 Kia ora kato, no mai haere mai. Greetings to you all and welcome to this EHF Live session with fellow salmon, Rob Reed, John Lee as panel members, and Chris Kanya as moderator. Today's topic is AI. What impact is generative AI likely to have on New Zealanders? We're not heavily digitised so we may have missed the boat, but was that boat the Titanic? This is the second of a series on regenerative AI. Chris and I have been having quite a few roundtables with fellows on what could the fellowship collectively create as thought leaders and users and decided to broaden this out to include the New Zealand ecosystem. During the series, the fellows want to cover risk and reward to New Zealand, security and regulations, well-being, right through to AI being used as a tool positively in business, like today's session. We'll be having about a 45-minute conversation with the panel and then moving into Q&A and discussion with you all during the next 90-minute session. Now about your amazing moderator, Chris. He's from Cohort 4 here and now lives in New Zealand and is a master in mentorship, coaching, agile training and mindful instruction. He does a lot behind the scenes of some of New Zealand's biggest organisations that you're probably not aware of and I'm excited to have him moderating this AI series over the coming months. Just a bit of housekeeping. This session is being recorded and it will be on the website afterwards. So just stay muted until Q&A where you can put your questions in the chat box and Chris will put those out there. And some of you may have to leave and that is okay too. Over to you, Chris. Hi, thank you, Michelle. So it's a real delight to welcome our panellists. I'd like to start with Sam Ng who has led product and service teams at UNDP's Chief Digital Office to support countries and their national agendas with a focus on inclusivity, human rights, responsible tech use, emerging tech, public goods and foresight. Rob Reid is a longtime tech entrepreneur, Rhapsody, current tech investor, science fiction author and a science podcaster. And John Lee was a product designer at Facebook, Facebook Reality Labs, Microsoft and Axon. And it's a real pleasure to have these thought leaders in the room and we've got some great questions lined up for you. So very much looking forward to sharing with you. So first, I've been in the thick of generative AI but I'm really curious to hear from you. How big of a deal is generative AI really? Is this a pets.com moment, an iPhone moment? I'd say iPhone badly understates it. From my standpoint, I would say there have been probably in the history of humanity three user interfaces that utterly, fundamentally completely reshifted the face, well, two, and I think this might be the third, the face of society and business. And to me, the first one was the duo of the telephone and the phone directory wasn't around then. But if we can certainly imagine how the rise of the telephone utterly transformed social lives, the presence of distant people in one's life, all elements of commerce, price discovery, markets became national, local commerce, just all kinds of things would have completely transformed. And if you held current, there were a lot of other transformations that happened at the same time, electrification and other things. But I would submit that if you had held everything else, the other technological innovations that defined that early 20th century kept them constant and you didn't have the telephone. There's no way society would have transformed to the degree that it did between, let's say, the 1910s and the 1950s. The next one without any question was the rise of the web browser and the search engine. And again, we need hardly, we can't even, it's impossible to overstate how transformative that was. But these are things that basically transformed the user interface through which we interact with almost all of the inputs of autonomy and information. And to a significant degree, with the first two and probably to a lesser degree with genera of AI, interpersonal relationships as well, just completely rearchitected. And the economy contorted and developed and just changed in completely unpredictable ways along with that. I believe that genera of AI has the potential to be that big. And the reason is really in that first word, genera of. It's transitioning from a point where most queries, most requests for information and even for media and ultimately I believe for entertainment, go from being a request for something that exists to a request for something that does not yet exist. And that is a really radical transformation. I mean, you think about it, you know, querying Google for a bit of information, querying Netflix, Flix for a movie, querying Spotify for a song. These are all basically asking for something from the great library that exists in the world that has largely now been digitized over the past few decades. And that's something that exists that's going to satisfy my need right here right now. And when that which we desire increasingly can be better generated than pulled from the archive, I think everything transforms. Now I'm not going to, I don't have total conviction on this, but it's high and it's rising and it's been rising at a very rapid for the last five months. So I'm going to say iPhone might even understate this. And don't ask me about this whole Apple VR thing. They will get John in that one later. If I can, if I add to that, I totally agree. Rob, I'm a bit of a designer from way back as well. And one of the things that really got me about this recently too is that sort of step you talk about the kind of conversational interface. The GUI made a huge difference as well as web browser, the graphical user interface, just kind of made a lot of communication that much more accessible. And one of the super exciting things I think about what we see with generative AI is the conversations that we can have using language. And I think language is such a big deal. If you pause and really think about language is really in some ways the sort of medium for civilisation. We can't build the kind of modern world without being able to communicate in the way that we do across these hundreds, thousands of different kind of languages. And that's what I think is super exciting about it. And also, I'm sure we'll get to this, quite frightening about it because that transformation is happening at that really core level of civilisation in terms of how we communicate. All the words that I'm making now and the sounds that I'm making that are being transmitted to each of you in the way in which it registers as a bunch of concepts. And as AI and as computers master that ability to communicate through language, it just opens up all these kind of different doors. Exciting, possibly also a little bit scary. But I think the one thing that's great about, regardless of what we think generative AI is a big deal or not, one thing I'm really excited about having these conversations mean that we get to choose what happens next. Because I think the thing that's concerned me a bit about these conversations has been a certain form of determinism that the kind of narratives that we've heard so far have kind of led us to believe there are these fixed futures that we don't get much of a say in. So I love to believe that we have more agency than we're sort of led to believe. Yes, totally. iPhone, that's just really going to be a footnote. I think if we look back even 10 years from now, these are going to be quite a significant inflection point. Yeah, I just want to build off of that in terms of transformation. Definitely think it's an iPhone moment, if not way more. Just, I think what blew me away was it took Twitter five years to reach 100 million users. And Instagram from the Facebook side of the world, two and a half years and it took chat GBT 64 days to reach 100 million users just for that scale of speed. And it just blows my mind every couple of days. You just learned about some new kind of gen AI tool. And it's just fascinating the pace of development. And I think what's also interesting in terms of the context was that with COVID and lockdown, Microsoft said that two years worth of transformation happened within two months during that lockdown time. And so you have that leading up to then all this gen AI stuff that's coming out now, just a really, really fascinating time to kind of see what's developing. Thank you. Thank you. And so Sam, you spoke to looking beyond fixed futures. So that's a great segue into our next question, which is what positive futures can you imagine with the integration of generative AI into everyday life? And we'll start with a global context to kind of set the stage. So what positive futures can you imagine with the integration of generative AI into everyday life in a global context? I'll pick up on that because I have a bit of a different perspective, perhaps the most in that in my day job, I tried to think and advocate a lot for the billions of people who are not even connected currently to the internet for whom any struggle of electricity and shelter and power, some of the basic kind of needs that we take for granted. Having said that, I think those prospects, that potential for some of this technology to really level the playing field, but also grant incredible access to people in the ways that I think were incredibly difficult to do before because of some of these more traditional barriers. And maybe just picking up on my point about language, for you to access the global economy, you tend to have to have a basic level of literacy and also just even ability to master a language, to speak a global language like English to better access scientific knowledge, to access different markets. And as the optimist in me, I think as those sort of barriers get removed and as we are able to provide greater access to learning and healthcare and all those kind of basics without the same kind of barriers, I think it's tremendously exciting to be able to craft possibilities for what's often termed as the global self. Clearly, there are scientific breakthroughs. There's a lot of positive opportunities in health and in learning that can happen as well. But beyond that, I think the sort of productivity increases which are more near term, how that can create more space for us as humans to even move into higher value work or just even for bud create better wellbeing and balance for us as humans as we come back to what we're good at and what we're intended to do as opposed to perhaps a lot of the busy work, which I'm really looking forward to a lot of that busy work being taken away. The more tedious and boring stuff that we just don't need to do. Personally, I think the kind of learning that that opens up to all of us on the school but also billions of other people who don't get the chance to even be part of this global conversation is going to be one that we get shaped which is going to be great. I'd say one of the most sublime advances that I fully expect to see from this and that I am definitely already experiencing is going to be really just tremendous potential in education, particularly self-education. So it is absolutely true that a very high percentage of people, not all, but in a very high percentage of learning circumstances, not all, but high in both cases, learn with incredible, not a very scalable enterprise for obvious reasons. But depending on the topic and depending on a particular person's learning mode, building on what Sam said, language is the most immediate way to get things in and out of our brains other than the visual cortex. It's certainly the most immediate way to convey thoughts and communicate thoughts. And as we scale up from, we already have things that are text-based, but as we scale up from that, there's already tremendous voice synthesis technology that's out there and video is more expensive and more slow, but it is going to catch up. When we get to the point where chatGPT is really chatting to us with a natural voice and for some people in some applications when it is also a human-like face that's conveying information to us, the ability and the large language models get more and more exquisitely tuned to have conversation and to detect holes in knowledge and to answer questions and to just very, very interactively instruct with a memory of what this student has learned before and with an increasingly well-refined intuition for what they may not know based on what they've said. I just think the ability for almost anybody to get educated, to self-educate, it won't be self-educated being collaboration with a model, but nonetheless to upscale themselves and we'll go far, far beyond the extraordinary things that are already possible with things like Khan Academy in Wikipedia. And I believe Sal Khan's TED Talk is now online. I was like to see him give it live at TED conferences just in April, but I'm pretty sure his talk has already gone up and if it has, Google it because what they've done at Khan Academy with a tutoring bot is already unbelievable and far more unbelievable things than what they've said before. So I'm pretty sure that's what they've done with the pipeline. Myself, I have this sort of this deep science podcast that I'm lucky to, if I can do three or four episodes a year now. I used to, there's times when I've been able to do a couple of months, it varies, but I interview people from a really wide diversity of discipline sport and so I end up usually spending 30 to 40 hours preparing for each interview because it's like, oh, I got to go from neuroscience or joy of learning and it's always under the tutelage of my guests who's going to be like a world-class scientist in one of these areas. So I've gone from zero to advanced laymen and a lot of fields over the years and as an investor, I need to do that too. And so I happen to be right at this moment in the conference room of a company that basically does enzyme discovery and development using machine learning. Interesting stuff. I did not know a great deal about enzymes and getting ready to understand this company and what they do is, was a lot like preparing for one of my interviews and so it's like, here we go again, I've done this dozens of times, I know what to do. What's different now compared to a year ago when I would prepare for a scientific interview is GPT-4 is an incredible way for me to navigate a new and complex scientific space. All my old tools like Wikipedia and other kinds of queries and finding lectures on YouTube and so forth, they're still there and I still use them, but the ability to get to something that is really targeted to my level of knowledge and I can answer my questions by finding the Wikipedia article on a particular domain of enzymes, but I can find it much more rapidly by asking GPT-4 and say, give me a 750 word answer. So little micro articles coming out, that's already just like a blinding increase in my efficiency learning. It's a narrow application of education and I think it's going to go across the board and I think that humans are going to become smarter and wiser at a much faster scale than ever before as a result of this. So that's what I get real excited about. Oh, that in shorter hold times in customer service. So that's my sublime thing and my mundane thing, but they both are important. I think you've been optimistic about the customer service thing. I don't think it's that good. We're still going to be put on hold for different reasons. But it will be your AI agent that's on hold. And it has all the time in the world. That's right. I have a point about agents, but John, I might let you jump in first. I think that was a pretty solid discussion. You guys covered a lot of things I would have said. Sam, would you like to add anything? Yeah, I was just going to say this. People have probably seen all kinds of different reports in terms of what that means for supercharging productivity both in terms of quantity and qualifications on jobs, or the rest of it. But I think this idea of a digital agent, something that is a virtual assistant for everyone in the organisation, for yourself. I'm sure we've all probably played with this already. And that kind of prospect is already gaining quite a bit of steam. And recently I had a bit of a joke with a friend that wouldn't be great that instead of us having this kind of conversation which is filled with padding just because we have this kind of conversation with each other, not necessarily just getting these generative AI models to work on the outputs, but perhaps to get on the inputs from humans, but have these conversations. And sure enough, these startups out there are really starting to do this. So in the near future, perhaps what we kind of think of productivity now for better or for worse might be handled by autonomous agents that we get to design and fine tune to our interests. So fully review the role. We're all going to be enzyme experts before we know it, and that's a good or bad thing. One of the quips that came up with the last panel is everybody has an intern now. Not yet, but that's something there. Getting there. Sam, I just wanted to build off of that. I think having designed for privacy, I think what I'm most excited about is a personalised AI is much more on-device as opposed to on the cloud. And therefore, now we're talking about AI and there's a lot of scary aspects to it, but now if it's purely on-device and it's not necessarily connected to the cloud, then that level of AI is much more individualised and I can trust that a lot more. And so I'm super excited about the future of AI plus these closed networks and AI and perhaps me made just within my household. I think that's a frontier that's going to be really fascinating. On that note, I guess we might get into it later on too. What the role of open source might be in that regard, in terms of who actually has access behind the curtains to what actually happens and how that might be controlled. Thank you. So we've discussed some of the global benefits of generative AI. What are the risks that you possibly see? How can we address these risks while still promoting the possible benefits? I just think there's going to be an ability to perpetrate fraud on an industrial scale that doesn't currently exist. And so in one example that I've been thinking of and using a little bit lately is there is a form of scam that we're all familiar with and we've all been exposed to, which is the Nigerian Prince E-mails Hue and asked for your bank details in exchange for a small or potentially huge benefit. We've all gotten those e-mails because they are free, essentially free to send on a mass scale. And they probably worked very briefly a very long time ago and we all have our immune layers kind of booted up. Now the consequence of booting up those immune layers are that when we in our normal life encounter Nigerian Princes who are real Nigerian Princes who want to hand us money, we're probably going to shut them down but that's a pretty low cost that we're incurring as a result of that. So move from that to a very labour-intensive form of scam. There's something that's got a terrible name that comes from the term of fattening up a pig for slaughter. And in this form of scam, and this will be a generic example but there's variations around this but basically somebody will find a target in an environment where people meet people online. It might be a discord, it might be a Reddit it might be a dating app. But the target is targeted and the perpetrator basically plays a very very very long game where they befriend this person over an enormous number of interactions. At some point say oh my God you work so hard, you deserve more from life whatever it is. I've got this really hot crypto tip. Now don't bet much please don't bet much whatever you do but here's the tip and here's the exchange. The person's directed to a very legitimate seeming exchange they make this small bet and it pays off huge. They may even actually withdraw a little bit of money but now they're hooked, the hook is set this is sort of the quote unquote fattening phase and then over another span of you know many many days weeks even months that hook is set and manipulated at some point the person's betting more and more and more they're getting addicted to their parent but you know ephemeral winnings and at some point their fake friend and their fake prospects and their real life savings all vanish. There was one of these cases was documented I think it was in Forbes the perpetrator and the victim swapped 271,000 words of text and I can tell you as an author that is equivalent of an 800 page book no scammer will however greedy they are they can only do that so many times a year now an LLM I'm sure this is going to start happening at incredible scale now LLMs will be able to perpetrate that you know turning test passing bots to not a few dozen of carefully selected targets but extremely broadly and now add the richer media add the voice and then later this really worries me a great deal you know eventually pixel perfect video totally photo realistic in this environment whether people are criminal scammers or they're more benign non-criminal marketers or their people have a political agenda who are trying to you know persuade somebody to come over to a fringe point of view or even mainstream point of view I worry that the number of fake people out there who are photo realistically seeming real might outnumber us online and those first of all there's just the tremendous runaway potential for scams but now think of how our immune layers start rising up I'm not just merely suspicious of an idea in Prince offering me money which doesn't really happen very often in my daily daily life so that's not a terrible expense but these bots are going to become masters of micro expressions they're going to become masters of feigning empathy they're going to be masters of eliciting sympathy and pity and they're going to become masters of all these subtle cues that basically make us human and what happens when our immune layers start blotting that out and you know start saying okay that's you know when I see that I go I shut down and we start doing humans as well as bots and we translate that through the offline world and online world I really really worry about that runaway potential and yeah so that's that's dark sorry one coming back to I think my this is why I think language is so transformative right as you kind of sit here and listen to us make arguments trying to be eloquent about our points of view you could imagine that we stand no chance compared to a supercomputer in general rating these kind of persuasive arguments I think Rob is you've kind of pointed out not only do they have the time but they've got that kind of brute force ability to you know really manipulate based on language and you know for thousands of years politicians and poets and all kinds of people have been using storytelling and language to really shape society and actually one of the things so putting putting on kind of again the head of institutions democracies ultimately fundamentally are conversations in public there's lots of people talking to lots of other people about what to do collectively and so as AI and computers begin to master language it can really shape those conversations I mean social media has shown us how that's plausible and when that AI has been unleashed it may be quite challenging for us to better understand like you say it's not just text Rob it's fully multimodal how do we ensure that we have that ability to trust how we have these conversations in public because this is the way in which we address and solve many of the other challenges that we face as a society but having of course having said that I think these are some of the current narratives that we hear over and over again I think one of the things to come back to that question how do we manage those risks while potentially still benefiting people and planet one of the key things is going to be the stories we tell each other this you know a lot of the things here now provide responses of I think ignorance or despair but really optimism and maybe maybe not maybe I'm just listening to wrong podcasts but I think there are real opportunities for us to begin to create broader civic engagement around a positive forward looking plausible vision of AI could be that isn't determined by some of the few actors that are having that conversation now and that's going to be a lot of things I think there's going to have to be much better coordination cooperation general this is obviously not just a US problem or a U problem or a New Zealand problem this is well beyond any geographic boundaries one of the big things I feel even for myself is just how do I keep pace how do I have adequate AI literacy and let alone the people who you know easy to pick on education and that the poor educators and teachers are already so heavily overloaded how do I keep pace with the opportunities and get a basic level of AI literacy to be able to respond same of course with our policy makers and our lawmakers and our leadership and our many civil servants who have to try and make sense of this so I think there's got to be a real response a balanced response for helping people come up to speed and at the core of it and I guess I alluded a bit to this in terms of how we function as democracy as we get the opportunity to I think rethink how we organize work a lot of work historically is based on technological as well as in organizational constraints I mean it's how many people we can manage how many hours in a day we've got these are the languages we can speak so some of these sort of assumptions and constraints may benefit from a bit of a rethink as technology has become a bit more mainstream so choose to be more optimistic I think choose to tell better and different stories as a bit of a starting point which is why I think it's exciting to have these sessions You can't have the future you can't imagine John Yeah or in thinking about some of the risks of AI I think it from the creative perspective there's a certain joy in my upbringing as a designer of just how many sketches I've done just from childhood and leading up to now as things become easier and faster and more powerful with all these AI tools I worry about some of the convenience and the speed of convenience taken away from the joys joys and slash tediousness of learning something and the practice that it goes into something and I I don't know what that means for the future of some of these pursuits but I do worry how fast and easy it is to generate certain things that might take away from the time that can be invested and eventually becomes quite joyful in how that skill set blooms Thank you I want to see if I can narrow it down with you a little bit. We've spoken about the global context and I'd like to get a little more local with you What do you see as the future possibilities that excite or frighten you specifically within the New Zealand context I wasn't really worried about our elections in September but I listened to the RNZ podcast recently like Bob Reid was it perhaps that actually kind of just caused me to stop and think a little bit because obviously for some of us in New Zealand we've heard about the kind of national political ad words, kind of fiasco where they use mid-journey to generate a bunch of attack ads and I would like to think New Zealand is not the sort of place where there is an issue but we also forget that just a year ago and this is what was highlighted in that podcast we had those parliamentary kind of protests so there is a kind of growing kind of undercurrent of people who may and bad actors who may use some of these technologies to their advantage and kind of so descent so in a more immediate issue I think that's a little bit higher I thought I think that a lot of that focus is of course on the US elections next year but in general I think this idea of how we have discourse in the public forum in the town hall in a way that can be trusted and it is respectful and how many of these technologies really accelerate that perhaps in the wrong way that's something that is definitely on my mind related to that I think it's just the sheer speed in which everything runs so this isn't New Zealand specific but I think part of what we underestimate is just how quickly things are developing and while some of these use cases at the moment and the applications are quite laughable we can expect things to grow only quite exponentially from here on in and any other thing I think that's both frightening but also quite exciting just New Zealand is a bi-cultural society how if we were to do this right and I think there are many possibilities for us to kind of think about how we might better encode the values of Maori perspectives and world views into large language models into algorithms into how we engage in public conversations and better understand each other I think there's a fantastic opportunity for us to globally be a voice and we can be recognised for how to do this well and that is of course something that I am putting on my more global hat I see as an opportunity everywhere but it's definitely one that is going to take a bit of effort for us to make happen I think Tain just what Sam said to the sort of the pig butchering thing not that I really like saying the words pig butchering but is generative models accelerate smaller targets become profitable and so currently the people get targeted with these kinds of scams or people have a lot of money to lose when they can be automated and talk about the pig butchering scam now they'll go after people with $5,000 in life savings you can't do that profitably now now transition over to sort of election mayhem that kind of thing I don't know how much effort went into QAnon perpetrating that but that was a pretty big fraud a pretty fat target the United States and its political system and all the money that courses through its veins that made sense to somebody and all the synthetic media and other things that were done to hijack our election now we're turning up in New Zealand elections I think these kind of like larger scale frauds and hijackings of democratic processes that probably only would have been drawn to a very very very large target is going to be is much more likely to be drawn to a 5 million person democracy whereas in the recent past it might have been limited to a multi 100 million person democracy and so I think that's probably a risk that we can see Joe would you like to chime in? So I suspect we'll find some policy makers will be here or watching this video and I'm curious what advice might you give to policy makers what are some of the levers that might be available to them and I know Sam I believe you spoke a little bit to AI literacy even just keeping up with even just keeping track of the companies that are popping up in this space let alone being able to understand their products is like drinking from a fire hose so I'm curious what advice would you give and pursuing AI literacy particularly here in New Zealand? I think that's a pretty good place to start and I think a lot of the attention at the moment turns to the heads of organisations so you have these big talking heads which absolutely that conversation needs to happen and they need to be educated in terms of the headlines of what's actually going on but often I see the middle management layer if you like the actual operators themselves you have to actually take these and these tasks and convert it into something meaningful and they get all kinds of pressures kind of squeezed on specifically targeting them and prioritising a bit of effort to do that New Zealand's a small place and Wellington's a pretty small place and there are a handful of people who are willing and able perhaps to provide that level of education and informed hopefully in balanced way without a bunch of agendas attached to it so I think that really is the core starting point and just to tie back to the importance of being a biocultural society as well ensuring that we have all of society around for that conversation and for different perspectives to be integrated in understanding where the technology is going and how we can leverage that is going to be I think a part of our strength because the temptation is very much to copy there's an avarability to copy what other countries are doing just because the sheer size of who we are in New Zealand but there are lots of different ways in which we can reshape this conversation as well to be a good treaty partner and think about how this works in the digital world and the digital era as well I think that's going to be something that EHF is particularly strong at adding value to as well at a really practical level and how we might translate that I think there's possibilities for us to think about what sandboxes would look like from a policy perspective as well and maybe the opportunities and spaces to create these sort of certification or auditing systems where again we could think about what are the things that are important to us as a society what are the sort of values and the outputs that are important and perhaps not to use the hard sort of stick of regulation but just to be able to hold up a bit of a light against the different tools that are going to be used in New Zealand to be able to say this is how well this might stack up against the values that we hold dear as a society or a bunch of things that are non-negotiables for us in terms of the different tools like we could do this and we could just kind of shout out some of the things that we've observed that contradict or perhaps it's almost like a warning label that we get on foods for different tools that come into common use so there's really some great work I think some beginnings of some great work there's that algorithmic charter in New Zealand it's ever fairly lightweight still lacks that kind of detail but I do think that sort of enablement of public sector that middle management and operational layer is going to be pretty key to getting real stuff done not just positioning statements ethic guidelines and all that kind of stuff which is part of the course I wonder what the support system looks like for automation and the layoffs that will come with automation and what that means for transitioning that displaced labour and how I think in Japan they had the system where in a company if there's a massive layoff that happens that company will loan some of those employees to other companies while they try to figure out how to transition those workers and just really curious what could be a textualised New Zealand cultural support system for displaced labour that comes with automation and artificial intelligence and what kind of system could come in place I think from my perspective regulation is public policy is in some ways fraught because as digital tools these are going to be innately international and they're going to travel at light speed and the people who seem to be most concerned about AI risk are really talking about international agreements and international accords like the nuclear testament treaty which I think is actually good that we're thinking internationally about this already it's so rare that we think internationally about anything as a species and it's good that this is ascended to that level but what I think is intriguing about New Zealand is it is so much more have an agile government than we have where I live and it's almost like a really speedy sports car compared to a great big lumbering truck and great big lumbering trucks are good for certain applications but when you want agility and speed you know there's tremendous benefit to that as well here the giants of AI are present here and there's going to be all kinds of contortions that enter any kind of legislative or public policy process as a result of their presence and the depth of their pockets New Zealand is a much more pure place for I think discussion and ideation and probably legislation and so I think it'd be really interesting if New Zealand really took a lead on certain forms of regulation ones that don't presuppose that you can tell Google what to do because I don't even know if the US government can but policies that sort of make the public digital sphere safer and I think New Zealand could plausibly create really good common sense legislation that then it's much easier to evangelize something that's already a law in a country that's well regarded internationally that is excelled in a lot of areas like public health and so forth so just a simple example like an AI building label and sort of you know to pick up when Sam said just something that says you are talking to an AI now you are looking at AI generated content that just needs to be present we're not going to ban anything we're not going to prohibit the creation of anything but if this is a bot if you are a bot that's you know generating content and you know a chat sphere you're talking to somebody one-on-one you need to put up the flag people need to know when they're interacting with AI and it's kind of coming off of that off the top of my head but that's the kind of thing that New Zealand is a big enough market that people are not going to want to simply ignore it it's a very light regulatory burden to place on entities and people obviously fraudsters are not going to go along with that but that's a light regulatory burden you can geofence it and implement it quite easily and it could be you know a very very powerful simple powerful thing that could be very contagious to the communities of other countries I think there's a real opportunity to innovate here in a country that can move quick and think deeply and I think that's kind of cool Yeah maybe to add to that too I think there's possibilities for us to demonstrate what different visions and different futures could look like so as well as the kind of dominant models that exist and we might touch on this later on but open source is catching up fast and you know there are new models now that have been tuned to produce different outcomes so imagine if we did bring a bunch of technologists to your point Rob I mean we can potentially move fast enough and get government buy in to create a model that is cognizant of Maori values and perspectives and how might we take one of those existing models tune it in a way that might just offer a different window into how these can be used to demonstrate that as something that can observe kind of how that might be co-created with the people of New Zealand as an alternative as well so Rob have again set this kind of formal determinism that what Google and OpenAI have produced is kind of what we have to put up with I think we've got that kind of pioneering spirit in that and then move quickly to try different things as well and with like you say with the blessing of you know some of the countries that are very accessible and as perhaps this video might show right one of the things that I'm really curious about listening to this and Sam following on from your open source comment I've read an article recently that Google is quite terrified of open source AI models so they're less worried about open AI and more worried about the open source models and so I'm curious what additional regulatory challenges might that present for the US can't tell Google what to do how might we be able to regulate when in a formal group of people can build let alone malicious actors how does that change regulatory model if there's no one accountable. Yeah let alone the global kind of coordination cooperation kind of a challenges right so yeah they simultaneously terrified me and excited me at the same time and depending on the day I think open source how we view open source in the past isn't necessarily how we could consider it through offering us the same kind of benefits going forward but there are some models which are closing in very quickly on GPT 3 and 4 in terms of performance and I guess it is again it's up to us as to how we make good use of those I'm with Rob in that and again kind of sitting where I sit within the multilateral environment I struggle to imagine how we can keep pace it's a necessary guardrail we are going to have to do this governments are going to have to signal intent and cooperate with each other but just the rate of change is just too difficult so we're going to have to find different ways I think more carrots than sticks knowing we're going to need sticks to demonstrate I think better uses for this and to a certain extent I think we're going to need technical innovation to be able to counter some of the challenges that even those examples you gave Rob we're just going to find technical innovations we're going to have to support technical innovations as well as grow awareness and literacy to counter this and expect that regulation will do its best but we're going to need more than that as a defence I think from a defence level layer standpoint just sort of going narrowly to that the way most of us are defended against spam and viruses is not the law it's anti-spam you know programs and software but you know Gmail protects me from the vast majority of spam that comes at me because I have to use Gmail or it's antivirus software and so I think you know we do need to regulate well but we also hopefully we'll have economic incentives in place for defensive layers to be built by people who are good at building defensive layers I was talking to the CEO of a company called resemble.ai which is a really good one of the top way synthesis companies I interviewed him for my podcast and actually just posted the episode if folks are interested and he was talking about a digital watermarking technology that they have that will that's pretty interesting so what they can do is you can have a watermark a person's voice and you could have your voice watermarked whenever you'd put any spoken content online and if somebody ends up training on the vocal content that you have online the watermark is kind of invisible that generated voice that's generated from you know the surreptitiously watermark content that you have out there will be discernibly it will be shown this was trained from a model of existing vocal content now that doesn't help somebody like me he's got many dozens of hours of my voice already online that's not watermark but I think that that's a really interesting go forward technology and another thing that they're able to do is they can go let's say you're in a family and you are very concerned about somebody scamming you by calling you and mimicking the voice of a loved one or you're an organization and you're worried about any one of your 175 people who have a voice of the CEO trained on a model he believes and we talked through some technical reasons why he believes it's feasible and it sounds incredible to me families and organizations can do much deeper training on the voices of the people in that family in that organization so that something that is even a pretty good digital mimicry will have a hard time fooling the system that's been trained on the people you're trying to defend an example in a domain of voice but I do think that these defensive layers will be developed and I think they can become very powerful and we definitely need those to be out there as well Rob just kind of riffing off of that if I was given these constraints as a product designer I would try to imagine it living on my phone and as I'm talking if I had a personalized AI bot working for me you could imagine a way to design tools to kind of help filter through a lot of different phone calls that you get or messages that you get I can imagine like a web browser and as it's blocking a lot of pop-ups and spam and a lot of details you could take that parallel example and then build out ways to show some of these AI defense layer that Rob was speaking to and I think my product design hack goes 100 miles per hour because I think there's a lot of ways to contextualize that information have an AI personalized bot that lives with you and then present that information to you so that you can kind of filter through what's human, what's AI what's good for your everyday use case yeah no one at a kind of more ecosystem level I wonder how we can better support and create incentives for a lot of these innovations that like you say might just make it a more efficient market and so one thing that I've heard which I think is interesting is just how might we ensure that finance people that finance a lot of the stuff have some kind of I hate to use the term but there's an aspect ethical and safe AI especially for those who are investing in this that is part of the criteria that is part of also in fact the appeal of financiers so when it comes to technology and when it comes to people building a lot of this stuff there are some sort of incentives to do this right and not only from a philanthropic perspective but I think from the perspective of you know those that control the cash some kind of influence over what good looks like and there are lots of different tools that are out there and it's just some basic kind of hygiene being put into thinking about or whatever the unanticipated kind of consequences of what we're doing what are some ways in which we can mitigate this and what are some of the tools that we might want to begin supporting but yeah having said that I mean there are lots of other I'm sure initiatives to even as a fellowship there are going to be lots of super talented people here I mean John your brains are really kind of jumping to a bunch of different possibilities how might the fellowship even be able to create space because it's a limited talent pool too at the moment the people who really know what they are doing you kind of have this opportunity to make millions of dollars a year in salary somewhere with potentially huge upsides or you can give up your job and you know work for a pittance and kind of join the what's and can sometimes feel like the failing side of defending and regulating some of these technologies so how do we change that equation as well and I think that's something that we need to kind of consider as a way of just building more and more of these kind of innovations to effectively prevent spam from hitting out in boxes because yeah that's already happening isn't it and it seems like even today with the current maturity of these models we're having difficulty detecting the text I see a lot of commentary on social media people that have been confidently accused of there by the professors of having models write their papers for them only to be embarrassed when the same model or the same tool said the constitution was 95% GPT so there's interesting challenges there thinking about New Zealand again New Zealand has a history of punching above its weight I think we're kind of known for that and we spoke a little bit to the adaptability of our legislators how that we can move a little bit more quickly due to our scale but I'm curious what other ideas do you have for New Zealanders either tech entrepreneurs or legislators about how we can really punch above our weight using this technology what advice would you give I'll offer a practical really simple one I don't know a lot about it but what I have seen in the film and gaming industry excites me tremendously in terms of industry that punches above our weight and I think we might have had one of these I've streamed before in this area but you know the notion that this undermines the creative industries is perhaps a bit of a short-sighted one the possibilities for the kind of resilient and pragmatic innovation that New Zealand is well known in the sector to really redefine what that looks like in the future I think is amazing like double down on it so I think that possibility for us to really do some pretty creative things and reimagine what that looks like at a practical level is really good back to New Zealand being a bit of a leader in terms of soft power on the global stage demonstrating how we can do this better in even our start-up space would be fascinating and Rob, I think it'd be interesting to get your views on how you see this sort of in New Zealand and the US but you know as well as more kind of formal mechanisms would there be ability for entrepreneurs in New Zealand to think about how they can use it but also be accountable to someone maybe there's some kind of peer accountability in terms of how technologists and boards in which they're accountable to talk a bit about what they're doing and what they're looking to adopt so against the small community it's probably easy in fact there will be lots of different ecosystems that are really talking a lot about this both in terms of the entrepreneurs the actual talent themselves as well as the people that finance it so building kind of generally more positive tools as a tech community and a tech ecosystem would perhaps draw more talent as well to New Zealand this is such a this is a field that was really dominated by academic world until pretty recently I mean there's obviously some vitally important research that's gone on in a lot of corporations above all the attention is everything paper Harold of the Rise of Transform models came out of Google but the history of academic labs really driving forward AI innovation is unbelievably important and you know this just popped in my mind about it but New Zealand is such an unbelievably attractive place to visit and visit for long periods of time goofy idea but you know it would be very interesting if the government said we're going to target every AI academic of consequence in the world and offer them the coolest six months sabbatical program come for a semester we're not asking you you know to move this vast distance you're in time but you know an incredible amount can be learned in a single semester from the right professor and I think if there was just a recruiting program to say every AI professor of consequence throughout the world we are going to give you a really really cool one semester sabbatical program to come down here you're going to do well financially you're going to be living in a beautiful place you're going to have this fabulous experience and potentially have an extremely high ratio of the world's great AI academics in there at any given time yes it will be a different group every six months but that's just fine I think that would be cool very cool I love it plus you get access and then you would train this incredible wave of students it's amazing how many people who are great in this field are right out of school just came out of these academic and sorry Sam I just got too excited no no Charlie and back to kind of our earlier point about educating our civil servants and our law makers as well and our business leaders right like I think the possibility or a small kind of cohort of people to be educated and to be really curious about what these kind of global academic experts are thinking about sounds like an EHF 2.0 John would you like to jump in? So question from Larry in the chat and I think this will be of interest to Erin as well what's your sense around intellectual property you know these models are trained on a variety of data sources some of which may be intellectual property what do you use the potential impact on creativity and innovations we know it so you asking sorry sorry you jumping masking the panellists now it's really tough you know how do you know what it's really trained on I mean you know stability gave itself away because it trained on so many getty images when it was asked to create an image in many cases it started reproducing the getty watermark and you know that's busted but it's really I think in many cases it's going to be very very hard to tell what body of work something was trained on and you know another this is obviously this is sort of a playful example but you know if somebody goes to school and they read a whole bunch of texts and they train their brain and then they start doing things because they've read a lot of knowledge nobody's saying they don't get to think new thoughts and create new ideas obviously it's there are lots of ways that that's inapplicable but it's hard you know I think it's hard to determine what something was trained upon and particularly if something novel unique and generated and generative is coming out of it I don't know if it's hard to throttle that too I'm not saying this isn't a problem it is one it's just a very hard one yeah I think it's it's the sort of thing that we're thinking about because wealth often accrues from kind of assets that used to be land and perhaps arguably in some ways intellectual property has sort of been that moat for a lot of different people and organisations and so the natural thing will be for interests to really dig their heels in but to your point Rob I think there's this kind of tsunami of just progress in terms of how we think about creativity how we and even our rights to some of this that have all largely been built on top of other people's work as well to not perhaps have the same kind of protectiveness and there are some areas clearly where you know if in pharmaceuticals you spend billions of dollars and the protectability and defensibility of that and entire business models have been been made but this the area where yeah at lighter kind of touch areas where we're talking about trademarks or copyrights that's really up for grabs isn't it in the next kind of five years as we think about what it really does mean for us to produce unique works and what kind of ownership we have over that given that again a lot of our energy can go there simultaneously our energy can go towards new possibilities and new forms of media new forms of creativity that we have yet to discover that might again supercharge personal and collective productivity if we were to channel efforts into the kind of more abundant will view rather than the more kind of you know scarce will view of you know or protect what's rightfully ours but it's easy for me to say because I don't hold a whole lot of an IP or copyright and a bunch of stuff that's been plagiarised yeah just really quickly I'm really curious what the evolution of ownership is and and given the 3 nft space I'm just really curious how this is going to rapidly evolve and we'll see how a lot of court cases currently are trying to set precedence for all this kind of stuff and so I'm just waiting to see how all this plays out and it's interesting that the United States copyright office has stated that works created by non-human entities like AI can't be copy written so that's kind of an interesting thing or okay I can build something standing in the shoulders of giants but then I don't get to just take it and I'm curious how that will evolve I think the same it can be said of patents as well you know can you patent something that was generated by an AI and how would you be able to tell is that a regulatory thing that we could do that might might offer a little bit of a safer harbor the USPTO has already said no you cannot patent something I think somebody actually created as a specific test of this somebody submitted a patent quite recently that was very specifically attributed to the AI that he worked with and the USPTO said no so that's at least you know narrow case of patent office but I think it's some ways these run the risk of being very naive regulations because I actually think we're entering a word of what I'll call centaurs which are you know the mythical beast half person half beast now half person half machine and I think that's certain great creative collaborations are going to be the results of somebody who's a very good visual artist working carefully with Majurni and creating something that's a synthesis of their strengths you know as a writer you know I haven't really done it yet but I could imagine at some point in the future I might be collaborating with an LLM to you know be more prolific than I would be otherwise and to create things that I'm perfectly proud of and willing to put my name on that are distinctively mine but we're in some way accelerated by an LLM and if there is a blanket you know okay fine if you hit the button and out pops of work and you did nothing but a four word prompt can we say that shouldn't be copyrighted sure I don't have a problem with that but when it's a really intense collaboration between somebody who is an artist and a master of a tool and that tool has generative properties I think it's a grey area we need to be very careful about there's a great book called Homo Deus that I kind of scanned recently that talks a little bit about those centaurs Rob that we're kind of moving towards this kind of digital reality in the next step of the human kind of journey I'll definitely plus one to that that was you've all harari right yeah yeah I'm also kind of wondering if there's like a second order derivative that might eventuate from some of this like in in covid for example we all jumped on to you know platforms like this and as soon as we're able to we flocked to events and we kind of were longed for this kind of human to human sort of contact and so as we kind of see this immediate kind of rise of a lot of automation we might kind of also move into kind of more analog there might be a whole kind of secondary market of much more analog systems patents that are exclusively human or perhaps artworks that were hand drawn and there might be for those who can afford that luxury I think as well there might be just a rise in this cottage niche industry as well yeah I love the example of someone pointing out you could get a highly optimized supply chain built Blu-ray DVD for $30 but yet $500 for this handcrafted pottery bowl and there's this difference between the handprint of a human versus this highly optimized consumer electronic and the value that's placed on both of those things it really seems that in the future authenticity and humanness are going to be qualities that will be especially valued and what does it mean to be human and even what is our truth if these models get so good that they can generate realistic content what can you believe if you don't see it or hear it in person and do we have to be so dogmatic about our definition of human as well because one thing we haven't really touched on here is it's kind of like the intimacy and the kind of loneliness problem that a lot of these agents are beginning to solve and we're seeing people bear all to a chat bot on your phone and actually form these very real kind of relationships with agents the likes of which have all kinds of consequences but as we begin to kind of form different relationships and write a new chapter I guess in our humanity but you're right there's some kind of very real face to face things that we're probably going to value and treasure much above or the more commoditized kind of services that we're going to be offered soon we really have access to I'm reminded of a movie that came out 10 years ago by Spike Jonze and just the future that that presented and although I'm not so keen on the ending of it there's a real I think hopefulness as well so I've got just one question for the three of you Bob and then I'd like to open up questions to our audience so what recent and coming potential development in generative AI has you most excited what's the one thing that you've seen recently or on their horizons particularly got you fired up and alive for me is what Rob has mentioned before I've dabbled in kind of educational technology in the past I realised that there was not a strong sort of middle management workforce and that was all kind of linked to education as well as Cambodia's kind of history and part of that challenge was of course there's just no quality Cambodian content for preschoolers and we can never get enough teachers to go to places where they're most needed and that's a human scaling, an institutional scaling problem a lot of kind of what has been done with MOOCs but supercharged now I think potentially with generative AI opens this huge world of possibility for billions of people to be a part of this kind of global conversation and all the privileges I think that offers as well I think that's super exciting I really do believe there will be some fantastic innovations I do hope that they get evenly distributed across the global south and not just the global north and I think that's pretty exciting compounded with that would be Cambodia's role in perhaps hopefully making that happen so that would be my my two big ones keeping a close eye on what's happening in open source which again that we touched on briefly before it's simultaneously terrifying as well as pretty exciting and it's immediate kind of application to education I think one thing I saw recently was a demo of humane's device how it lives on your chest and conversations within can speak to other people in different languages and translate things for you I think there's some definite pros and cons there but I think as a personal computing device that works with you in your own context I think we've barely started to see what's possible I've been seeing some research that's coming out there's a model called ESM it's an LLM that came out and stands for evolutionary scale modeling and I think ESM2 just came out and ESM1B was out a little bit earlier and what it does is basically instead of letters it's amino acids and it's basically every protein every significant genetic sequence probably a lot of just natural peptides small as that are in this model and it's been particularly good for antibody research at this point and I think the potential to create novel proteins that aren't currently being coded for in the human genome or any genome but have the chemical properties and the dimensional properties the folding properties which were now much much better predicting than we used to be thanks to AI as well a lot of work that a deep mind on that one has just extraordinary potential for life science and so that's something that did create this really interesting LLM people are starting to get particularly in antibodies that hasn't apparently done as much as enzymes yet but that'll probably be coming soon I think there's just a lot of life science potential here because life uses just a different alphabet with 20 letters instead of 26 or 4 depending on how you look at it or 64 if you count all the codons but you know what I'm saying and of course there's the corollary risks of that of people using generative AI to create novel pathogens as well particularly with the availability of desktop CRISPR printers oh yeah that's something I worried about professionally for about two years John would you like to jump in yeah another thing that came out of Facebook I still call it Facebook just because of my type there is this tool at this AI tool where just I have a one year old and a three year old when you do it all when you upload a doodle this AI model can animate it for you and I just love whatever doodles my boys make to then use this AI tool to create these kind of like story books for them and that relationship I have with them then to showcase this tech at their own level is really fun for me I'd add one more which is in the hitchhiker's guide to the galaxy there is the babel fish I deal with a lot of people who are forced to learn English and I am in constant admiration for how they have learnt to communicate so eloquently in a language that is not their native language and I hope perhaps there will be a babel fish coming up soon because of all these LLMs even so but there is even working communicating with animals right through AI that can open up a whole new kind of world of possibility for us in terms understanding our place in this world and bring about a whole lot of empathy and the way in which we communicate with each other so yeah go the babel fish I'd like to open up to questions from our audience we've got about five or six minutes of questions and I invite you to either raise your hand physically or virtually or post a question or chat would love to hear your questions why people are just thinking of one I'm just going to throw something in the chat it's a link that the previous panel recommended for reading it comes through daily but they all thought it was their best site that they go to but a quick read on companies that are using AI that you've been talking about today Thanks MC It does make me wonder at what stage people actually still be reading things given that so much content's been generated I don't have time in space to read this exponential kind of sorry do I have to cut someone off No one of the analogies that comes to mind for me is the song The future's so bright I've got to wear shades and it's this incandescent bloom of content that we now must filter so in fact it may be an apt metaphor for how we might be looking at the world in the near future we need better filters Yes please Yes Trent yes what's your question Hi so it also reminds me enough to line is nobody knows what it's all about but everyone thinks it's great which is from They Might Be Giants but I'm interested I saw this in the idea that indigenous knowledge is not often written on the internet it's not documented or not documented in the ways we know and obviously as we move forward if we're training things based on information available it may not be available and may lead to a further demise of deep knowledge of place and etc and I'm interested maybe Sam has a comment about that but maybe anyone does I'm not sure I certainly do Trent that was one of the first things we raised is a agency in collaboration a little bit with other agencies particularly groups like UNESCO whose job is to preserve heritage and indigenous knowledge but you're right that ability for us to double down not just on the digital divide nearly a billion people still have no kind of access to internet let alone all the more advanced things but our ability to digitise there's some technologies like voice I think Rob you mentioned this a little bit earlier that we're going to move off written text which also lowers the barrier from a literacy perspective and supports oral traditions but the importance really for us as a society to deem that valuable to be able to be put effort into digitalising that because otherwise we do end up and we already are in a place where we've got a monoculture in the same way that corns dominated our kind of agricultural landscape we sort of exporting these you know it's a form of digital colonialism it's just kind of exporting a bunch of values that are baited in into these LLMs and into these kind of models and so if we're not careful and we kind of just willingly kind of accept market forces will determine that for us in large part we will get increasingly homogenous in terms of how we view the world and that's again back to where I think New Zealand can really lead on the world stage we can demonstrate how we can do this better and demonstrate that there is an alternative and that's kind of something that we hold dear as a society we'll tie for one more response or one more question what's emerging in the room there was an example of that actually at a level in an event that we're at Sam when everyone was doing their reflection circle and they were doing their thank you of the event and someone read out the response from chat GPT that covered everything everyone else said in the room except for the part on honouring Iwi and all the rest of that with it it was actually because you're right because there was no content on their effort to pick it up so it's already been it was a really prime example of how it came through and we're like wow yeah it was really interesting A couple more questions and comments came in one from Kelly is there a common public myth about AI you think we need to dispel That it knows everything I think this is one that most people who are in this space already are quite aware of we're lazy people and we are going to basically turn to this oracle that is the AI personal assistant for everything and that is terrifying but also or too attractive so yeah I think that is a myth that we might willingly buy into John What came up for me is that there's a lot of concern for AI and the thing that I hope people can be more precise about that is I'm less worried about AI I'm more worried about the people and the companies directing AI and I think the focus has been on artificial intelligence and what things are coming out of it but it's still at the end of the day it's the people and the companies intent that's really driving things I think one myth might be that governments are in charge if you read some of the people who are worried about the existential risk posed by AI not generative AI specifically there are some very very sound arguments there and Sam Altman himself said that he fears that this has the potential of nuclear weapons or worse for all of the imperfections of the US government in the 1940s it was nonetheless the delegated arm of a large democracy of people that ended up developing tremendously powerful weapons and the people who had stewardship of those weapons over the subsequent decades were not all certainly coming up from democracies but they were representatives of governments that in some way or another were offshoots of societies that in some way or another represented at least some perspectives of those societies in this point these unbelievably powerful and potentially wildly destructive tools are coming out of private companies which in some cases are quite tiny and in no cases were elected by anybody and you know the amount of resident knowledge within open AI about AI probably exceeds that of any government on the face of the Irish and so we are looking to make us to rain things in and put guardrails up and we should but you know we're entering an era in which you know governments were more or less hegemon when it came to you know technologies processes units of people that could shift history that is over it's in private hands it's in garages it's in universities it's on servers and you know it's just something we have to get used to yeah definitely want to plus one that yeah I think our ability and our courage to rethink institutions it has to happen because they were designed for quite a different era and expecting to behave differently is full hearty I see there's a question about disruption from Yacal yeah we've got a couple of minutes just to explore so past empires have often fallen with the arrival of new resource bases is there a way to avoid cultural collapse due to disruption and the final parting quips for moments of them I worry a great deal about something that some people have termed reality collapse you know if you'd ask me in the early days of the internet where was the internet going to take us I would have thought although I wouldn't have had the vocabulary yet be some combination of Wikipedia and Snopes just an incredible self-correcting self-organising hive mind that relentlessly seeks out truth and untruth would have a very hard time surviving the onslaught of fact checkers and the natural human desire not to be suckered or hoodwinked but instead it's the natural human desire to always be right that seems to have won out and you know many people have come to the point where they believe that they're entitled to their own facts not just their own opinions and reality collapse to me it's a term that some people use in a differing set of ways is whatever set of facts a certain group of people whether they're entitled to the media could be synthesised the video the audio the news articles the you know articles of you know by allegedly certain very well branded publications can be synthesised to support that point of view whatever it is and that you know I don't know cultural collapse reality collapse you know yako's question says cultural collapse I don't know how we avoid this but we better start thinking very hard about how to do it I'm just conscious of time we've got a hard stop for a couple of people including our panel on the hour so I just want to say a huge thanks to Sam, Rob and John particularly Chris for moderating it's been a great conversation and you can find the recording it will be on the website and we've got one more part coming up to the series on the 11th of July just furthering the discussion and Chris will be moderating again so hopefully the team will see you there on the 11th of July but thank you Kei Kete and enjoy the rest of your days thank you