AI and Accessibility

Speaker A: All right. Hello and welcome to Artificial Intelligence Insights, a podcast by S and P where we will have uh, the great fortune of discussing AI and accessibility. Um, we've got a great panel of guests and we don't want to waste any time so let's get started. Um, perhaps let's move um, from east to west. So Miriam, I think you are the furthest. So do you want to introduce yourself?

Speaker B: Yes, Miriam Fernandez. I'm also joining from SMP Global, co host and I work with SUDIP in the analytic and innovation team covering artificial intelligence research and my background is on the financial services industry.

Speaker A: Thank you Miriam and Chris, I believe you are slightly west.

Speaker C: Hi, Chris Holmes from the UK House of Lords, uh, with special interest in all things technology, AI, dlt, cyber digital assets. With the fine golden thread of inclusion and innovation running right through all of that.

Speaker A: Perfect, thank you Chris, welcome. And Tricia, I think you're slightly east of me. So

Speaker D: awesome. Uh, hi, I'm Tricia Maya. Uh, I lead product at ted, like TED Talks. My background is in product management, design, data and analytics for consumer and SaaS products. And yeah, I'm really excited for this conversation. I'm very, uh, interested in exploring how AI could be used in service of something bigger for mission driven products and use cases and really creating space for everyone to reap the benefits of this amazing technology.

Speaker A: Thank you Tricia. Um, I'll go next. I'm um, Sudeep Keshe, Chief Innovation Officer for S and P Global Ratings. Um, I share a similar passion where I think that um, uh, I can really be a force multiplier in helping us really uh, kind of reach our sort of greatest human potential. Which you know, to me that sort of, that performance indicator includes bringing society along and being able to really manifest all of the good, um, that we can do and be able to kind of augment ourselves and our abilities while you know, mitigating some of the risks and things like that and not, not having sort of adverse consequences. It's going to be a challenge, it's going to be uh, a journey. But I really, really believe with strong conviction that together, um, when we're working together, thinking together and having conversations like this, that we're really going to be forward. And last and not least, the brainiac Diogo. Do you mind introducing yourself?

Speaker E: Yes. So I'm Diogo di Lucena, I'm talking to you guys here from la. So thanks for having me. It's really great to be part of this conversation. Um, my background is on rehabilitation and robotics for people after stroke and spinal cord injuries. So helping people gain and regain agency after a neurological injury. Uh, right now I am a chief scientist here at AU Studio. Um, at au, we have a really interesting model where we do work software and data science consultants by day, but by night we do research that is impactful for the world. Uh, that was on. That has been on brain computer interface for a long time, working with a lot of the labs in the US that are building, uh, implanted bci. Uh, and with AI kind of really, uh, taking off, we moved on to doing a lot of research on AI safety as well.

Speaker A: Awesome. Thank you, Diogo. So with that, uh, let's start conversing. So we're going to have a couple questions just for all of you as panelists, but by all means, I think that all of your experiences are extremely rich. And by all means, if you want to, um, jump in and contribute a comment or a suggestion or anything like that by, like, let's please do so. Um, so, Tricia, let's start with you. Um, so last summer is when we met. We were both presenting, um, you know, keynotes at the Imagine AI conference. And one of the things. And I was really just grateful to just be in the front row and be able to kind of watch your presentation because all of my neurons were really firing once you showed some of the, uh, advancements that you guys at TED were making in terms of accessibility, um, not just for, um, um, um, in terms of, uh, language. So I think language was one of the things that you guys had showed, but also having things that were compliant with lip readers, uh, for people that are hard of hearing and so on, and also being able to have those channels where TED is actually connected to a lot of, um, jurisdictions where other, uh, modes of communication may not be. So, like, all of that just really said, like, this is just a quintessential example of technology kind of pushing the boundaries of really helping, you know, smarten society. Do you mind just telling us, um, you know, can you walk us through some of those initiatives that you, um, guys were demonstrating and also, like, what's some of the spirit behind it? Because I really, I just really appreciate the backstory.

Speaker C: Trisha.

Speaker D: Yeah. Awesome. Thank you. Um, yeah, so, Ted, if folks don't know, our mission is really to spread great ideas to everyone everywhere. We are a global organization. We have had, uh, we've really invested in our. Specifically in our language and global abilities over the past 40 years or so. Um, we have an amazing volunteer translator community. And so we have just a lot of goodwill around the world to make ideas accessible in different formats. Until recently that's really been around the written form. So like I mentioned transcripts and subtitles. But more recently sudeep, as you mentioned, we launched uh, what we've been calling our AI dubbing initiative. So we basically will take the speaker, uh, the speaker's tone, their voice, uh, and translate that into multiple languages. So it's as if the speaker is speaking French or German, even if their talk was in English. And one of the elements, uh, in addition to the voice cloning is the lipstick. Like you mentioned sudeep and we've paid a lot of attention uh, because we have this amazing volunteer translator community. We're not only feeding the AI engine human quality transcripts, it's not auto generated, we're feeding them human quality transcripts. And then those translators are reviewing and quality controlling the end output. So they're saying this looks sounds natural in my language. This is as realistic as it can be. So we do spend a lot of time on the, the AI and the hi the human intelligence and making it something really special. Um, and actually even before we launch this, we have integrated AI into our process to do the first pass of translations and then the human translators will do the final touch. So we've really been thoughtful over the years. Um, and so it is really amazing. You watch a speaker like our head of ted, Chris Anderson, he's giving a speech who's a native English speaker. He's delivering his talk fluently in Italian or German and his cadence and his tone is preserved. And so it's not just about the translation, it's about making the content experience as natural and authentic as possible at scale for people with so many different language and learning needs. Um, you know, sometimes people don't want, you don't want to read a TED Talk. Like TED talks are about the passion, they're about the emotion, they're about conveying that. And so you want to sit back and have that immersive experience. And now it is possible and in a way that again we spend a lot of time on the quality and of course the permissions. We will never do this with someone's voice and likeness unless we have explicit consent and approval from the speakers. And so we've worked this into our operations as well. Um, we're building it into kind of our legal agreements from the get go. So like how do we do this at scale going forward in a way that's really again authentic. Um, and, and, and focused on, on quality.

Speaker A: Amazing. Trisha it brings me back, I mean, when I, when I had saw the, the, uh, the presentation initially, it brought me back to my university days because I went to rit, which also houses the National Technical Institute for the Deaf. So most, if not all of my, uh, college, um, classes, the coursework, we had a translator in the front. And I'm just kind of thinking that, what a game changer, especially as lots of universities are expanding their international footprint, um, that you could actually kind of use technology like this to really help, um, you know, galvanize education. And like, you know, you noted you're preserving the soul. You're actually using technology to preserve the soul with the, uh, tonality, the intonation, all of these other kinds of things in that speech, which I think it's like we're beginning to understand how important that is. You know, even just kind of from, from standards or verbal communication. So it's just like that exercise is recursive by nature. It's just, it's, it's helping us understand, uh, what matters more. But it's also, you know, helping bridge the gap. Excellent. Um, Chris, let's turn over to you, um, because you've been at this a while, I think you'd be mad at me if I told everybody how long, but let's say it's been a little while, um, that you've been both a member of the House of Lords, uh, in the UK and you've been really helping address a lot of things that are really passionate to me, including innovation, inclusion and then crafting regulation, championing issues in terms of financial inclusion, things like that. So, so quite a lot. And if that weren't enough, um, and for those who don't know, uh, Chris is a bit of a legend, uh, in the world of sports. Um, so Chris, can you give us just a little bit of a sampling of your story, your origin story, if you will, and how you can see and how you've literally helped, uh, make technology sort of a force multiplier, uh, for helping, sort of accessibility issues.

Speaker C: I think I've been, thank you. Very kind, uh, introduction. I've been incredibly fortunate from the outset, really to understand the critical power, uh, and extraordinary force of inclusion even before I would have even known that word probably. So to have, ah, grown up and had sport and education as such, key forces in my life. I've been fortunate to go to four Paralympic Games and lead the GB swim team as captain for five years and come back with a few lumps of tin from various games. Very, very fortunate to have those experiences and to take that then into a corporate law career. I would have stayed in the law had the bid for the London 2012 Olympic and Paralympic Games not come along. And I thought I have to get involved with that and to thread inclusion innovation through everything that we are trying to do with those Games. In the summer of 2012, as a result of the work that I did on that, I think I was asked to join the UK Parliament in the House of Lords and carry on with a load of those, those themes, those threads, those enabling, empowering, unleashing of talent, ideas, policies, legislation. And it's what has taken me to seek to legislate in this area of AI. And largely because of that sense of, as people have already rightly said, the extraordinary human potential that there is to be unleashed through these technologies, it always has to be seen through the human lens, human led, human in a loop, human values, human beliefs, right through all of it. And there's an extraordinary false dichotomy which keeps recurring with some tedium across democracies and nations around the world, that you can either have innovation or you can have regulation, but you can't have both. But everything that I have experienced and everything that I've seen suggests to me that right size regulation, good for innovation, good for investment, good for consumer, good for citizen, good for creative. Now we all know bad regulation, there's a fair bit of that out there, but that's bad regulation. That in no sense means that regulation of itself is bad. And when we get it right, what you really see through right size regulation is that sense of the human touch, our human values running through the very words in those regulations which enable, which empower, which unleash. And again, right back to the outset, when we get this stuff right, I think we should see gleaming through everything those golden threads of inclusion and innovation.

Speaker A: Excellent, Chris. Yeah, one analogy I think that really comes across to me, and granted some of this is because I'm a musician, right? But it's just sometimes the limits breed the innovation. So for example, you have 12 notes in a scale and because you only have 12 notes, you can actually have some of this limitless amount of amalgamations of these notes or patterns and all of these other kinds of things to form all of this beautiful melodies and things like that. And it just, I agree that it's just like there's not a continuum with innovation on one stop and regulation on the other, but actually one, one feeds the other in bidirectionally, but. But I think that's one of those things that we do have to be really mindful between not only just good and bad regulation, but good and bad innovations. Right. Like where, where it's. If the, the innovation is misappropriated that that can lead to danger. And that's something to think about and something to you know, risk manage and balance and have a conversation about and be inclusive about, you know, all of the people that would sort of be impacted by you know, that decision. Being able to have some level of representation, to be able to form could be a good innovation and a good regulation, allow those to work intent. So let me um, I'll just take a pause and then Miriam, um, why don't you host a little bit and let's go to Diogo.

Speaker B: Uh, yes, thank you. And I couldn't agree more with uh, what Chris and Sudeep, you were saying about good innovation leading to good regulation, uh, and those being combined because that when you uh, when you relate it to the work that yoga you're doing, it builds on trust, on trust for the patient and for the, and for the user. And uh, I know Diego that you've been uh, in your, in your research both at your current company and before, uh you've been covering uh, quite a lot on AI safety, uh, but also on smart prosthetics and as you mentioned earlier on brain to computer interfaces. And when we think about this type of technologies, assistive technologies like the bcis, uh it's true that the commercial uses are still maybe in early days but uh, the fact that this is happening at all and that people for example uh, that have a paralysis can communicate with computers using their thoughts, uh, uh, it's quite surprising and breathtaking for and it will be something new for some of our listeners. So could you share some examples of your work um, and also your thinking for how these technologies can uh, help drive opportunities particularly looking ahead?

Speaker E: Yeah, absolutely. I think ah, BCI is uh, really moving forward quite quickly. Uh and it's been a really exciting time for it. It was maybe 12 or 13 years ago that I saw a video from Brown University where a woman that paraplegic for um, 15 years or so for the first time, ah, since then she managed to sip uh, a glass of water just using her thoughts, um, doing that completely independently. And that was really moving to me and it was kind of what brought me into the space. Uh, luckily after um, doing my PhD and a postdoc, I got to work with that same group and they continue to do really, really amazing work in this space.

Speaker D: Uh,

Speaker E: we developed some at Harvard we were developing ah, soft robotic systems for people after stroke, um, and spinal cord injury. And you can start really kind of pushing that forward. We were doing only a glove for example, but now the system is kind of doing the full arm. And you can imagine as you connect that uh, to a brain computer interface that someone that didn't have the ability, uh, to move their arm and actually do things independently with this kind of assistive device, they can really regain independency and agency. Uh, there's been a lot in this space that's been really exciting. Uh, we worked for a while on Imagine, uh, handwriting. This is a work coming uh, from Stanford where people uh, with spinal cord injury had one of these chips implanted into their brain and, and they were just imagining handwriting things in the air. Uh, and we can decode the neurosignals that come from that, uh, imagination and actually translate that into text. Uh, so people are able to actually type and it was at about the speed of someone typing on their cell phone. So that's really an amazing accomplishment. Uh, and we have reproduced that into other studies as well. And then now we've been having more things on speech neuroprosthesis as well. So people imagining they're speaking and we can pick up the signals that they would be sending to their muscles to make that speech movement. And we can get things to be really almost at the level of regaining full speech. And you can imagine that being used, uh, for someone that is losing their ability to speak. For example with uh, als and that being like I completely game changer in terms of how they interact with the world. Um, and these are using like the current vci, um, that is in the market that they are actually uh, FDA approved and uh, can be used for commercial uses. Of course there is the fact that you need to have a brain surgery to get them working. Uh, so there is a big bar there to cross. But uh, there is something that is already impactful in two people's lives. Uh, and we're moving to things that are a little bit less invasive as well and that have really high capabilities. Um, we've been really excited about the work from Forest Neurotech. Uh, they're a, uh, focused research organization. So a new funding structure that also helps push this kind of technology forward very quickly. Uh, where they do this, uh, basically full brain, uh, computer interface where they can read the entire brain at once with just a few devices that are implanted onto the skull so they don't go as deep as the other ones. And so it's a little bit, uh, it's more safe and it's easier to do. Uh, but they can also write back to the brain so they can send uh, ultrasound waves that change the brain activity. So you can imagine that being used for someone, uh, with depression, for example, that you're measuring uh, specific brain signals, uh, that you can notice when uh, they're getting into a specific type of mental state and you can help uh, drive away from that. So there's a ton of applications and really uh, the field is moving forward a lot quicker now with AI, which then also brings some of the concerns that we have with AI. One of the uh, participants in one of these studies, uh, it was set up something for auto completion to help them really type faster because the systems are so somewhat slow, uh, and AI was helping them to finish their sentences. But then at the end of the study, uh, they reported that it didn't really feel like the language and the way that they were speaking was their own. Uh, so that is one of the issues with implementing these systems. Uh, and as we get better systems, better AI that is more personalized, we can get over some of those issues. But uh, yeah, even in these early days they are already really uh, helpful. Um, and I think as we get to really large AI that is, um, even maybe smarter than humans are, we get into much bigger challenges. And that's what we've been working on on the AI safety side. So how can you actually build systems that can model the users well, that can represent the users well in their internal world model and that they can with that really align on what they need to do and what the user wants to do. Uh, we've seen with these models that they are capable even at the current stage to um, deceive or to uh, do things that would not be aligned with humans when they have a specific goal that they're trying to achieve. Uh, and of course today it's a little bit, it's still somewhat easy to capture those, but as we move to larger, smarter systems that, that will become more and more of a challenge. And that's, that's what we've been mostly researching, working on now though Diogo, I

Speaker A: imagine that there's, there's quite a lot of work in terms of the, when you start finding these issues, right, as well as some of the solutions and things like that. The first thing is that you're, you're embedding some telemetry onto when any of these different sort of uh, you know, thoughts are present or these electrical impulses are present and all of these other kinds of. And then once you have sort of that map, it's almost like being like a child, where it's just kind of like that, that environmental interaction becomes the information by which you can then actuate and then potentially actually like, come up with some creative ways to mitigate some of the risks and so on. Because at least now you're, you're cognizant about them, where I think even with, you know, folks who aren't having their brains measured. Right. Like, a lot of these things are happening, like in the background silently, and you would never really know. Is that, is that advancing some things as well in terms of the conversation?

Speaker E: Yeah, that's an interesting, uh, perspective to it because we have seen studies coming out of Google and some other places as well where they can map, uh, this, the reasoning that's happening or the activity that's happening. In AI models and in humans, you can try to create, uh, methodologies where you align those better. So that would be one side of that. Uh, and for folks that were not measuring brain activities specifically, uh, we've been doing work in trying to replicate, um, some of the features that we see in human brains, like empathy and how that works, uh, into AI systems and try to get that to be emergent in AI through their architecture and training.

Speaker B: It's so fascinating, the fact that in order to have AI and humans more aligned, you can do that by measuring the brain signals. It's really fascinating. I think that, Diego, you've started to touch, uh, on plenty of opportunities, uh, for humans, uh, with some sort of disabilities and impairment. But, uh, we would also like to talk about accessibility from a different angle, from the economic angle. Right. Because when you don't have access to the Internet, or for example, you don't have access to the AI models or you cannot afford it, or you don't have enough, uh, literacy, digital literacy, or you cannot speak a language that also, uh, prevents you from accessing AI and, uh, reaping the benefits from AI. So if I turn now to Trisha, uh, I would like to get your views on how do you think, um, AI will impact developing economies and economies and people with less accessibility to AI and what opportunities, ah, do you see for the future?

Speaker D: Yeah, it's a great question. Um, I mentioned my background was in product development, but before I got into product, I actually, right out of university, um, I focused on economic development. I thought I wanted to go into international development. I was like, I was about to be a foreign service officer with the Department of State. I was on a very different career path and Diogo, I didn't know you were from Brazil. And I actually spent some time in Brazil working at a microfinance bank in the Northeast. It was called Banco Palmas. There was a community of regional banks. And what was so interesting, and is actually drawing some parallels to your question, Miriam, is um, the whole focus was to give economic opportunity to people who couldn't get loans, they couldn't open accounts with traditional banks. So there was really a focus on how can communities provide the collateral and um, sort of like attestation that like they know this person, like they, you know, how can we build our own credit? How can we build our own systems of access so people can open a storefront or get a loan for their house or for their children's education. And so there are so many different facets to that. So the access, but it's also the uh, financial literacy and teaching people about loans and banking that maybe they're not getting in their schools because they're in an under resourced community. So um, I think there are a lot of parallels to AI and technology. And that's why I think this notion of like a shared infrastructure for global access, like how do we make sure that we have open models, there are lightweight tools that can be accessed on multiple different types of bandwidth. Whether you're in a low bandwidth or a high bandwidth environment, um, you know, cost, like access, like a lot of the tools that are out there are, they're cost prohibitive. And so I think it's really important to think about how too much of AI innovation is really gated by compute or capital. And so it's like how do we make those, how do we decrease those barriers, eliminate those barriers? Because I can't speak for you all, but just even in my, in the product community and tech, I see a lot of people learning about tools just by using them. It's like brute force experimentation, just trying things, using it every day, making it part of your workflows and your habit. And maybe it's a little uncomfortable at first, or you're not sure, but that's the way to learn to, to even understand what's possible. And so if people don't even have the basic tools or they don't know how to access them, or they can't afford them, then we're cutting off like a whole swath of people from the new global economy. So just kind of further, you know, increasing that, that divide. So um, I think that's why it's so important to keep that in mind. Um, and think about systems similar to the, you know, the economic development angle. What are the systems we can create for people in our and other communities that don't have as natural of an advantage?

Speaker B: Thank you. Yeah, I totally agree. We are doing also some research on AI and accessibility and also exploring that angle and exploring some data on uh, what sort of factors and variables influence uh, accessibility. Like on what you touch, of course, uh, on the Internet access, on the, on the access to the models data, but also on the, on the literacy and the knowledge gathering that will allow you to continue improving as we use the technology, uh, more and more.

Speaker D: Um, just maybe just one more note, like I've been super impressed by the no code or low code development platforms like Replit. I'm like a huge fan. I'm not a coder, I do not come from a technical background, but just the ability to create new career paths for people just like who literally you don't have to have a computer science degree. I know we've, you know, we've said that for a while it's been sort of a trope but now it's a reality. You can create apps and sites and all of these great experiences with just a willingness to experiment and learn as you go. Um, and so yeah, I think there will be really huge gains we can make across just career development, um, technological literacy. Uh, but it's getting, it's letting people know that these things are out there. Like if you don't know where they are, how to access them, it's like, it's like speaking into a void. So um, anyway, yes, very, very important points.

Speaker A: Yeah, similarly on that point. And then, and then we'll go to Chris as well. But um, the um, I gave a talk at um, it was a pretty large forum and most of the questions I was getting were about um, you know, older adults that were asking about how do I get involved with artificial intelligence. And when looking at a lot of the coursework, there's not a lot. Right. And it's just. So I was thinking that like how do we make that more sort of accessible to you know, an older population and things like that. And I think it's like you said Trisha, where it's just kind of like if we can create some, you know, tangible use cases. So for example, can you make me a bottle that can um, I can put my zip code in and then it'll my postal code in and it'll find the nearest grocery store, find out all the items that are on sale, create a menu for four people or what have you. Maybe I'm an empty nester. Instead of four, it's two. It can modulate that for you. Um, so it's just like you can, you can start to create some training programs to say, okay, well this is how to use some of these tools to be able to harness the power of that. Or for example, like if I was an older adult that's been working at the utility company for 40 years and then it's just kind of. I always had a passion for wine and I'd like to do that in retirement, but I've never worked at a wine shop. You could ask for some course suggestions of like how can I make myself more marketable for a part time job at a wine shop. So that way I can enjoy my retirement the way that I'd like to, even though I've been working at the utility company. And AI can help with that as well. Um, and I think that's one of these, these brilliant sort of um, force multipliers of being able to leverage the technology to actually include people that have been sort of uh, digital on the other side of the digital divide. Um, I'm yapping. So Miriam, let me come back to you for some questions.

Speaker B: No problem. Yeah. So please turning to you and now looking at it from the policy making or government side of things, um, we know that with your work, I mean you introduced um, the UK AI Bill a couple of years ago. Uh, you're very cognizant of the importance of ethical principles such as transparency, accountability, interoperability. But we also know that AI and the data that relies on it, uh, can be biased or it can maybe fail to represent certain types of uh, population, groups of population can misrepresent them. This could lead to discrimination. Uh, and uh, this is one of the, of the risks that um, we are very cognizant of AI. So uh, I was wondering, how do you think about all of this paradigm in terms of your work, uh, to ensure that the regulation incorporates some of these risk mitigants by design, to be inclusive and to be equitable. By design, Very much so.

Speaker C: And really echoing and agreeing full throatedly with what everybody said on this point. Because no matter how powerful the technologies are, no matter how much potential they have, that's potential. It's not an inevitability. And there's just as much chance that these technologies could exacerbate existing patterns of digital exclusion and financial exclusion, which all too often walk hand in hand. It's already been mentioned, rightly by Trish, this sense of the Global South. There's a lot of talk understandably about biases within AI systems, within the data that's been ingested. And that's certainly a key point. But what's perhaps talked about less is the stuff that isn't there. How much in these large language models, for example, how much data are they even able to take from many, um, of the communities in the Global South? Not only is there bias, there's an increasing, if not complete in some aspects, lack of representation in the system. So how can those systems work for the benefit of those communities? For those countries to have an approach? I think we should always go back to the principles and ensure that what we're doing is principles based, outcomes focused and inputs understood. There's a lot of chat around the black box can't possibly know what's going on in there. Well, you can to a large, if not potentially ultimately complete sense. But it's whether organizations choose for that to be the case and structure it to be the case and whether we as legislators, as policymakers, even more important as the public determine that's how we want it to be. Because ultimately AI is nothing without data. And when we say data, that's our data. So if it's our data, it must be our decisions. And if we can have that human led approach, public engagement, absolutely critical to that. We've said that the most important clause in my AI bill is the clause around public engagement. Because without that public engagement doesn't matter how good the technologies are, people aren't going to avail themselves of the opportunities and largely then they won't get the benefits, the upside, but almost certainly, uh, they'll be saddled with the potential downsides and exclusion. So our data needs to be our decisions and then we have a really positive chance of enabling our human led digital futures. That's I think how we need to conceive it.

Speaker B: Thank you. Um, and I mean to add to your point, um, and around governance, I think apart from what you said, we also need some global standards that are unified across jurisdictions because right now we have some sort of spaghetti ball where there's not enough unified governance, um, standards for that to really represent society,

Speaker C: um,

Speaker B: and be able to take advantage of it. But let me pass it to Deep. Uh, I know, uh, you have some further questions.

Speaker A: Yeah, let's try to land the plane actually, just because I, uh, know we're coming up against time and I think that that was kind of a beautiful segue to tying Everything together. Because I think, like, you know, Trisha started. Started out with this, this thing of talking about when you're doing a lot of translation and things like that, you have to take the words and what is said and recognize that those are different things. So like the intent, the spirit behind the words, the tonality, the intuition, like all of the things that we would sort of consider the audio part perhaps of nonverbal communication very much part of the message. Right. It's an incomplete message without that. And then I think, you know, with um, you know, Diogo's remarks is when, um, when you embed a lot of these things together, like the brain is not just electrical signals. There's a lot of kind of magic that sort of happens happen. And by, by going through a lot of these things with BCI controls and so on, we're able to have better cognizance as to what's going on and be able to make better decisions. But we also need to, to then put ourselves in that perspective, you know, connecting with what Chris just said about what's prudent, what's the right thing to do. It's not just what's possible, but what is the right thing to do, um, you know, given these, these particular circumstances. And I think that's. To me, it's actually like, it's a great time to be alive because it's just like we, we get to be at the forefront of really developing entirely new sciences that like, in terms of like even, you know, my, My friend Mike Carroll, he turned me onto this book called the Book of why, um, which goes into this, this notion of causal inference, um, which kind of gets at sort of the base basis of morality and ethics and all of these other things and understanding it sort of from a scientific perspective. Um, and I think a lot of the things that you, you guys were talking about is exactly that. It's, it's, it's really sort of connecting the dots and recognizing that we know what we know and we have some level of management of what we think that we don't know, but there's just this wide corpus of things that we really just don't know. And now we're starting to allow the artificial intelligence to help us understand and then make decisions and be accountable for those decisions that actually kind of improve society. So the exercise, let's do a little round robin is, um. My friend Brian Evergreen calls this future solving. So he says, uh, his notion is that problem solving is inherently backwards. Right? You're only eliminating what you don't want by eliminating some of these problems. Future solving. He says, let me imagine a future and what needs to be true for me to have a role or a positive role in that. So, Tricia, if you don't mind if I start with you first, um, just what kind of future do you sort of imagine for us in terms of connecting on AI accessibility and all of these things in terms of risks and opportunities just for either for you personally or through Ted? What does that future look like to you? And can you just name like, one thing that needs to be true to actual.

Speaker D: Yeah, um, I feel like you're speaking the language of product. We're always like, what problem are we solving, what opportunities out there? Like, you have to crystallize it for people and make sure you're solving the right problem because there's a lot of problems to solve or there's a lot of opportunities out there. And so I think in my dream world, we would be solving all of these problems that Diego and Chris and everyone's talking about. Because I feel like a lot of the news cycle and hype is like either how companies are exploiting people through AI or how we regenerating more effective cat memes or something. There's just such a either superficiality or, uh, evil undercurrent to the conversation. But we need to tell more stories about how AI is serving humanity and not just disrupting it. So the things that we create will kind of manifest based on these examples, like we don't know what we don't know. There's an endless world of possibilities. So how do we start to talk about the really meaningful, deep, like, impactful use cases and give people really clear use cases like you said, sudeep of like, this is how you can use it, this is what you can do. So I think for me, what needs to be true to unlock all of the possibilities we've been talking about is really, really paying attention to the stories that we're telling. Um, and that really, uh, basically I think will choose what we elevate in terms of what gets built, who it serves, why it matters. And so I think just being really thoughtful about, um, what we're building and why is, you know, will need to happen to make that a reality and

Speaker A: being explicit about that. Like, even if we're, even if we're wrong, at least we're being transparent. And that may trigger somebody else's imagination.

Speaker B: And that links to what Chris was saying about choosing the content. The content is the data that then fits the model and you're Able to choose the stories, choose the content.

Speaker A: So on that Chris, uh, what is the future that you're imagining and what needs to be true to.

Speaker C: Actually Wittgenstein said the limits of my language are the limits of my world. I think we need to ensure that the limits of my large language model aren't um, the limits of my world. To not be trapped by the past and trying to solve the future, walking backwards into it, to consider a whole array of other possibilities and not Trisha, not to be trapped by the incessant news cycle which is predicated largely on fear, uh, and um, keeping people hooked in that sense. Because there's no question right now, geopolitically, geo, economically we're in very tricky times. So many of the issues facing us are ah, global in nature, are uh, existential in nature. A climate emergency, energy crisis, conflict, cost of living. And yet, and yet human led, human, imagined, human developed technologies have the solutions to these and so many of our issues. And if we get this right, if we get this wrong, this won't be a failure of the technologies, this will be a failure of us. The technologies are just tools, extraordinarily powerful, but tools in our human hands. And again we decide, we determine, we certainly always need to ensure that it's to use cases or to go even deeper, that it's purposed. Purposed. Ah, and through that you get the human connectivity. And if we get this right, and there's no reason why we shouldn't, it involves everybody to play their part, everyone to play a leadership role. But if we get this right, we could have the economy, the society, the social mobility, the health systems that we've always imagined could be possible, but we now could. And in many ways I'd say we now must bring that into being. And what a potentially positive, as you say, time to live. What a phenomenal mission to get you out of bed every morning,

Speaker D: Putting you

Speaker A: in the odd position of having to cop that. But um, let's let you land the plane, um, in terms of that future because I think this is like the bread and butter of a lot of both your research world as well as sort of the implementation. Tell us what's some of the feature that you're imagining and what needs to be true.

Speaker E: Yeah, I don't think I can top that. But uh, let me do my best here.

Speaker B: Uh,

Speaker E: we know that AI will be the uh, interface for basically every technology that we have moving forward. Uh, I think there is something that we call a slope blindness, exponential slope blindness that exists, uh, if we look, two years back, what has happened with AI to this day is really amazing. And it has really a big impact on what's happening to AI tomorrow and the next day. So things are going to be increasing, um, getting more powerful, uh, and smarter, uh, exponentially. And it's hard to predict that really well for us humans, uh, and we've seen with these models that it's very easy to get them to misalign and to do things that are not aligned with what the user is trying to do. Uh, so I think what we need to do, what needs to happen, is that, uh, there needs to be a lot more public and private investment into building safer AI. Like right now there are orders of magnitude more investments to just building more capability, uh, more, uh, features like that, but not so much on alignment. Uh, and we need to build alignment to be almost like an emerging feature within AI, so that when people are trying to use the systems and it is the thing that it's in between them and every technology that it's really doing what they want to do and what's good for them and for society. So I think, uh, that kind of investment is what needs to happen. And more research like the one that we're doing at ae, but a few other folks are too, uh, and hopefully a lot more in the near future will do to help increase that kind of accessibility in a more flourishing future for society.

Speaker A: Excellent, Diego. So with that, um, yeah, so, so first of all, so thank you all for, for your remarks. Um, because I think your experience, your thoughts and the way you've articulated them so clearly I think is extraordinarily helpful. And the, the next thing I'll thank you for is that I can imagine I'm coming away from this conversation very inspired to really just kind of take accountability for the seeds that I'm, I'm planting right in terms of some of the future and things like that. And um, it's one of those things that it's very much, you know, up to me and up to those who are inspired to kind of take those seeds and make sure that we're, we're trying to plant healthy trees that can then, you know, give us some, some, uh, fruit that we can harvest. That, that really helps, you know, society and helps sort of our mission as human beings, but that the accountability resides with us. And that comes from talks like this and being able to kind of imagine together be inspired by each other and be able to carry that mission forward. So from the bottom of my heart, really thank you for, uh, all of that and, um, be well. Um, excellent. So thank you.

Speaker C: Thank. You.