AI trailblazers putting people first

0:07Welcome to Practical AI. If you work with artificial intelligence, aspire to, or are curious how AI-related technologies are changing the world, this is the show for you. Thank you to our partners for helping us bring you practical AI each and every week. FASI.com, fly.io, and typesense.org.

0:39what's up friends ai continues to be integrated into every facet of our lives and that remains true because you can now index your database with ai you can write more code become that 10xer you always wanted to be and you can even draft a letter for a lease on an apartment or a new property AI is everywhere and it might be time for us to start questioning is AI our friend or our worst enemy and that's the focus of the three part season opener of the award winning podcast called Trace Route Podcast you can listen and follow the new season of Trace Route starting November 2nd on Apple, Spotify or wherever you get your podcasts and this show is all about the humanity and the hardware that shapes our digital world in every episode of Trace Route a team of technologists seeks to untangle the complex question who shapes the internet.

1:26Seasons one and two gave us a crucial understanding of the inner workings of technology while revealing the human element behind tech. And season three tackles not just AI questions, but also how can we use technology to preserve the earth? Who influences the technology that gets made? And what happened to the flying cars we were promised? I think it's safe to say that the future of AI is both exciting and terrifying. So it's interesting to hear the perspectives of experts in the field. Listen and follow this new season of TraceRoute starting November 2nd on Apple, Spotify, or wherever you get your podcasts.

2:22Welcome to another episode of Practical AI. This is Daniel Whitenack. I'm the founder at Prediction Guard and I'm joined as always by my co-host Chris Benson who is a tech strategist at Lockheed Martin. How are you doing Chris? Doing great Daniel. It's having a nice day here and having lots of interesting conversations about just all the things in AI. Episode last Last week really hit a nerve, I think. I think so, yeah. And I'm actually in London this week and giving a talk tomorrow about, quote, trustworthy AI, which I'm hoping that our guests can enlighten me on a few other aspects of that prior to my talk tomorrow, which will be convenient.

3:06So if I have to change my slides tonight, that'll be useful. You're cheating. Oh, boy. Listen to this. we're privileged to have back with us Solana Larson who is the editor of Mozilla's IRL podcast or online life is real life and Solana it's so great to have you back I'm so glad to be here hi I was just looking up the date and the episode number so we had you episode 187 back in July of 2022 talking about the podcast, the season that you had released all about AI. And we talked about concerning trends and how the technology was transformational, positive signs of change. And of course, it's just so interesting that it's only been up until now.

4:01So call it a year. Yeah. What a year. So much has changed. It almost seems like AI has been invented. I'm wondering if you could just give a little bit of context for why and how Mozilla is putting together the seasons of the IRL podcast and why the focus is on AI. And then we can go from there. Yeah, sure. Yeah. What a year. What a week, right? What a month. I feel that way all the time. It's just going so quickly. Yeah. I mean, I guess people who might be familiar with Mozilla might know Firefox and think, what does that have to do with AI and why are you talking about that? The Mozilla Foundation, the nonprofit arm of Mozilla for the past couple of years, has been really focused on what we call trustworthy AI, the term that you just used.

4:52And, you know, I guess you have a buffet of choices when you're trying to talk about ethical AI or fair AI or equitable AI. Trustworthy was the one that we went with as an organization. It checks a lot of the boxes, especially at the time that we chose it. and it resonates in policy circles in particular in some of the contexts that we like to be in. And so there's this element of what is the future of the internet? And if we're an organization that has a manifesto that cares all about making the internet healthier, creating tools that enable people to create and be a part of the internet, if AI is the future of that, then what's our role there and how can we help make sure that all the mistakes that happened for the internet in all these, you know, 20, 25, 30 years aren't just all being repeated again in AI.

5:49And so we're thinking about privacy, of course, but there's also consolidation of power, you know, the way that big tech kind of take control of everything and squeezes out opportunities for smaller players, and a whole bunch of other things, right? So we have a whole foundation of people, We have fellowships, we have grants that we make for different people who are innovating and trying to think about AI in a way that big AI tech companies aren't thinking about them. Just so that we can have an alternative kind of conversation happening around these things that are changing our society and our industry and everything so quickly.

6:28And so part of this is like, we do have a big mouthpiece, I guess, as Mozilla. And one of those platforms that we have is this podcast, IRL, that the Firefox team started years ago, and that we took over on Mozilla Insights team a couple of years ago. and we're now doing a season seven. It's the second season that we've dedicated entirely to AI. And part of that exercise is thinking about, well, who do we want to lend the microphone to? Who do we give the microphone to? What kind of voices would we like to have in this dialogue about AI that we don't maybe get to hear as often when we're tuning into the US mainstream tech press?

7:11Yeah, that's really great. I love in the announcement of the new season of the podcast, you talk about, like you just said, it too often feels like we've kind of lost control of the Internet to these larger players. And, you know, you want to speak to those kind of reclaiming power over AI to put people first. I'm wondering, as you were preparing for this season of the IRL podcast, you know, it has been a transformational year in AI. And, you know, we talked about some trends, both positive and concerning last year in terms of the IRL report and season. But I'm wondering if you could talk a little bit about how all the unique things that have happened recently, and especially around the kind of public discourse around AI and the public adoption of this technology, weaved into how you wanted to curate this season and particularly how the topics that you covered kind of bubbled up, which we'll get into here in a bit.

8:19Yeah, I think front of mind, a lot of people are curious now in a way that they weren't before. I mean, you must experience this on your podcast as well, that people now have this hunger to know about AI, where a couple of years ago, they were like, Oh, what's that? Or what? How does that concern me? Now everybody's like, this really concerns me, what should happen? And I think there are a bunch of areas where, like, nobody is entirely sure what to do. Like the first topic that we took on in episode one is around open source in large language models. You know, this whole question where you have on the one side, you've got folks who are saying it's got to be open or we can't audit the models.

9:01We don't know what's happening with the data. And then on the other, you've got the saying that it'll be the doom of all of us and everything's got to be shut down and closed for security purposes. And so you have these, I think, like a lot of discussion these days, like it's really polarized sometimes. And so it's trying to figure out how do you make a nuanced argument that explains not just different sides of the story, but just explaining how, you know, there's a spectrum. and there's a lot of AI topics that get sandwiched together just under this umbrella that's called AI. And it's just so many different contexts and so many different business purposes and so many different, like it almost less and less makes sense to talk about it all as one thing.

9:51We're right on the cusp where we're still talking about it as one thing and we're still trying to grapple with how we should regulate, how we should build, how we should design, what we should think about it personally. And so it's a really exciting moment to try and figure out those things. And I mean, the challenge as a podcast creator is that each of our episodes is like 20 minutes long. So we pack in three, four different voices. There's some really deep analysis. We work with our host, Bridget Todd, who's great. A whole bunch of people work together on this thing. And it's like this very highly polished, produced, lovely kind of white paper in audio almost of a big issue, like a big topic.

10:34So I'm really proud of it. And last season, you know, which was a little bit ahead of the curve in terms of talking about some of these AI issues, we actually we won the Webby for a best tech podcast. Congratulations. That's awesome. Congratulations. Yeah. Wow. I was surprised because, you know, we're the only tech podcast in the nominated group there that was hosted by black women that was featuring voices from Africa, from India, that were really kind of digging into the corners, I think, of thoughts around AI that aren't just concerned with how much money a technology is making. You know, that that isn't necessarily the criteria of success for, you know, why you would elevate somebody's voice.

11:17There's so much that you just said that I want to go into even more. You talk about the larger public's discourse over these topics, and topics with an S is crucial. It's very nuanced. And yet you kind of alluded to that sense of responsibility that you have in your own podcast about trying to, what voices do you want to raise and where do you want to get them? And I know that Daniel and I feel that weight very intensely. We're at a moment now where the whole world is really hopping onto to this topic. So you're bringing the people we had before, but there's so many new people that want to understand because they really do get that it's going to affect their life.

11:55And I guess selfishly, as we have tried to do that, trying to avoid just like people wanting to self-promote, there's always these efforts at that, but trying to bring the right conversation to bear. And how do you think about that? How do you and your team think about the fact that you're in such a responsible position in terms of being able to either have voice or lend that voice to others at such a crucial moment in time. We're at a unique point in history. How do you parse that? How do you help the public get that discourse right and talk about the right things? And how do you recognize, for instance, on this one, you know, this is too big of a topic to be thought of as one unit anymore.

12:37We now need to kind of segregate through the different concerns within it. How do you approach that? because it really affects how the public thinks about it. Yeah, I mean, I think we have an advantage in the sense that the organization as a whole is also thinking about this on a daily basis. So when we think about who do we give fellowships to or who do we give grants to or who do we partner with when we do different things as part of the thought process, and it's extremely difficult because it's almost like every single AI startup or project is something for good. Right. Like everybody says that they're doing it more ethical, making the world a better place.

13:19Everybody. Right. Like it's just but you need to figure out, I think, what are the values that guide you when you want to make a decision about what that means for you? And, you know, even in an organization like Mozilla, there's a lot of diversity in terms of like, what are people comfortable working with and what are their opinions on this? we chose the theme people over profit on this. And the idea wasn't to just only look at nonprofits. It was to also be able to look at, are there ways that you can profit, that you can make money and still kind of have a sense of putting people ahead, which is what we try and do with Firefox and with other Mozilla products, right?

14:01Trying to figure out how do you make money and not sell people out, not sell their data. And so there's that sort of critical lens on it. And what happens very easily is that you end up veering a lot to the people who are criticizing, right? They're being critical, they're criticizing, they're pointing out the flaws and the errors. And you can also get a little bit too much of that, I feel. And I think that's where we really put a lot of effort in is to figure out, you know, we listen to those voices very intensely, but who's being constructive? Who's kind of trying to rethink this from a different angle that we hadn't thought about before?

14:41Like one area in the second episode, we look at content moderation and we look at data work, right? The ghost workers and how exploitation of labor and data and how the content moderators in Kenya were being paid really terribly and not treated well and how they're fighting back. This story, right? But then as part of a package, you know, we also talked to an organization called Karya in India that's trying to rethink, okay, well, if we're doing data work, how could we renumerate people differently? And what they did is they have these voice data sets that they're making in different languages in rural India in particular.

15:22And they're working with a lot of women and they do it as part of like this educational project and have people, you know, be able to do work from home. And there's a whole philosophy around it. But every time they resell a voice data set to a new client who's building some kind of voice recognition tech, they send more money to the person who donated their voice to help train the system. And so what they're asking is, well, if we're paying pennies for this work in the industry and companies are making hundreds and thousands of dollars on people's labor, why don't we just give them a bigger cut?

16:02We can still have a really good business, but we could be thinking differently about what a contribution is and have kind of royalties that build over time. You know, as long as this data set has value, why don't we think differently about how we share that value across more people? and that's like a very simple thing where you hear about it and you're like oh yeah right we could just be thinking differently about this and there are a lot of examples like that in ai where you know karya is a business maybe they're not a unicorn but that's not their goal right that's not their ethos and so when we're challenged i think by people who are innovating in terms of in entrepreneurial ways as well.

16:46I think it really, it helps us see how we've been goaded into thinking about AI in ways that are really defined by an industry that has a specific set of norms and values, also around how data is used and how humans are treated. And we can rethink a lot of those things, you know, it could be different. And when we're talking about now in like the regulation space, how do we make things safer? and how do we stop harms from happening? All that stuff is really important, but we also need to have people who are working on figuring out, you know, where do we want to end up? What is the vision for what we want to accomplish with this tech?

17:28Because it's not going away. And yeah, so we need more examples, I think, of what is good, what could good be. That's our kind of guiding star for how we pick who we choose to feature in the episodes and how we kind of build up a story that has a bit of tension, but also like a silver lining. I'm already super excited to dig into these various topics. And I think people at this point, maybe they're wanting to just hop over and start binging the IRL podcast. And I totally give people permission to because it's just so good. So you can pause this podcast you're listening to and hop over and jump right into IRL.

18:10But I do want to take the time to talk through some of these subjects on this podcast. And because these are topics that have come up in various ways and bringing these other voices into that, I think would be really good. And the first episode, which you title with AI's Wide Open, I think is super interesting and relevant even this week with, you know, OpenAI releasing a whole new set of features, which are just incredible. I mean, like, I think anyone would have to admit like these, these features are incredible and like the vision stuff. And it seems like there are these sort of proprietary model providers or API providers that are really leading the way in some of this functionality.

18:59But there is this really amazing undercurrent of open models. And it's multifaceted, as you've alluded to, some people approaching it in various ways, like stability, releasing some models that are amazing performing models, but maybe licensed in a restricted way for research use and other purposes, others that are releasing under various licensing. So there's the licensing side of this. There's also, as you alluded to, hey, if we start opening up these models, can we really say to OpenAI, hey, just open everything up and everything will be okay? Are there reasons why we shouldn't be saying things like that?

19:42So this is a very multifaceted topic, and I'm curious if you could just kind of lay out how you thought about approaching this topic in particular and who spoke into this and what were kind of some of the highlights that were mentioned. What I was most concerned about in the beginning was not knowing where to end up, if that makes sense. Yeah. Because the jury's still out on a whole bunch of things related to this. I was worried about AI doomerism. I was a little bit worried about making it sound too scary. But then I was also worried about not making it sound scary enough. What's the right balance?

20:23Yeah. We called in David Evan Harris, who used to work on the AI responsibility team at Meta, who's probably a little bit more on the worry side than I am, at least, but also has a history in open source and was able to, I think, also speak in practical terms about what is and isn't being done at an organization like Meta and the reasons that he's concerned and how it connects to, for instance, like social media, which is something that everyday people are really concerned about, right? Like, how is this going to affect elections and that kind of thing? So that's something that we looked at. And then we talked to Abiba Birhane, who is a fascinating researcher, very vocal on Twitter as well, who is looking for openness in data sets and who's been doing a lot of auditing of data sets and trying to, you know, among, I think, a group of researchers who have really tried to engage with companies and try and encourage them to be more responsible about what they do and do not include in the data sets in regards to hate speech, terrible representations of women, black people, and so forth.

21:35And that's a really important, I think, tenant in the way that Mozilla thinks about why it's necessary for these models to be open, because the companies time and again have proved that we can't really entirely trust them. And so open source is like our safety net in a way. So I think on the spectrum, that's kind of, we're leaning more towards the open side than the closed side, even taking into account all the things that we know. We talked to Sasha Luccioni from Hugging Face, who does research around climate change and open source models, and just hearing just the simple perspective around or simple, but just this idea of if it's open, if we're working together in large communities, we might actually have these AI models have a smaller carbon footprint, you know, because they gobble up lots of energy.

22:32And so they're kind of concerns that are outside of this framework of dangerous, not dangerous, where we need to think about what would it mean to have researchers who speak different languages working on these things? What would it mean if you could have a more diverse set of people than the people who work at open AI working on some of these large language model capabilities. And then finally, we end up with an interview with Andrei Mulyar from a startup called Gnomec, and they've made the system called GPT for All, which you may have played with at some point. But, you know, it's like an alternative to chat GPT that you download onto your computer, and you can ask it questions and chat with it offline.

23:15and it doesn't take your data if you ask it not to you know it doesn't take your private information and use it to retrain models so it's like again like a different approach to thinking about oh well what if we did want this stuff to be offline what if we did want it to be totally compressed so you could have it on you know an everyday computer in a low bandwidth society you know what could we do then how would we be designing differently i find that inspiring but you know it's not clear cut right for any of these things nobody has the answer but the important thing is that we need to be like really contextual about how we think about these things a lot of the concerns that you just enumerated are certainly daniel and i share this it's been an interesting conversation on the safety side but then in the last week you know we had the bletchley announcement in the u.s we had the executive order which was quite detailed in terms of that.

24:12In the context of we agree with you that open source is a good basis on which to try to work on solutions that are in the daylight, so to speak, that everybody can be part of and see and verify. How do you see regulation that it looks like we're right on the cusp of after many years of talking about it, it's starting to happen. How do you think it will affect some of the ideas that you were just talking about, do you see us making some turns or do you think it's not going to have much impact? Because I'm still trying to process it myself. I think it has yet to be seen, you know, because if we're thinking about risk, for instance, like in the EU AI Act, if we're thinking about who is responsible for the risks and upstream and downstream risks and how would you regulate how that affects communities or how would that affect individuals or small organizations or nonprofits that were working on these things, I think we still don't entirely know.

25:10I think part of what we tried to communicate, and then I see Mozilla communicating in other contexts as well, is that, you know, it's not a matter of like, open isn't always just like good in its own right. It's like open for a purpose that is good. Yes, it's a start. It's a start, but it's also like you can be open in different ways. You can be open to consolidate the market in your favor and stamp out competition. That's not the kind of openness that we're in favor of. We're in favor of openness that leads to transparency and accountability. We're in favor of openness that enables collaboration between people who have good intentions.

25:52And so openness that allows people to build and create and do things for their own countries and languages and societies and stuff. So if we start thinking about how do we protect those functions of open, then it's not, you know, it's not just open for openness sake. It's open with a kind of, yeah, a purpose. Right. So I think it's important for the regulation to start thinking about those things. And, you know, consolidation of power is a big thing, like enabling free and open competition on some of these issues. I think it's really important. Having like a global lens on the effects of these technologies is really important.

26:34I mean, overall, it's good to see that things are happening. And certainly we agree with a lot of things that are being put forward. I think we just want more in certain areas. Yeah, on the data set front, you mentioned some of the necessity to have some sort of transparency about what went into a data set in order to proceed with kind of efforts that are reversing bias or trying to prevent hate speech or toxicity coming out of these models, which I think is really good. I also think I really highlight that kind of global piece of this. So one of our engineers at Prediction Guard, we were working with Lama too the other day and doing some experiments in a variety of languages.

27:26He's a Hindi speaker. I think he speaks a few other languages as well. And we were trying some things in Hindi and he's like, hey, I like this doesn't seem quite right. And then he went to the Lama paper and looked at kind of the distribution of language data through in the data set. And he was able to very quickly understand like, oh, this is why this is why this is this way. Maybe if we do this or that, you know, we could we could make some improvements in the tasks that we're trying to do. So even just I don't think it always has to be. I guess what I'm saying is maybe like every single thing needs to be open and in all the same ways.

28:03But even being transparent about the makeup of a data set and where it came from and provenance, that can be actually quite helpful and powerful. And maybe that ties into this kind of second topic as well, which is some of this data comes from these sort of data workers or crowd workers or ghost workers. And highlighting that as you label it as the human in the machine. Why was that an important thing for you to include as part of this discussion? Because it's invisible and it's overlooked at all parts of the food chain, right? From the consumers at their computers, but also the people actually developing AI.

28:44Like the systems that you interact with in order to hire thousands of task workers to help you do things with your data, like they're designed to shield you from actually having any kind of human connection or, you know, sense that you're dealing with the human. The whole thing is meant to feel like a machine. and unfortunately I think it also shows the the callousness of a lot of the industry because thousands of people are saying that they're traumatized and are suffering and can't eat and yet you know these practices continue you know as just like a cost of doing business and And you've got, you know, millions of people in countries like Kenya and the Philippines and like India, places that have a lot of tech graduates who are doing this kind of work, which is thoughtful and requires reading long policy documents.

29:42And people take a high degree of responsibility for the outputs a lot of times on some of these projects. And yet they're treated really terribly. And so it just seems like, well, does it have to be this way? No. it wouldn't have to be this way. And it's tied in with this, you know, like we can do better. Like it just seems like an area like we can do better on this. The fourth episode that we look at actually, I hope you weren't expecting to go through them chronologically, but we actually. No, no, you're all good. They're all interconnected. Yeah. Yeah. Well, because we returned to this question of like open, not open.

30:19The title of it is lend me your voice and it's actually about voice data sets and what it means to belong to like a small language community and you build a data set to be able to do voice recognition tools in your own language and there's a lot of open source ai that's really useful in that context for them to be able to build stuff in ways that are affordable right like sustainable also if you're trying to build a non-profit. And what some of them describe, we talked to this one person, Keoni Mahelona, he's based in New Zealand, he's actually Hawaiian, but he's working with the indigenous community in New Zealand for Maori language.