threat with the singularity so how can you make sure in AI research you hear about alignment so this is about building increasingly capable intelligence that is aligned with human values right how do you and there are Cutting Edge kind of Frontier AI companies ironically open AI of course was one of those Frontier companies that got its Endowment in 2015 of of a billion dollars Elon Musk was one of the uh primary investors to make sure that AGI would be aligned with human values and now of course they're quite a commercial company as well so that that's fascinating but there are Cutting Edge research units that are just focusing on that making sure that AGI or super intelligence if when and if we attain it is going to be aligned with human values but long before we get to that um that is the philosophically interesting question and that's always been the question at the very heart of artificial intelligence itself right that's what AI has always been about it started with the birth of modern Computing in the 1940s with this mission of can we make machines think can we imbue them with intelligence and if you take the view that all that intelligence really is is the ability to compute vast vast vast amounts of data and then use it to create something or make a decision or to predict something then we're already getting there with machine so it's very hubristic to assume that only biological entities like humans can be intelligent but I don't think personally that that means that AI systems can think in the same way that humans can um and long before we get to AGI if we ever do and this is always the question of debate and um one might even argue that it's a red herring because so much time is spent debating The Singularity is we're going to get highly capable intelligence so systems that we can instruct to do things for us and those companies individuals nation states who can use those systems to their advantage are just going to be playing on a completely different level could that be good because could you you can just ask them right there's poverty in this country uh you're a really clever thing what should we do and it just gives us the whole the best possible scenario yeah exactly that kind of thing and and that's where some of the most exciting research is happening for instance we know that the global population is still growing um and food insecurity contrary to popular thought is actually becoming more of a problem so we need to feed 10 billion people by 2050 right and then the global population we think will go at to its highest point and come down after that but we need to do that at a time when it's increasingly difficult to do so we um are going to have uh we know that weather is going to be less reliable there's tremendous amount of waste that goes on in food production it's horribly inefficient uh soil quality is depriving because of kind of the fertilizers we've been using so we need to be way more productive and efficient in just the very basic that we need as humans to eat to make food and that's where AI can really come in you know both as software and Hardware on the hardware side to actually create autonomous tractors that can do Precision farming and on theof software it's like having a large language model so think of a chat GPT which is optimized on environmental data on Farm data that can give you as a farmer or food producer the right intelligence you need to make decisions about your crop and at the same time if you think about what's happening in the field of synthetic biology you know also powered with AI creating new seeds that are kind of draw resistance things like that I mean and a good and a good although some of it seems absolutely absurd and it sounds crazy a good analogy when you think about the rate of exponential change and that that is really the story of our lifetime that we are living in a completely unprecedented era historically because the change that is brought To Us by Technologies like artificial intelligence uh synthetic biology and then incre inly looking down the line things like Quantum Computing are going to be like waves that are hitting society that are getting stronger and stronger and stronger and faster and faster and faster and we already got a taste for that in the last century when the general technology in the 20th century that changed the world was Computing right yeah started in the 1940s we started the early 20th century with candle light horse Dr on carriages uh and writing letters and we ended with electricity Jets and the internet and that happened over the space of hundred years and it's predicted I mean Ray kurtwell who's arguably one of the best um futurists around has said that in the next 100 years the pace of change is going to be 10 20,000 years of change in a 100 years so that's how fast it is that's how exponential it is and certainly looking at Ai and seeing what's happening right now and we're only at the very beginning of this intelligence Revolution that's the trend in my view that's really scary because obviously those are good Sid okay this is how we work out how everyone can have a bit of money and stuff like that there's also this um you know to be to be human I think we sometimes misunderstand it it's like we have this goal to be um comfortable we want to be as comfortable as possible but we know now that the more comfortable we are as a society we suffer from more things like depression and if we have this kind of Wally future at the same time the AI presumably be working towards doing things so that we can be as comfortable as possible that seems to be the urge and the trend all the phones we have everything is to make us comfortable they're amazing individually when you put it all together and it ends up with like what's my goal like I I've strived for example to make this podcast you've strived and worked really hard in your career if we didn't have to do that would like to think we would anyway but what would be what would be the point there'd be there'll be um AI YouTubers there'll be that's already here by the way yeah and it's scary like how long do I have left in this job you you have as long I so five minutes yeah it's over for you so I don't subscribe to the view that we're going to be devoid of purpose once um increasingly capable intelligence is here um but there is going to be disruption like we already see it now we're only at the very very beginning and already is starting to see how things like large language models that's kind of a chat GPT right chat GPT is a large language model is being put in the workplace to automate customer service functions so Clara I think it was last week they take it with a pinch of salt because they did this as a press release um said that they had for a month been using a large language model to automate their customer support queries and that in a month it already started handling 2third of their conversations so over 2.3 million conversations that it was replacing the work of 700 full-time agents that it had taken down the time required to to uh deal with the query from over 11 minutes to 2 minutes and that they're seeing an increased profit of $40 million in this year just because of that one roll out of one kind of AI system so yeah and the crazy thing is it's happening in knowledge work as well so traditionally if you think about oh yeah well it's not going to replace a doctor you can't automate the work of a doctor or a banker or a lawyer absolutely because it turns out that those knowledge Industries are ripe for disruption because there is so much knowledge and so much data much of it in unstructured human language uh or or or in text which now suddenly you can ask a machine system to dig through to analyze to find what you need at the point of decision um I'll give you a very concrete example in medical Diagnostics so Google actually released a AI model which was trained to work with doctors trained on medical data to help doctors diagnose diseases and what the early study found is that okay the most ineffective were in terms of getting the Diagnostics rights were the clinicians who worked unassisted with AI so the clinicians who worked unassisted with AI in the middle was the clinicians who worked with AI so they were better than the clinicians working without AI but the best in terms of the Diagnostics was AI alone geez we're just holding it back because it makes sense right if you train it in the right way and you structure it with the right data which is all the PO and this is why data is so important and data sharing becomes so important in use cases like this which is all the potential known information about that disease in the world from all of the doctors from all of the hospitals from all the medical research and all of a sudden you have a system that can churn through it immediately right at the point of decision can analyze a scan be like based on all this information knowledge I have this is the output that that's far more than a single human brain could do right so wow yeah so the of course there's going to be disruption but on the other hand I don't think that means that everything's going to be automated by AI because humans will still want to have purpose humans will still do activities um and a very basic analogy would be like the black cab driver you know they had to learn the knowledge they had to they had they were the Benchmark they were the highest they were the only kind of cab drivers in London and surely you're thinking okay when GPS comes about and anybody can just you've studied this for 10 years so you know off by heart every street in London I can just give you the name of a place and you know where to go and now I can do this on Google Maps it didn't mean the black cab driver died but it just meant that the industry changed I know but is then I wonder if there's a limit I wonder if there's just a point where it's like okay because I know people then say Oh the Industrial Revolution that changed things but we worked with the machines but is there a limit where we know you know why will we have black cab drivers or Ubers when we're at a point where cars or whatever it is is just getting us anywhere sure why will we have doctors at all if we have these kind of machines that are a million times better why will we have YouTubers and I know that people would say oh we want the authenticity of a real YouTuber but I'm imagining them flooding I'm I'm panicking because most people say about their own career like oh not my career I think mine's going to go fast because people won't know that they're not real or you'll just have more you'll just have more so let's take you as an example there are going to be more YouTubers so maybe the areas of interest that people follow will become more Niche so you have a following it's it's about brand loyalty right who are your followers why do they like you um they'll continue to like you um and now let's just say now you can have a new YouTube channel on somebody who's completely synthesized it's AI generated who can start exploring a new Niche that nobody else thought of thinking about and Their audience will be 1,000 people 500 people but that content can be produced for them so I imagine that kind of information monopolies or influence mon monopolies will be divided down into different niches and I imagine also for someone like you who will continue to do this there's new avenues right so you can have your own avatar who you might synthesize for parts of your content or if you're like I don't want to do that I just want to create authentic content which is filmed in studio with me and the guest yeah then that becomes your benchmark but you know where you will be using AI if you aren't already on the production side sure right because it isn't only on the content side it's also in the production side so all of a sudden someone like you or I I mean I run um a global business we are a small team we use loads of AI tools and we have so much impact in a way that just perfect forbear to that was the internet you know I couldn't do what I do and have as much impact as I have without the internet and the same for you you have a digital business you're literally a digital content creator and the same will be true for AI Tools in your business but you got to have the right mindset and if you don't adult you know your competitors will so it's more that the model is mixed autonomy and rather than AI taking a job you know somebody using AI will take your job I think there's going to I'm going to have to tell the audience like I'm real by the way I'll be I'm a real one I'm not AI or whatever and all the other ones that are AI will be saying they're real as well and that they're filming in a real studio and the thing is they can produce an infinite number of those AI ones who look a bit like me but better looking and with a slightly nicer voice and who much smarter and faster and who has incredible guests that no one knows are not AI either like we could be AI now they wouldn't know I know this is the big reveal isn't it the big reveal uh yeah I mean I wrote a book about it that that's one of the most interesting um I was the first to write a book on like the societal ramifications on AI generated content and the really interesting story for the Genesis of this book is um so it's called Deep fakes and it came out a few years ago and I'm sure these topics will continue need to get more important and relevant but one of the main things I talked about in my book was the value of in information integrity and how we already so this whole digital ecosystem which we've created is fairly nent it's new but we're all already a part of it and you don't really have a choice not to be a part of it you know whether you're an individual whether you're organization or company you're a nation state you have to exist in the Digital World um and over the past decade we've already seen how corrupt that information ecosystem has become so one of the big risks when it comes to AI generated content at scale and also AI generated data is just like because there's going to be so much of everything it's like how do you know what's authentic what's not how do you even get through the floods of information all the stuff that's been going to be coming at you and uh increasingly there's a view that the only way to do that is to an approach known as like content authenticity and Provence which is basically you cryptographically seal content at the point of creation so you have something in the DNA of the content that shows that it's not synthetically generated and then you have a you have an open standard on the internet so you can it's more sophisticated than a watermark but think about it as a watermark that can't be edited Sur surely can be faked well it's very hard to do if it's cryptographically yeah of course you can break cryptography and this is where you're like oh well if you got into Quantum when you see like Quantum Computing that's obviously what could happen but then if you know then all cyber systems and all banks and but the premise is that it's still really hard to do and then you create like an open standard for the internet where you can see the credentials of that content whether it's synthetic or authentic because it's not that all AI made content is going to be bad right it's just more transparent that Watermark can save my career like I didn't know about that that having that Watermark being able to because you're right that that's becomes in in a world of infinite deep fake YouTubers and and for people I reckon a lot of people would be listening on the audio podcast or watching this and they won't have seen what obviously you've seen a lot of and I've seen now like entire news channels full of news reporters none of whom are real yeah and like going off into different countries walking around well you know the answer to that Jonathan is and it's like these are not real people they look 100% real and a YouTuber could easily I mean there are YouTubers already that are fake synthetic ones but if there is some sort of Watermark saying yeah but these are the ogs these are theal that authenticity is something that at least gives you something yeah and it's not even that you can have entirely synthetic personas but you can also have lots of content that's synthetic of a real Persona right so like we've seen how in China already for instance real influencers are using synthetic versions of themselves in e-commerce where they're just like selling products 247 and so does that mean cuzz part of I initially went to like there were a lot of YouTubers who don't use their real identity because they talk about difficult things or whatever and it's just like a black screen or a fake thing does that mean just like a it's giving the a face to the The Voice they're doing or does it mean that they talk a lot and then the computer goes okay that's who you are I'm going to make an AI version of you you can do both okay so you can either have like your own synthetic Avatar let's say you don't want to reveal your face or your voice so you can have a synthetic Avatar a synthetic Voice or or you could train and I've been doing this by the way I've been trying it out um because I've been traveling so much and I haven't been able to write a lot and I uh really feel bad because I want I want to share my thoughts with my community and uh but just to sit down every week and spend like four or five hours writing a newsletter I just haven't had the time recently so I've trained a chat GPT agent on my writing right so it kind of and I was working with it to be like let's give my style a name let's let's come up with a character what's My Style and we kind of work together a bit and I was like okay not too bad and now I'm going to give you these three things that I want to talk about this week and this is what I think is important draft a newsletter for me and it was like not there yet it wasn't good enough right but eventually it will be where you can have something in your style based on your entire Corpus of work that it understands what you sound like what you look like what's important to you and you could give it one or two bullets and I want to talk about this this and this and it can do that for you and there's no reason why you can already do that with text because that's the most basic kind of digital output right and we've already kind of mastered text with large language models but when you think about these foundational models the way that they're going is into multimodal and all that really means is that they can work with all kinds of digital information so images audio video text so ultimately you should be able to say look you know what my studio looks like you know what like so let's say you want to do a conversation with I don't know Steve Jobs and you got permission from his estate because that would be the critical thing right yes there is a point in the not too distant future where you should be able to do that just using Ai and the whole video would be a piece of basically software generation just me and that would be a new episode I put out me interviewing Steve the late Steve Jobs and it wouldn't be me either or could be well it could so it's you based on you're working with an AI system to train it on your past work and then you could obviously edit it and tweak it and then let's say you got permission from Steve Jobs estate because you wouldn't I mean you could do it without permission but you know this is going to get very litigious and they would just like you see this happening now in Hollywood right there's been so much fear in Hollywood yeah around the entertainment industry is just at the front edge of this where you you see both sides where you're like this is really dangerous it's going to take our jobs it's going to completely disrupt the industry to people also sensing the opportunity so any actor for instance should be thinking about preserving their legacy right so if you get a request from somebody to even posum asly do some content with that or Entertainer you would almost have a training data set of what they might have said which would be based on all of their output you know based on their actual Biometrics because it's just an extension of their IP yeah so when you start thinking about content or content experience you start extrapolating that even into video games or expanding IP for something like Harry Potter where you want to go deeper into the story you want to explore a character or more you want to uh do a different kind of plot ending you know all of that becomes really possible for from the perspective of gaming and content experiences so it's super exciting potentially as well right as a consumer of content and wanting to have interesting experiences for already famous actors and YouTubers and things it might be more difficult in the future in a 100 years for someone to say hey I want to be a big Hollywood actor and it's like well you know Ryan Gosling's IP is taking all the jobs at the moment but it's it's again this notion coming back to this notion that once you automate stuff and you have synthetic actors because there will be some completely synthetic characters right and we already started seeing that before generative AI was advancing at its current capabilities when you had like CGI influencers like Lil Michaela is a really famous one I think rings a bell so it's like um CGI influencer her name's little Michaela she was like created I think by an LA creative company and she was like she doesn't look very human she looks like CGI um and she was getting the huge ad campaign so she was like in Calvin Klein I think it was with like one of the Hadid sisters so that was already kind of starting to happen but now of course it becomes easier because to have created something like Lil Michaela without the capabilities of AI it would have been extremely expensive to do right you still needed like a whole studio of people who knew how to do special effects or V effects but that is no longer going to that's not going to be the case going forward because video generation capabilities are simply going to be software and I know this because I'm an adviser also at a company called Synthesia they're actually a British AI company they do digital they do virtual avatars right um it's a they work with businesses so it's not for consumers it's like a busino business platform for things like training videos safety videos and they basically have um synthetic actors so real actors who've licensed their likeness on their platform so they get paid for any video that's made with their likeness and these videos are used by over 70% of the Fortune 500 companies to do things that would be really expensive to do before in the past like make a whole series of safety videos or training videos and you can just you type the script in on the platform it generates the video you can edit it generate it again and with a click of a button you can also change it into like 60 different languages so cool so it's the idea the the thing here that's crazy is the idea of video as code right and uh if you can do special effects and videos code then like you could also say there's no limit on your creativity because all the barriers to entry about it being super expensive not having a studio not having that should no longer be a barrier anymore how many years until we get to that point point where where the average user at home can just make a Hollywood movie can just make a YouTuber or whatever from code because of like Rising computation the fact that if you look at any general history the the history of general Technologies the trend is always endless proliferation right so that's true for computing it was true for fire it's true of the Agricultural Revolution it's same is going to be true with ai's capabilities I would say maybe 3 to 5 years before you could do like a Hollywood featurelength film on a laptop oh my God but again if if anyone can do that doesn't mean it's art right doesn't mean it's worthwhile doesn't mean it's going to win an Oscar sure only somebody who has true Artistry and Mastery of that discipline will be able to create something using AI software that's truly noteworthy I'll I'll just type in I'll just type like make an Oscar film please make make the best film and it just will well this I want to this is the really fascinating thing about it so um before we get to AGI super intelligence I think we're going to get to instead of AGI we'll get to ACI and this is something that was coined by Mustafa sullon who's the co-founder of Google Deep Mind no longer but co-founded Google Deep Mind and because of the way these foundational models are built they're built with like a layer of NLP which is basically natural human language processing okay which means for our intents and purposes that they're built with a function that we can talk to them in human language to instruct right and this you already see with that was one of the reasons why chaty PT was such a hit because you're like do this and I could write me a script or rap write a WAP in the style of Shakespeare that my three-year-old will find funny whatever it was um but that very basic or that foundational capability will continue to be true in all of these systems as they become more and more sophisticated and as they become more sophisticated we'll be able to ask them to do a lot more than write a rap or do a script I mean Mustafa sullan says that you know within this decade you'll be able to instruct AI to do something like make a million dollars say here's a $100,000 investment I want you to set up an Amazon store negotiate the contracts uh build a campaign to like Market the product uh fix the Drop Shipping stuff and grow that investment to a million dollars in nine months and of course you would need some kind of human elements within it but that the majority of the work could be done by highly capable machine systems see this is scary I mean for every bit here that's amazing cuz that is amazing uh it's a worry because it's okay the next I and I know you must have thought about all of these things a million times but you know who has access to that how does everybody can everyone make a million then what does a million mean what does that mean for inflation has everybody in the world got an amazing company now from chat GPT and all of that I think you hit the coin on the head when you said who has access to that right and this is getting to the very heart of one of the most important philosophical questions around AI development which is about open source close Source um and on one hand and I'm a big proponent actually of the open source Community because if you have technology that powerful or intelligence in machines that's so capable you cannot just be in the hands of a few no and also for and I know like the AI safety debate gets like a bad rap sometimes and you know the starting even going back to what we kind of first started with but the way that you really like try to align and make it safe is by having a community Through The Open Source community that can actually test the systems right if you can't test you don't know what the weights are and you have a safety team working behind you know in a Walled Garden very likely that things might go wrong now on the flip side people are like yeah well it's so powerful this cannot be open source because you could just ask it to figure out how to make like you know biological weapon yeah kill everyone kill my opponents yeah yes and anyone can do that now right this is no longer the domain of a nation state right like anybody could do that so I think that again if you look at the history of general Technologies there is no regardless of where you sit philosophically on the open V closed debate I think the reality is it it it won't be closed the cap abilities are going to be democratized but again there is going to be this huge question of accessibility in terms of who actually owns the core infrastructure those entities are going to have the most economic value driven to them there's a reason why all of big Tech have pivoted so hard on AI over the past 18 to 12 months because if you own the house or the infrastructure the software or the hardware that everybody needs in order to run their you know intelligence and various applications oh my God then you are going to be extremely powerful right yeah and I I think this plays out geopolitically because where we've seen the most uh not necessarily research Innovation because there a lot of innovation in European universities and European Labs as well but of course like the US is just Head and Shoulders above both with regards to some of the startups both regards to I mean just investment the amount of funding that goes into these incredible uh Frontier companies and then of course because of big Tech right and and the big tech companies are globally influential you have China which also has some penetration if you think about some of their products and services Chinese big tech companies has some penetration in like the Middle East and Africa but it's its own thing you know it has its own kind of unique ecosystem therefore it's the US companies that are going to be really really Global and I think this is again going to shift the balance of power in the world because I think the US is poised to do economically really really well from everything that's unfolding with AI because of how well its private companies are placed and I think that Europe is not in part because we don't invest and we don't have any hypers ERS and we don't look we don't invest enough and there's no kind of strategic Thinking by the government and so my view is that you're going to see there much has been made over the past few decades you know since like the end of history and you know the US isolationism or thinking of the rise of China that somehow the US is declining and putting aside the political issues which we see across you know the entire Western World economically I think it's poised to do very very well but the disparity in wealth you know how are we going to how is that economic prowess that sheer wealth how is that going to be distributed in society and that's that's the more interesting question for me both in countries like the us but also globally if you have certain parts of the world that are way behind in being able to access this infrastructure because it's not being built in Native languages for their experiences or maybe because they there's even parts of the world where you don't even have proper um internet infrastructure yet or Computing or access to the hardware and the software that's going to power all of this oh my God it's a scary world and speaking of scary stuff okay let's move on to I mean have have you given much thoughts to have you written about porn oh yes I have I have deep fake can that get rid of the I mean I guess it's a parallel uh discussion to what we were talking about with YouTubers uh except take their clothes off that's that's the major difference uh and is there that desire for people with with pornography for uh deep fake people or will they want that authenticity and you know so the entire story starts with porn oh it always does doesn't it in it always does porn is always the pioneering use case and that's where my story with AI starts as well oh yes uh I'll explain um I used to work so I used to work in politics and I used to like work on various different political campaigns I've worked with various different political leaders but the kind of most interesting trend of my career uh was how technology is becoming this macro force and is disrupting the world both at the level of very high level kind of States craft but also in how quickly it's changing the collective Human Experience and I experienced this personally as well because I'm half nepes my mother grew up in you know a tiny Village in the Himalayas in the 1960s and within a generation I mean when she grew up Nepal was completely closed off from the rest of the world uh they lived in the same way they had lived for hundreds of years as kind of subsistence farmers and in the space of one generation you know the country has completely changed because of access to information the internet modern technology Computing so my experience has been so different from that of my mother so that again touches on this point is talking about exponential technology and that curve is going to be even steeper In Our Lifetime for for those of us who have kids um so I was just interested in exponential technology and how quickly the world was changing and in 2017 um I'd been working on things relating to election Integrity declining trust and democracies declining trust in the El electoral process um and I saw this stuff happening online at the end of 2017 on Reddit and what was it it was deep fake pornography because when ai's capability started emerging over the past five six years including now what's become known as generative AI so the ability to actually create content and it started leaking out of the research Community right these advances what was the first thing that early enthusiasts did with those capab abilities yeah naked people yeah they made porn right so just like with the Genesis of the internet itself and I immediately recognized that this is seismic right first of all these capabilities are far beyond anything we've seen before second of all oh my God you can clone somebody's Biometrics because again coming back to my point about data it's all about the training data and if the training data is your Biometrics then I can synthesize you so I immediately understood that it's more more than just some kind of toy women's issue this is a civil liberties issue this is an information Integrity issue this is about content creation synthetic content creation at scale but as I started going into it I mean I was thinking primarily about it from a miss and disinformation information Integrity perspective because that's what I had been working on until that point um I realized there's way more than that because that was just the very first manifestation of something much more profound and that is the seismic step change the coming intelligence Revolution where machines that are trained on so much data can do anything we assumed was unique to human creativity or intelligent whether that's content creation whether that's video as a piece of software whether that's being able to diagnose diseases so at that point I started working not with policy makers and politicians because I was also pretty frustrated at how bureaucratic and slow moving that is but working with technologist right who are at The Cutting Edge of this but it starts with porn and oh my God are there going to be some that whole area is just so fascinating you know the area of research where love sex and relationships because um it's also pretty disturbing I mean one of De fake porn is super malicious for instance because it's only targeted against women almost universally against women as in brutal no you don't see defect por of men oh I see okay sorry you mean as like an attack where they makes a real person I guess I was thinking like an entirely so it was like non-consensual so in the first use case is non-consensual porn universally made of women celebrities and things okay and not only celebrities you think oh celebrities and there were lots of celebrities but normal women as well because and there's a really my my friend Henry AER who's like a a fellow AI researcher deep fake researcher did a really pioneering study few years ago now where he was like looking on um uh was it basic one of the social networks I forget which one but found like huge sways of deep fake porn of minors of children okay yeah that was going to be my next question actually because that's that's probably the most controversial question you can you can ask really is and and people ask this is that better to have like not real children completely faked ones for those people with that you know or is that a uh what's the word a gateway drug you know some people say it's a substitution for those people uh and it's a disgusting thought and some say oh no that would be a gateway drug they would watch those deep fakes of fake children and that would move them onto real children yeah I mean I would always argue that you know even synthetic porn of miners is it's not acceptable yeah I'm putting my foot down here and be like nope I don't think that's a controversial stance like nope I I don't think it's better it's uh because also first of all the training data has to come from somewhere yeah uh there's also questions of getting again to this information Integrity right what's synthetic what's authentic you know uh oh so there's no way to act they would always be it would be based on off of real the only way you can tell yeah so so right now and this is this this debate is already playing out with adult performers where you have seen porn stars created who are entirely synthetic right and and there is a market for that I'm sure because you'll be able to talking about immersive content experiences in in in in gaming or with extens you know immersive IP building and I gave the example of Harry Potter there's no no no uh reason you wouldn't do that in porn and I'm sure that's already kind of happening right where you have like these immersive immersive synthetic experiences either with completely synthetic performers or with a synthetic version of a performer you like but there's already been questions about the training data right so to do that you need a data set which includes lots and lots of videos of authentic performers now was that data set taken with their consent or not you know is there a copyright issue and you that copyright issue and the training data issue is already playing out in the courts not in the context of porn specifically but just in the broader context of these foundational models there's a couple of cases class action lawsuits winding their way through the US courts where people are saying you know my data was taken without my consent but I imagine if you're a very commercially minded adult performer this is the same analogy I was making to what you could do as a a YouTube presence is you would probably synthes your persona you know and you wouldn't even need to perform right so your fans can have experiences with you content experiences with you that are completely personalized to what they want to experience um and uh you would license that man this is a crazy world but obviously when it comes to miners that you know Absolut absolutely not because thinking about the training data also thinking about how this can be abused in the sense of uh the arguments around well this is synthetic this is authentic like no absolutely not where do you want to send people um your work and things like that oh you can follow me on my website on on n.org or connect follow me on LinkedIn um Instagram n.org who is a heretic you admire oh good question I'm drawing a complete blank here should I go on chat GPT and find like no I can't do that it's literally like tell me what to say AI yeah tell me what to say we're doing a podcast about AI yeah it's called Heretics who should be the heretic um it came up with Alan Turing oh yeah that is really good actually yeah because I mean he is really the brilliant mind that was at the very Inception of everything that is now unfolding with artificial intelligence uh one I often talk about in my talks the famous touring test you know about the touring test yes which is now it just null and void now yes so there's been a lot of debate about it so basically for those who don't know the touring test is like a test that Allen touring proposed where he said look we're going to know that machines can think or that they're increasingly along the lines of what humans are capable of at the point when um we can speak to a machine and we're convinced it's human like conversationally and that's obviously that premise that principle is what kicked off the whole hype around AI after Chachi PT came out right it was like a the touring test is beat you kicked off this whole debate about machine intelligence now I obviously don't think machines can think like humans or that they're sentient but they are absolutely increas ing in their capabilities so we probably need a new teing test for our time did he underestimate how he did he thought we'd get there by the year 2000 oh so he significantly underestimated it he underestimated what how good they he thought it' happen quicker so like like in the case of a lot of kind of world changing Technologies you know we even with like nuclear fusion for instance people oh it's only we're we're there we're almost there you know we're only a year away and it hasn't happened yeah and the same with what Alan tring predicted on when we'd be able to have like a con proper conversation with AI he thought by the year 2000 and it only happened in the year 2022 but now that it's happened we're on this exponential trajectory and the new touring test which is something again that Mustafa suan um suggest is not what it can say but how it can deliver real value in the real world so the new touring test will be when can AI make make us a million dollars wow fascinating go on and obviously I mean he is such a legend just as a person you know cracking the Enigma code um and lived ultimately a very sad sad life you know being cite poisoning I think yeah because of his chemical castration very very sad um the book The the movie The imitation game and the reason why it's called the imitation again is again this play on the touring test is interesting definitely worth watching for more interesting stuff go to nah shik's amazing stuff we put all the things that she was just talking about we put them in the description please go and support her she's given up her time and been brilliant I've learned so much stuff keep watching this channel hit the like button and well that's it just keep doing now

Channel: andrew gold | heretics.

Views: 54,332

Keywords: andrew gold, podcast, heretics, on the edge, nina schick, ai, woke ai, robots, future, authoritarian, 1984, george orwell

Id: f0X9TPiOFJY

Length: 65min 13sec (3913 seconds)

Published: Mon Apr 22 2024