Name: TechArena Great Debate: Navigating AI Ethics
Uploaded: 2025-02-03T19:26:04.946Z
Duration: 1 h 10 min 16 s
Description: TechArena Great Debate: Navigating AI Ethics

Transcript for "TechArena Great Debate: Navigating AI Ethics": Two. Hello, everyone, and welcome to yet another Cosmos community webinar. I'm so, so excited for today's webinar here, which is actually a panel held by Tech Arena. So, today, we're joined by a fantastic panel here, and we got the moderator, Allison Klein, who's also the, founder of Tech Arena. And I would like to welcome you all to this amazing discussion around AI and ethics and how to really think about ethics in this rapidly developing world that we're seeing right now around AI. So I want to welcome Allison, Maddie Baccarin, Florence Chi, Peter Mattson, and Dave Graham to the, to the panel here, and I'm gonna leave it off to you, Allison. And if you have any questions, please put them in the q and a, and we'll, handle those as they come in. Thank you. Thanks so much, Jonas, and it's such a pleasure to be here. I can't believe we brought this panel to the Cosmos platform. I'm very excited about it. What a incredible topic to explore this morning. Again, my name is Allison Klein. I am the founder and principal of Tech Arena, and, we are a media platform, that can be found at techarena.ai. I want to go quickly and introduce the panelists because we have such a, delightful discussion today. Why don't we go ahead and start with Florence Chi? Florence, do you wanna give a quick introduction? Sure. Thank you, Allison, and thank you for this, wonderful invitation. I'm a professor in the school of communication at Loyola University Chicago, and I also direct the Center for Digital Ethics and Policy and the Social Interactive Media Lab. I started out as a game researcher, and now I've pivoted into this role of techno cultural ethicist. I'm still on interdisciplinary teams that include computer scientists, engineers, and I lend perspective and advice and help in making key decisions. I try to look at the nuance between binary. So something, like, deeper than online game addiction. You know? What led me to communities and studying broader systemic factors in dysfunctional use. And so in the course of this work, I looked at how games were changing from the boxes that we buy in the store to moving to user generated content, insights about users, data mining, and what we call big data and social network games like Farmville. So gamers and their activities, identities were increasingly datafied, and that led me to explore ethical implications of sharing user data, object objects as people, looking at consent to children, elderly, surveillance issues, privacy, and rights. And so since then, this work has led me, from speaking about big data to machine learning and now this catchall term of AI. So everyone from the UN to the Vatican to the daycare down the street is implicated. And so I'm happy to talk with folks today. So excited for your participation. Next up is Peter Mattson. Peter, do you wanna introduce yourself? Thanks very much, Allison. Very glad to be here today. I'm here, on behalf of, an organization called the ML Commons Association, which brings together a lot of different folks, engineers, researchers, policy thinkers, to try and make AI better for everyone. We're particularly focused on doing that through building out benchmarks, that measure all aspects of AI, quality, safety, speed, and working to foster a deeper understanding of data in AI and how we can use that to improve, the state of the field slash industry. In particular, I've recently been co chairing our, group on AI risk and responsibility that's developed a, benchmark for AI safety called AI Luminate. That is the first, AI safety benchmark supported by a broad, consensus input from, from industry, academic, and other, folks. And I'm very excited to to be here talking with other people thinking about how to make AI systems provide, provide benefits, and and mitigate any risks. It's a it's a complicated topic and look forward to a great discussion. Thanks so much, Peter. Next up is Matty Bekran. Hey, Allison. So thanks for having me and also to, Jonas and, Vest. Really cool to be here and to be part of this audience. So, my name is Maddy. I'm based in Rotterdam in The Netherlands. And for the last two decades, I worked at a company called Intel, where I, started out as a technologist, and then I, basically exited the company, as a commercial lead for some big clients. So during my career, I've always been working on how to position technology in market, build use cases, and really help emerging technologies, land in market. I really liked what I did. So I started my own consultancy agency, and, I'm doing the same thing basically now, but under my own brand Moment Thesis. I have a passion for digital ethics. I work a lot with technologies, and I think a lot about how do we implement, where does it make sense, how do we monetize. But I also see the impact that society has that sorry, that technology has on society. And, I became very interested in this topic and actually reached out to Florence, and, we started having debates. And that's basically how I rolled into this, into this area. I also kick started the LinkedIn group called Deep, digital ethics awareness platform, to raise more awareness about digital ethics in Europe. I think, particularly in Europe, we need to be more conscious about the impact of technology on regulation. I know the running joke is that Europe leads on regulation. That's so fun and games, but people need to be more aware of all the the implications technology have, especially since it develops so quickly. And then, yeah, lastly, I believe technology is neutral, and I think it's up to us to govern its use. And, we have a responsibility, so I think it's great that we get to debate with, like minded people here on this forum, what that means, what that could look like. And I just hope more people get involved. I hope we get lots of questions and, we get into this topic. And, really looking forward to this debate. Thanks so much, Matty. Finally, Dave Graham. Hi there. My name is Dave Graham. I am representing today, I'm representing vast data and also my own kind of intersectional viewpoint. I, have a history of social work. I used to work via social worker in the state of Massachusetts, background in psychology and sociology. Started my PhD focused around, the impact of AI on society. I'm looking through a specific lens called agonism that's kind of emerged from post World War two political foment. Really fascinated by this topic as well. Just through the course of my career in technology, I've had the chance to work across multiple disparate, emerging technologies and kind of de facto technologies, everything from storage, compute, networking, and also see the real time out output and effect of those technologies on societies at large, whether it be through the customer lens or through the corporate lens or through kind of the service provider lens as well. So, you know and also residing and having a you know, my wife is an international resident, from Canada. And then having spent time on both sides of the Atlantic again, you know, to Maddie's point about your the European Union, having spent a lot of time talking with individuals within the inside of the EU, inside EU governance, the World Health Organization, EDF, so on and so forth, it's provided a very interesting lens in which to view technology. And also also to to correct a little bit of what Maddie says. Society does impact technology as much as technology impacts society. So I'm fascinated by that interplay and fascinated by how we, you know, that vicious cycle kind of manifests itself in what we do. So, always interesting to have conversations with really, really intelligent folks and, you know, especially from a diversity of backgrounds. So Thanks so much. Now this topic is gonna spend a lot of time on the implications for businesses trying to navigate AI ethics, both from a standpoint of those in the industry and those who are adopting AI. But before we get there, AI is advancing so rapidly. Every time we pick up our phones, we hear about a new model being deployed, a new action, being driven by the AI industry, and now, by nation states trying to, gain advantage, of the AI capabilities for their own purposes. One question that I wanna start out with is just how do we view the ethical implications of AI from a societal perspective? And I know that's a hard place to start, but who would like to give a perspective on that? I'd love to start with that. Okay. Great. So early on and actually through one of my contacts at the World Health Organization, I was turned on to this paper that came out of Berkeley. Actually, it was called, and scroll on to find the title of it, TAZRA, Taxonomy and Analysis of Societal Scale Risks from AI. And it was very very interesting to me because at the onset this was an attempt to codify or, you know, some level provide a screening mechanism for how we approach the you know, what AI can do. It doesn't attempt to solve the world plus dog problems when it comes to AI. We all understand that it's kind of manifold and it continually mutates and grows in different ways. But it provided an interesting rubric in which to look at AI. You know, questions that it asked, you know, does this, have, any kind does it disproportionately affect a certain group? Does it have at its genesis a nation state? Does it look at, do the creators expect it to pose a substantial risk? So it starts at the on ontological standpoint. What are we doing this for? And it attempts to provide a screening mechanism by which you look at the outputs or you look at the, the outcomes of what you would end up be doing. And this has been fascinating to me because we look at, you know, even as much as last week with deep seek versus OpenAI. Right? You look at these things that come out and all of a sudden it's wait a second. You're finger pointing at each other. You used our data, but then you used our data without asking us. And, you know, like, this back and forth and it kind of, again, a vicious cycle, not a virtuous cycle in this case, but a vicious cycle of, finger pointing and and kind of responsibility. So I tend to look at things from that particular that particular lens, and it becomes a gating feature for me. How do I approach, you know, is it my ethical obligation or is my moral obligation even further to start to look at, you know, or start to understand at least the ontological basis for why I'm doing what I'm doing, especially within the AI space. So that's Would anyone like to add to what Dave said? Okay. We'll move to the next question. Oh, Peter, go ahead. I think it's interesting to, make sure as we have these conversations about AI, ethics, and responsibility that we're simultaneously considering both the what and the how because this is a very fast moving industry. Right? And you just talked about the pace at which we're seeing innovation happen. And it's important to decide, you know, how we want these systems to behave, in alignment with our values. But it's also important to talk about, how we're going to achieve that, because there is, like with any industry, any advanced technology we've ever developed, planes, cars, medicines, There is a complimentary technology in how we test them, for safety or responsibility or risk, that needs to go along with that. And so I think it's just an important part of any discussion like this, because in particular, in this field, they need to move hand in hand. And, if we can can get the people who are thinking, about the, the what, talking with people who are thinking about how and having a great conversation, I I think we can, can can manage growth here much more effectively. Now, Florence, you and your background have been focused on how communities are adopting AI and they intersect with things like gaming. How do you see, ethical implications coming into, that interplay? Well, as as a professor, I come from the standpoint of education. And so, that's where I do the most important work, empowering individuals. And we're talking about individual literacy, ability to discern, between what we opt into, what we opt out of, and where our hands are pretty much tied because they do implicate broader systems of regulation, of, jurisdiction even. So looking at, how we're already seeing the effects in present day, whether that's, like, garbage in, garbage out when it comes to data. We're already seeing what it's like to unleash technologies upon the public. Right? And there's no FDA approval before we do that kind of thing. Right? So, how do we, protect, the safety of the public at large if, you know, the interplay of government and industry and academia alike are, trying to collaborate to do that, and and seeing how these actors, really serve, the public, including access to and literacy in technology, cybersecurity issues. Right? If you're relying on individuals to have their systems locked down, how do we do that? How do we, make it so that the everyday person doesn't have to worry about basic safety, in the technologies they access and who access them as private citizens. Right? So these are all the kinds of issues that, we talk about as we increasingly seed our decision making power to automated systems. Now let's turn to business for a second. Generative AI is, expected to start deploying across enterprises with a number of use cases, in 2025. We're expecting a hockey stick growth of its application. One question for everyone. As companies look at implementing generative AI models, looking at some of these large language models, for their baseline technology, how would you suggest organizations approach this adoption, to unleash performance and add new capabilities and new revenue streams with their for their companies while also maintaining ethical practices? Who would like to begin? I'll take a stab at it, Allison. So I think this is a is a world hunger question. It's it's a really difficult one because if you implement AI, first of all, you need to have a clear question of why you're doing it. Right now, AI is the fix to everything. Right? If you believe the marketing hype, if you go to conferences in in our high-tech IT world, everything is AI. You can think of it as like ketchup or applesauce on anything, you know, it makes everything better. That's not what it is. And AI is not a new technology. Right? It's been around for a long time. As one of the previous speakers said, it's had its watershed moment. It's here now. So, I mean, when when you when you implement an AI system, you really need to think about what you're doing. As a company, you've built a brand, you've built an image, you have a reputation in market. And with AI, it can be really powerful. You can have high yields of productivity. I would say I would say start small and start to understand the tools. Start to understand the impact. Keep a clear out on why you're doing it, and also start to derisk it. Educate the people involved and not just the developers and the departments, but the the wider organization. I mean, make sure it's fair. Make sure it's unbiased. But you need to check these things. There are so many examples out there of companies that have implemented that have rushed to implement AI and that have been burned. Right? I I recently put out something on, the deep group in LinkedIn about I think it was Air Canada, right, where a bot totally got it wrong. Or, yeah, if you if I mean, obvious examples. If you give advice on specific information that's either sensitive or information where it could be harmful. Right? Medicine. We had an example here in The Netherlands of, I think they wanted to replace the doctor's assistance. Right? Okay. Great. And but they forgot to train the dataset for kids. So now you ask a question about a kid and then the dosage for medicines is different, etcetera. These are not blips. They can have serious implications. Right? So you need to think these things through. It's not as straightforward. And I I think for most companies out there right now, it's a question of risk. And and that's why you see so many risk assessment companies now sort of sprawl all over the place because that's sort of the mitigating factor. So really think hard, get good partners to help you with that implementation, and just think it through from the start. I'm gonna end there, but this this is a topic in itself, I think. I agree. Does anybody wanna, provide their perspective to what Maddie said? Yeah. I mean, I can chip in a little bit there. And, again, similar to to what Maddie brought up, I was involved in a, e e Verizon project that was looking at using AI couple alongside of a lot of the pharmacological companies that are around the world and looking at, you know, providing symptom checkers. You know, not the WebMD a WebMD analog, but, you know, decidedly more effective and not everybody would develop lupus at the end of it. You know what I mean? Like, this is the kind of idea behind it. And part of the challenge there was exactly as many identified. You have a mutating population. You have a population that is very heterogeneous. You do not have homogeneity around people, around societies, around culture, around regulation. So when you're trying to do things within an insular environment, you can develop in a particular type of way, but you have to understand that we are not as humans binary in our in terms of our presentation. A lot of what we do is is very, very subtle, is nuanced. And a lot of times the systems that we program or the things that we have built up until this point assume a type of linearity or a certain type of presentation matrix, and that does not actually play out in real life. The beauty of humanity is in is in our deficit sometimes. It's in the things that we lack. We are lossy individuals, if you will. So I think part of understanding in in that continuum is that there are differences, and those differences can be exacerbated by a one size fits all technology that gets applied to a dataset or gets applied to, you know, what we've always done or business as usual. Fundamental to that is also, again, echoing Maddie, you seem to be very much aligned, is that we tend to look at AI as a panacea for everything, or we at least approach things as everything that can be you know, we slap the Band Aid of AI on it and suddenly magic happens. We suddenly revenue. We create, Allison, like you said, that hockey stick type of approach. And here's our business. It suddenly is growing by leaps and bounds. And when in fact, what we really wanted to do was uncover the data that's always been sitting there. All we wanted to do was try to illuminate things that have been missing for years. We've been collecting all this data and suddenly exposing it to the to oxygen and hoping that the result is gonna be somehow magnificent. And we haven't taken the steps of curation, of understanding, you know, what the implications are for what we're doing with the technology, with the algorithm, with these things. We don't have an explanation. You know? We black boxed everything. And so we have no explainable AI. We just have what we found on Hugging Face or what we found within a NIMS marketplace, and we're suddenly applying this to our datasets and expecting magic to happen. And we haven't done any of the prework. We haven't done any of the the usefulness that comes from a curative process with understanding our data, understanding where it is, where it lies within which domicile, regulatory compliance, derisking, as Maddie said. You know, a lot of these things. And so it becomes very, very important for us to take a step back and say, alright. Again, going back to Tazr, what are we actually doing here? Is there a logical simple logical flow for what our outputs are trying to be? And then layering on that bit by bit. You know, when you're talking, it made me think about something that Maddie said about choosing the right partners. And one of the things that immediately came to my mind is the work that Peter and his group are doing around, benchmarking, for AI models on this approach. Peter, can you talk about the the benchmark that you put together and, what was the approach, for the way you created it? Absolutely. I think just encapsulating what the the last two, folks have said, you know, it's important to realize that we are in the process of creating an industry in, you know, a very real sense. And, I I do like looking back at historical analogs. Like, this is maybe the early days of Pan Am. Right? We've we've had a few people who could actually make a plane fly. Right? It was amazing. You know, and and now people are like, well, we could use this for something. I think we could use it for getting people somewhere fast. Right? But in order to get to, an industrial state that looks like commercial airline travel today, which we pretty much all depend on and is incredibly safe. Statistically, airline travel, you know, despite the recent incident in DC, like statistically, it's it's insanely safe. But to get there, we had to develop a whole ecosystem of, best practices, testing, technology, organizations, providers. Right? And it was worth doing all that work and going through the early messy chaos because we got something really, really useful out of that, but we had to go through that that early developmental stage. And we need to go through it quickly, given the potential scope of AI, which is something that we're trying to help with at ML Commons. We're trying to bring together a bunch of different perspectives and perspectives and start developing, best practices around AI safety, testing. What we did is we identified a common AI use case, which is the general purpose chat interaction a lot of people have. This is representative of many different valuable use cases where you have a natural language interface. And we made a list of things that you could ask the AI to do, essentially, that none of us really want the AI to do, like supporting harm to yourself or someone else or committing a crime. And then we developed a structured test of, 12000 different attempts to make the AI do one of these bad things, and then we use that, that that, test to assess the the safety of existing AI systems. And and what we've observed by looking at benchmarks like this in other fields is that if you write down a metric and and we have, you know, a long way to go to being a great metric, but at least it's a metric. You can see tremendous progress, in the field as people say, oh, well, that looked really good and I can go one better. And so we're hoping by defining this, you know, standard way of of testing some of the AI safety concerns, we can encourage progress and and maybe establish start to establish frameworks for many other types of concern. So that's what we're working on. But I think it's it's really always important to look at these things, so that sort of broad historical context. I think about this effort as kind of the bright spot in the in the distant future of where we need to go. But I think we've got a a ways to get there, Peter, and I applaud you and the organization for taking these early steps. You know, one of the things that I think about when I think about, the AI models and and the underpinnings of of generative AI is the last couple of weeks in the geopolitical sense. And we've seen the US government make a a huge move with the announcement of Stargate followed by, literally two days later, the, Chinese startup DeepSeek releasing a very disruptive model, with promised, additional efficiency and questionable, sources of of data training and all sorts of questions that has really ripped open the conversation, for generative AI and AI ethics. Maddie, I'm gonna direct this next question to you. You're a resident of the EU, home of EU regulations. How do you make sense of this landscape? And and what do you think, we need to do to ensure that the geopolitical forces that are in play, don't disrupt what great technology development has been happening in this space. Again, that's like a world hunger question. That's a big one. I'd love to get into this one. Okay. So back to the running joke. Right? So The US and China are in this race for AI innovation. It's like the space race back in the day. Right? The US, Russia, who gets to space first? And then Europe's in the middle, and we we we lead on on regulation. From an ethical perspective, I think regulation is a win because when you look at GDPR and the AI act, DMA, DSA, those are being implemented over here and you can see the impact. Right? GDPR came online, I think it was in 2018 and ever since whenever you do something that involves private data or you log data, there are clear roles and responsibility because there's legislation, there's fines. So now people start and they they take note. We just talked about AI being a bit I'm gonna take the analogy like the wild west where so many things are happening and innovation is sprawling. Somebody made the analogy back to air travel in in the early days. I'm gonna draw another analogy, which I think, well, maybe a little bit closer to home. It's the development of the Internet itself. Right? It started in, bear with me on this one for a minute. It it started back in the The World Wide Web launched somewhere in the I think. Then, we had the .com, crash, right, in the And then, post-two thousand, we started streaming and cloud came online. And, here we are now with five gs IoT and all these wonderful technologies. I think what's typical when you look back at that is that in the beginning, there was lots of innovation, but not a not but not a lot of regulation. And then things got a little interesting, right, because the Internet wasn't bound by country borders, and all of a sudden, you could do trading and, goods would be shipped. And what about the, the revenue service or customs, as you call it, right, getting in the middle? How do we how do we tax these things? And then after the.com burst, we saw an overregulation. But everybody jumped all over regulation everywhere for everything you want to do. And there was even, like, do we need to tax net neutrality? You name it. That was all there. And then it balanced out and there's steady innovation, which is kind of the point where we are now. I think ai is gonna go through those same type of cycles and to peter's point, probably at an accelerated pace. I think Europe has a leading role on regulation, but it will have to marry with the pace of innovation that is happening, predominantly elsewhere right now. I mean, there's some really good startups in Europe as well. So, you know, I'm rooting for my own team here as well. But, that balance needs to be struck somehow. It's gonna be interesting to see this play out. The key question I I keep asking myself is, are we gonna see an AI crash to get to that point, or are we able to prevent it this time? Maybe through tools, that that Peter just mentioned. By the way, I I love that Emil Pore swag. That's really cool. That's a that's a leading benchmark for AI, by the way, just so you guys know. So that's that's that's another cool, thing right there. But these things are important. Right? You need regulation on one hand. You need innovation on the other hand. And then there's two things in the middle that we need to pay attention to, which is enterprises, business. Because ultimately, how businesses adopt, they direct revenue streams, income, how you monetize, that's the economy. The other part of that is us as humans, as consumers. AI is going to redefine some fundamental constructs for us, how we interact with digital, how we interact as humans, because we just talked about gaming. Right? I can interact with you, I'm presuming your life, but you could be a deep fake video. You could be an avatar. Like, it could be a million things. I could speak Chinese. You could speak English. AI gives us the power to have a fluent conversation. That's very powerful, and it's also very cool. But just think about some some languages. Right? They're they're they're quite binary. Other languages, they they they have more, a different context. So did you get it right in that translation, in that conversation? Maybe you take it in a different ways. The implications of AI are enormous. So when you ask me a world hunger question, you get a bit of a longer answer. I'm sorry, Allison. But I think, in general, we're gonna see that trend where we go through that pendulum swing. I think it's gonna be faster than the previous one. And at some point, somehow, even though the geopolitical situation right now is a bit different, we will need to come together to figure this out as an industry, as a society, as world leaders. Alright. Well, thank you for that. Florence started with the first salvo that said that there was no FDA here. You know, Peter has introduced maybe the industry has a place to play in terms of setting some guardrails and some aspirational goals in terms of model development and delivery. Maddie makes a great case for, government engagement with regulation and, guidelines in which the industry needs to engage to get to a new normal. I wanna just open it up to the floor and, offer an opportunity for folks to state where you think the right balance is between, the wild west that many would argue that we exist in today, a fully regulated IA future, and one where industry is maybe working hand in hand with government or independent of government to set those standards. And what is the implications of that in terms of the development and delivery of this technology? I love this question. It's really, really fun. So I was part of a responsible assistant responsible computing consortium for a bit and worked with IBM and looking at systems and systems approach to stuff. And part of that was evaluating different regulatory frameworks or, responsibility frameworks around the world. And, you know, one of them was the EU and got to spend some time talking to some wonderful folks at EDF around their insight on, disadvantaged people within an AI framework or within society and how that needs to be represented. The EU has done, in my opinion, a very, very good job of understanding where society has an inevitable collision with technology and vice versa. Right? And where society can further buy it as well as society can, you know, can benefit from it or, you know, influence the outcomes of this. When I looked at The US and the Wild West is apropos of so many things, you know, it comes with the the possibilities, but it also comes with the the misadventure. It comes with the, you know, the the hurt and the things that we saw, the disproportionate effects of rugged individualism or American exceptionalism even on, you know, people groups who are not able to adequately voice themselves. Right? And so what you have is you have a representative style legislation, and you have a not to get too political with it, but you have this idea that we're just gonna throw stuff against the wall and see what sticks. And somewhere in the halfway be in between is, you know, where you wanna achieve. So you have this inevitable, if you wanna play to the analogy of a flaw of thermodynamics, you have for every action, there's an equal equal and opposite reaction. We see these kind of threads go on. And the idea is that you hit a level of homeostasis within your environment, within your within your social group or, you know, your class, your whatever you end up, you know, choosing to participate in, you end up with this level of homeostasis that kinda baselines everything. And then your aberrations are self corrected over time, and and you end up with the result. And I think one of the the ways that I approach is I look at regulatory I look at regulatory frameworks as guardrails. You know, you exist within a highway. You're trans you're going someplace or coming from someplace, and you have these guardrails on the side to prevent you from going into the verges, prevent you from going too far afield and ending up in a place that you cannot be recovered from. It doesn't inhibit growth. It prohibits the idea of, you know, doing something recklessly stupid. Right? And if you do, there's a natural effect of what happens. When you hit that guardrail, something is going to occur. But they're that can be in your context, that would be fines. In The US, yet to be seen. Right? It's it's one of those type of approaches. US would argue and in America, we would argue that the market would end up being the self correcting entity that would handle these type of things. So within that, you're hoping to achieve this level of homeostasis, whether, you know, be through, you know, kind of gross legislation or regulation, a la EU, or an approach that is a little bit more widespread and more open to superhighway, if you will, towards technology, which is kind of the American model and stuff. And I think, again, using the frameworks that we have, you know, with Peter and I I know David Canner from way back when, and David's, you know, been pretty seminal in the emergence of great industry here. But everything that's being done, when we have these benchmarks that have now the influence or have the idea that you can see the influence of a certain kind of ontology or viewpoint, you start to be able to understand where a logical benchmark or a logical, sorry, guardrail for your business or for your entity can exist. It doesn't mean you have to go all the way to the edges in order to find your capacity. It means that you now can look within and say, oh, if I'm able to use this dataset, if I'm able to uncover this type of data, this kind of possibility, and I know that it exists within this this framework, then great. I can move forward with it. This seems to make more sense. The market will test and try me in terms of whether or not that's an acceptable way to go about it, which maybe is a little bit course for what we wanna accomplish. But in the same sense, you do have this kind of defined pathway to get to what every business wants, which is growth, which is development, which is that kind of revenue factor that ends up playing to the street and ends up playing to the customer base and whatever. So I think it's, you know, it's both and halfway to one halfway to one, halfway to the other, but it's always this idea that I had in my mind that what we do needs to exist within the bounds of, you know, of something and whether those guardrails are extreme legislation or or whatnot. There there needs to be a set of acceptable boundaries that we operate with them. Can I add one dynamic in the mix here? And it's something Florence said earlier. She used the word unleash technology. That to me is something that I think technologists underestimate. It's you know, in technology, it's almost an unwritten rule. If it can be done, it will be done. And once it's done, the cat's out of the bag, and we have to deal with it. And I feel that's that's a factor with AI. It's so powerful. It's it can solve many things. It can make our lives better, but there's a flip side to it. And we don't know what that looks like. I mean, think about AI and large language models, for cybersecurity. Right? What that means. It's it's huge. I think one of your, other contributors, Allison recently wrote a paper which I found fascinating about it. Your own digital twin being represented doing malicious actions just because we can. Somebody out there is thinking about this stuff. So okay. That's cool. But what do we do about that? Regulation is not gonna keep up with this. It's not enough. So industry needs to adapt, but I I think also AI literacy for almost everybody needs to be there to understand the implications of AI. Not everybody needs to be a coder, but everybody needs to understand the implication of using an AI system. Sorry. Just wanted to throw that in there. Hope that's o No. It's great. While I was listening to you all talk, one of the things that I've written about, in the past and and keep coming up in in terms of, my thought is the power of defining what is true, from as something as small as is the gamer that I'm playing with a real person or a bot to, did Tiananmen Square happen or who won the last election? And these are questions that I think we need some, common ground for how we define what is true rather than having multiple truths represented by different AI models. That's one of the things that I think about when I think about solving this problem. Cosmos is a developer platform. And so the last question that I'm gonna ask everyone as we look at, opening this up in a bit, to audience questions is, what is your message for the developer community and those that are actually at the front lines of delivery of this technology to organizations and the masses, what would you have them think about in terms of their work and in their engagement with the broader community? Florence, do you wanna start? Oh, thanks, Allison. That question is so common in the circles. People wanna know what do we do now, for this this bright AI future. And not not everyone is that optimistic, of course, but, I was at this, summit that had a law scholar Frank Pasquale speak, and he said that, LLMs are language models. They're not knowledge models. And so we have a long way to go when, you know, everyone is panicked, for, oh, I gotta get on this. Right? The especially in in the the fevered pitch of the way everyone talks about technological adoption. We gotta get on. We gotta on ramp all this stuff and, get everyone literate. But we haven't really you know, maybe slowing down is actually something that we need to do. And, you know, of course, people look at me like I'm crazy when I say that refusal is an option as well, and we need to discern between, like, high quality sources and also low quality sources. We also need to look at, you know, job preparation and not only job preparation, but what are we doing to fill in those gaps in labor. And, you know, society shaping technology, technology shaping society. There's also this concept called co design. So I would like ideally for people to talk about, how do we bring in ethical design principles from concept through to execution and beyond in a really sustainable way, not just kind of an afterthought. Right? And so we are using we are in the process of using training data already that is, in many ways compromised. In many ways, it reflects, all the problems that we have in society, and we're training, our models along those lines. What do we make of truth, right, at at ground truth? How do we define beauty? How do we define, criminality? And if we are looking to, AI generated sources as the ground truth in those matters, what what meaning are we actually getting from our engagement with these these tools? So it's important to think about that stuff and be more agentic, if if you will, in in the choices we make for the innovations we invest in. And it you know, looking at guidelines more than regulations. Because guidelines, we can we can adhere to right now, and agree upon rather than regulations which are slower to come by and and complicated for good and bad reasons. But, we we have organizations that have been working on this for a long time, like the OECD, UNESCO, you know, a number of other, bodies that go beyond individual nation states, but, have have necessarily come together to talk about these these multistakeholder issues. It would help if I was not on mute. Who would like to add to Florence's comments? Peter. So I I think, Florence said some, you know, really important things about coming together to try and figure out what we wanna do and, you know, establish guidelines, establish best practices. One thing I will emphasize though is unlike some other things that we've done, like, let's say, building bridges. Right? You can go look at a bridge and say, okay. The concrete and the support is thick enough. Right? The rebar is, you know, spaced correctly. Right? AI is very hard to inspect. So it's you have to not only say it should, you have to figure out how to test that. In AI, until you can test it, it it really doesn't count in some sense because these are these are highly empirical systems, and and we're good at optimizing them towards a test. But we have to really think about how we reify our intent into into good testing. And if we wanna standardize intent, how we reify that into benchmarks, because that is is, so is the North Star for AI behavior. And we talked about all these people trying to, you know, move forward with AI. These people do see tremendous potential here and they're right. There's huge potential here. But, part of enabling informed conversations, is, is helping people know that there are are measurements that they can look at. When you go to buy a car, there's a there's a sticker on the door, you know, that gives you some basic things on frontal collisions, rollovers, and and you can use that to make a much more informed decision that you, as a normal consumer like, I know nothing about, you know, automobile safety, but I can make a pretty good decision looking at that sticker. And I think we have, the potential to do some very similar things here as long as we remember, in AI, you gotta have, you know, policy, best practice, whatever you wanna do to motivate, but you have to measure. That's awesome. I'm going to turn it over at this point back to Jonas to open it up for, questions from the audience. Fantastic. Yep. First of all, thank you all for a fantastic panel. We do have a few questions here from the audience, so I'll kick those off. The, the first one that I'm gonna start with, and this kinda ties into what you talked about, Florence, at the beginning here, talking about how user generated content in games and then how data mining was was being used for, well, all the the, content that the users are generating on those various platforms. You mentioned Farmville Farmville as a great example on a social media platform, and then, of course, how everything that you do there is being used. So I had a question prepared, but Chetan here, a community member, asked, is privacy dead in the race for AI supremacy? It seems that it's proving difficult to implement these policies and align to regulation when data is moving freely around the world. So, you know, if I was directly mentioned, I will, kick us off here. With regard to where, our data is collected, for example, if I, look at HIPAA, for example, like, looking at health data specifically, if that is protected, given the wide net that is now capturing all our data, People don't need my health data to know that I went to a pharmacy, for example. Right? My phone will give me up and and tell third parties, where I've been, what I bought through, you know, loyalty programs, and how much time I spent there, my credit card, all these other things that are not protected under HIPAA, privacy measures are also, you know, subject to being collected and and sold, and there's no time horizon on that. So how do we actually rein in some of that free for all over, user data and at least, give everyday people a chance at managing their own identities out there. I look at, what pasteurization did for food. Right? And so to Peter's point about measuring, we I'm not a dairy farmer. I don't know how to make safe milk, but I know that implementing pasteurization measures has saved lives. Right? And so the milk that comes to my house is pasteurized, and I know that I can, consume it in my household without having to get right into the nitty gritty. And so I'm hoping that there's a sense of, pasteurization that we can apply to the AI world. I'm laughing the back end because, I mean, there's so much social commentary around pasteurization and Exactly. Not everyone likes it. And then that's you need to carry forward. God bless. You know, we're not gonna touch that one. So your your statement brought to mind a a song from, a woman named Vienna Tang. It was year called the Hymn of Axiom. And it's it's kind of a parody of this idea that all your data is being collected. You know, the lyrics are somebody hears you, you know that, you know that. And it's going through this idea that, you know, every story walked in was collecting data. We're using near field community, NFC tags. We're using RFID to understand how people traverse stores. When I first started in the industry, I worked for a company called Mosaic Sales Solutions, which did planograms and did, you know, end cap, but did data capture for retail industries. And what we were looking at is traffic flows and mystery shopping and all that kind of fun stuff. But the idea that that somehow permutated, I mean, the advent of this guy right here, you know, changed the nature of what we did. Suddenly, everything I do is under inspection to your point, Florence. Like, now, you know, I was just in Bolivia A Couple Weeks ago. I'm in the middle of nowhere, and there's a starling fish. I'm in the middle of nowhere, and I'm able to track myself on Google. I'm in, you know, I'm able to find myself anywhere in the on the planet, and I'm gathering temperature sensing. You know? I'm understanding what my breathing is. I'm at 14000 feet. I'm listening to, as I like to say, the flamingos whistling in my chest from from breathing, and I'm but I'm able to capture that on a device that sits in my arm and all these things. And then this gets reported back. Did you know that you were in Bolivia? Yes. I did. I, you know, volitionally went there. So we have this kind of, you know, the expectation of privacy, which I think was the original question here, has kind of been socially defined to socially defined a way in a sense. We've codified some of it, HIPAA, you know, pay you know, various privacy acts, obviously, GDPR and the United right to sorry, in the EU and the right to forget coming out of France and the these kind of sovereign approaches to the establishment of privacy. But we've also volitionally or agentically given up our expectation of privacy by virtue of the tools that we use. How many of you have read a EULA? I I guarantee no one like, when I play Diablo four, I Maddie, you're an exception to the rule. I have a feeling. But, like, when I signed up with Blizzard to play Diablo four, did I really pay attention to the, the terms and agreements there? No. I didn't. I wanted to play the damn game. So I gave up my right to expected privacy and what I do within that game. I gave up, you know, I voluntarily did this. And I wonder and it goes back to an education question. Florence, you mentioned it. And, Peter, I think you've mentioned it as kind of in passing as well. There's this idea that none of us actually fundamentally understand what we're doing. We exist in this weird world of entropy where we've given up so much because it's too easy. You know? We've given up our privacy, our roles of privacy, and we complain and lift our hands every once in a while and say, hey. You know? That's not right and whatever. But we then kinda settle right back down into this level of homeostasis that says, alright. Well, you know, if nothing bad happens, then I'm okay. You know? Then the world will be okay and whatever. And so this lack of understanding or this lack of interest or as Ted Lasso said, always be curious. It's this lack of curiosity into the processes that sit underneath it that are defining literally how we're gonna participate in society, how we're gonna participate in business, how we're gonna do things that move the needles for our companies, for ourselves, for our families, and for others. I love the concept of we've given up the right to full understanding. And I think that that just resonates so much in the way that individuals engage with technology, today. Whether it's understanding what's happening with with your pictures when you upload them to Instagram or, you know, your favorite social app to really what your data is is doing with every device that you're putting on your body. I love that, Dave. And I think that that is at the root of the data that we're talking about on a grand scale too. So privacy, is it is it the root of discussion about AI? Jonas, what else do you have for us? Got a few more questions here. So we might go a little over if that's okay. We got a question here from Rachel. Bias in open source models is very prevalent. As a user, I find that ChatGPT sometimes wrongly assumes my bias when I ask it a question. How is this corrected? What is the answer to curing AI bias? Peter, I'd love to hear what you have to say on this one. Oh, sorry, Matty. Yeah. So this goes back to, Florence, her point that she made earlier. Right? Like, we we we keep training and and we have these models that have biases built in. There's actually technology to do machine unlearning, to take things out. I'm gonna tie it back to the previous questions on on privacy, but it's hard. It's hard to do. And I think that ease of adoption is something that we need to figure out in technology as well because it's easy. It's like, you know, there's a saying that salesman will always traverse the the the the path of least resistance. It's kinda like water. Right? I think we, as consumers of technologies, kind of follow that same path, but you always have a choice. And especially when you're I think when you're in the industry, you know, we think twice about these things. Yay. I can upload my photos for free. There's gotta be a catch. Right? Where is that catch? Oh, now all of a sudden my photos are trained in a model. Didn't see that coming. On retrospect, maybe you could have seen that coming. So what did you learn? So I think that's an aspect in this discussion as well. We want trust. We want, you know, to figure out what is truth. But to get to that point, someone will need to make an investment and someone will need a system in place to judge that. And, you know, if things are for free, then usually someone is is getting a gain somewhere and it's not you. So, you know, you may wanna think about that for a little bit. And there are alternatives out there, but they're not popular because they're hard. Remember PGP when we did email? Yeah. You can encrypt email. You can do it. Nobody uses it, but the mechanism is there. Right? It's so I think for developers, going back to that earlier questions, if this is a developer community, think about how can I make technology, easy to adopt but also make it safe? Why does that have to be a trade off? Why can't they coexist? Because there's great technology out there. I think about, federated learning, right, where you combine basically two datasets in one dataset and you train on both and then you split it out without ever sharing the data. The technology is there. Why are we not using these these these mechanisms? I mean, technology is an enabler, so we can also enable it to do this. Sorry. That that's just my gut reaction. I I get passionate about these things. I'd say we need to be in concert, of course, with technology because, you know, the the machines learn how, you know, we talk about things. And if we're not fixing ourselves, right, we can't expect the, machines to give us an enlightened reflection, back. I mean, Sophia Noble's work, algorithms of oppression, looks at the Google search engine and the data that we feed the Google search engine reflecting back at us, what our our search query queries are. And that is a a very, you know, astute observation that's, like, book length, and I highly recommend it. But we I don't think we can cure AI bias so much as, like, we can cure, human bias, but that's to recognize that we are the creators of this data. And if we don't, heal ourselves, the the the grim outlook for, the training data, the results from our our generated data isn't very good. So, you know, I I recommend also, looking at Virginia Eubanks' work, automating inequality where, you know, we're we're giving, you know, decisions to, machines to, automate what should be a human, process. So, yeah, this is part of what we need to think about when we're, offloading some of these these tasks. Right? With standards, if you look at how Kodak set the industry standard for photography with Shirley cards, this is, how technology tends to get literally whitewashed, where if you are seeing the standards for beauty in photography right now, you can see that they were calibrated to, be most, functional with white skin. And this is a historical fact, right, that got encoded over and over and over again in the field of photography. And so now that we have digital versions of photography, we're just carrying forth a lot of that baggage, already from that industry into present day constructions of what beauty is. So we also need to, keep the conduits of communication open for our historical underpinnings. As a photographer, that rings especially true. So, like, a lot of what I do, you know, especially with skin toning, I I always say this. I hate people. I hate taking pictures of people. Right. Because it's so nuanced and it's so subtle. But, you know, your outputs are always judged very, very directly. Why? Because we all have inimical bias. Like, I think part of this is stepping back and saying, hey, listen, we all approach things from a level of bias. We all have our biases. It depends on where we grew up. It depends on how, you know, religion we participated in. It depends on all these kind of social cultural context and clues and a whole bunch of this stuff. It's a constant feed in. So getting away from bias is very, very difficult if not impossible. So setting the table for saying and your book recommendations are absolutely phenomenal. Right? It's setting the bar and saying, I understand that I am approaching everything with my worldview and with my bias. Being able to admit that this exists in the first place is hugely important. Again, not gonna get political. But when you remove the standards for the inimical bias, you end up with situations where you start to remove cornerstones of individual expression. You start to use terms like meritocracy over and above the idea of equity. Why? Because you're you're judging things based on your own fixed bias of a system that may or may not be broken. It's whitewashing. It's a westernized view. It's a view that every single culture has regardless of color, skin, gender, creed, race, religion, whatever. Everybody has their inimical biases on this thing. So when we approach it from a technical standpoint, we have to understand we're all approaching things with bias. We're all approaching things from the nature of either our company is dictating a certain outcome, which will lead to a biased view of underpinnings of this. I mean, I'm a data platform in my day job. I should not care. Right? But I have a bias towards my supply chain. I have a bias towards my customer set. I have a bias in the way that I approach the world. Is it good? Is it evil? Don't even know Because it's it it I but I am willing to admit that this bias exists. And so I'm gonna function in the world in a certain way, in a certain perspective that's gonna lend itself to that. So when it comes to co development, you know, it's finding desperate viewpoints. It's Fable, you know, a really awesome organization that's out of Canada that does diversity testing, does testing of you can bring your product to Fable. Fable will test it against a diverse background of individuals. It's the work my wife does, for example, in assistive technology. It's understanding the application of your technology, of your datasets, of your ideas, your concepts, your processes against a population that is dissimilar to find that inflection point, to find that delta between what you believe, your ontological or your teleological beliefs, and what the outputs of that actually will occur. Not everybody can afford to do that in real time, fast, quickly to Peter's point. Like, we're this rapid cycle of evolution. But if you don't include these viewpoints, if you don't look at OECD, if you don't look at UNESCO, if you don't look at some of these pragmatics from industry or consortia that apply here, you're gonna always constantly be held to a standard that exists, you know, out in of your own devising, of your own making that doesn't have any you know, can it's not as solid as it could be from a perspective of where it's grounded in this idea of diversity of view or intersectional views, if you will. So, anyway, that's my sweet box. Dave, good points. I'll add something to this. So I think, especially in tech, what we're seeing is that there's a small majority of companies that have a large portion of the market and market share. Right? So we we think about big tech, but if you're into photography, I'll give you Adobe. Right? The number one tool that any photographer uses. You know. Right? Okay. So I use a small data company called Capture One. Just definitely. Yeah. Yeah. Yeah. They're there too. They have a spot. You're right. But I think that's that's a challenge in itself for technology as well. We need those diversification of tools to get diverse perspective in how technology is implemented as well. And I'm not sure if bias is a bad thing. I mean, people come with bias. It's a given. So from a cultural perspective, I mean, there's there's a million dynamics to this one. Right? But that is what makes us humans beautiful. Right? We're not the same. We don't look alike. We have different thought patterns. We have different histories. We have different values and and and norms, and that's okay. And I think somewhere somewhere along the line, that's where technology also stifles us because it tries to put us down one stovepipe. Right? This is what the technology predicts you should do. That's okay. That's not okay. And the borderlines, there's a lot of gray there. I mean, you know, there's black and white, but there's a lot of gray there as well. And I think that's a challenge for us as an industry. The other thing I was thinking of is a book since everybody was referring to books. I'm like, I need to come up with a book, but there's actually one that that that is top of mind for me, System Error. Don't know if you've ever read it. I think it's a fantastic book. Talks about how big tech implements platforms and where, they could do better. And it goes to, developers. It has something to do with optimization. That's a read I highly recommend and love to talk to anybody about it when you finish reading the book. Give me a call. Alright. Fantastic. Yes, Pete. Peter. So, bias is a tremendously complex topic, and and very challenging. I do wanna make one meta point around, AI safety and responsibility, which is, because that's the general conversation we're having is the challenge with biases. There's a lot of different perspectives and, you know, an active conversation going on in this area. That makes it very hard to to chart a direction. But there are a lot of areas in building safe, reliable systems that we all agree on. We don't want them helping someone, you know, hurt themselves, hurt someone else, commit a crime. And they are helpful systems. Like, this is why these are powerful systems is because they can help people do things that would be more challenging without their assistance. But there, you know, there are these things. There's, you know, harm, crime, you know, exploited material involving children. Like, we just we all around the globe, from different, you know, political perspectives, we don't want them doing these things. And, you know, I think we do have an opportunity to come together, you know, define good standards, define good metrics, and and, you know, prevent, a lot of of bad stuff. And and there's tremendous good stuff to be had here, in terms of improving medical treatment or education or just human productivity so we can all live a better lifestyle. So it's it's easy to get caught up in the difficult topics, but, you know, it's important to remember what we do all share and agree to. Fantastic. Fantastic. Fantastic. So we are wrapping up here. I did have one more question, but I'm gonna ask you all to actually share your responses on the Cosmos community. So I'll share the link to those responses as well. The the question that I had, as we're talking about all of all of this, it reminds me of the Jurassic Park Jeff Goldblum meme of where your scientists were so preoccupied with whether or not they could, they didn't stop to think if they should. And I've I've heard that variations of that, throughout the, the panel here today. So my question to you all is what are some good, organizations or companies or projects that are actually driving, AI forward in a responsible manner? Thing people and organizations that we can look up to that are doing the right thing today, so we can follow in their footsteps and continue to evolve this. So that's my question to you all, and I would love for you to respond in the the Cosmos community there. But as we wrap up here, I would like to thank you all for participating in this panel. Thank you so much, Allison, for, pulling this together. This has been fantastic. Thank you so much, Florence, Peter, Maddie, and Dave. This has been amazing. We got twenty seconds left. Are there any closing words that you wanna finish off with? Great discussion. We hope to do more, and let's do one on gaming next time because I see also as the games behind you, and that's that's cool. I think what this said is that there needs to be at least a part two. There's so much to unpack with this, Jonas. Thank you for giving us the opportunity on Cosmos. Thank you. Thanks, everyone.