EP 148: Safer AI - Why we all need ethical AI tools we can trust

This is the Everyday AI show, the everyday podcast where we simplify AI and bring its power to your fingertips. Listen daily for practical advice to boost your career, business and everyday life.

Do you trust the AI tools that you use? Are they ethical? Are they safe? I don't even know if I ask myself those questions every time I jump in and I try new AI system or a new gen AI tool. But it's important to talk about having trustworthy AI and that's what today's show is all about. So I'm very excited to have someone from Mozilla join the Everyday AI Show. But before we do, as always, we're going to go over the AI news and and welcome. If this is your first time to everyday AI. Everyday AI is for you. This is helping everyday people like you and me learn and leverage AI with our daily Live Stream podcast free daily newsletter. So let's go ahead and jump into the AI news real quick and there's a lot, so bear with me here. But there's more drama at OpenAI over the weekend than a Shakespearean tragedy on fast forward. So the OpenAI board fired founder and CEO Sam Altman on Friday. By Saturday, there were rumors that OpenAI was trying to bring him back as there were dozens of high ranking OpenAI employees who were signaling they would leave and follow Altman to wherever he may go or whatever he may start. By Sunday, it appeared the two sides were in agreement and reports indicated that Altman would probably be coming back if the whole board was kind of replaced. Microsoft CEO Satya Nadella was reportedly leading or helping lead the discussions. And it was announced late last night, on Sunday night that former Twitch CEO Emmett Shear was announced as the interim CEO at OpenAI. And then Nadella also a couple hours later announced on Twitter that Sam Altman and Greg Brockman. Greg Brockman, who's the president and co founder at OpenAI and together with colleagues will be joining Microsoft to lead a new advanced AI research team. Obviously this is all way too much to unpack in one single day. So join us tomorrow. We'll be having a dedicated conversation about this to unpack it and talk a little bit about what it means. So entire show tomorrow dedicated to that other things we have going on in AI news. Meta maybe snuck this one in over the weekend, but the Facebook parent company disbanded its responsible AI team. Wow. Okay. And then also Germany, France and Italy came together to reach an agreement on future AI regulation. What a jam packed weekend of AI news. As always, we're going to have much, much more so make sure you go sign up for the free daily newsletter at your Everyday AI dot com. Check out the show notes if you're listening on the podcast. It's in there. Make sure to go sign up because my wow. My gosh, after all that, I need a breath. But more importantly, that's not what we're here to talk about. Although we're probably going to touch on it today because it's very relevant. So please help me bring on and welcome to the show. There we go. We got a Mark Sermon, who is the president of Mozilla Foundation. Mark, thank you so much for joining the Everyday AI show.

Thanks, Jordan. Really excited to be here. And what a weird day. I think maybe what happened is somebody just told chatgpt to write a Shakespearean drama about OpenAI and that's what we're watching unfold.

Yeah. And it was definitely fast forward. So maybe, maybe Mark, let's start super high level. We'll probably, probably hit rewind on that one. But just tell everyone a little bit about what you do at the Mozilla Foundation. I'm sure a lot of people know Mozilla, it's one of the most popular Internet browsers in the history of the world. But maybe tell everyone a little bit about the Mozilla foundation and maybe even a little bit about Mozilla AI.

Yeah. Thanks, Jordan. Well, Mozilla, as many people know, but actually many people don't because we've been around 25 years, we were started as an open source project in the public interest. The idea that people should have a browser, they control this open source. And that was at a time when Microsoft really dominated the web world. 98% of browsers, the whole technology stack of the web were really in Microsoft's hands. And so in many ways, Mozilla started as a counterbalance to corporate power and the idea that people should control the tech. And with Firefox 20 years ago, we were successful. We kind of took back a bunch of share from Microsoft, but also laid the path for things like Chrome and all of the websites, Gmail, Facebook, everything that emerged during that time which needed open technology to get built, they all are based on open web technology like JavaScript and, and HTML. And so that's what Mozilla stands for. Who we are today, we're at the spot of saying the same risk is there, that one or two or a few companies are going to control the technology that defines the next era, and that's AI. So over the last few years we've set up a bunch of fully public, purpose oriented activities, our own activism through the Mozilla foundation, things inside of our new Mozilla ventures, inside our main Firefox company, but in particular Mozilla AI, which exists to build open source AI that's trustworthy and in the public interest. And I can talk a little bit more in detail what that looks like, but that's something we're really proud of as a very tiny, but hopefully over time, a part of a bigger coalition of people trying to build an alternative to just AI in the hands of a few big companies.

And let's actually go there and then we're going to get back to Mozilla AI because I have a lot of questions and I'm sure our live audience does as well. And as a reminder, if you are joining us live, make sure to get your question in for Mark. What do you want to know about responsible and ethical and trustworthy AI? Because he's, he's one of the people out there leading this whole movement. So get that question in. But I do want to talk about this, Mark, because usually the big breaking news doesn't always line up with what we're talking about on the Everyday AI show, but today it definitely does. So maybe I'd like your take as someone, like I said, who's helping push forward responsible, ethical, safe AI use in the ever changing world of generative AI. What's your take on what's happening at OpenAI with Microsoft now obviously very involved and now Sam Altman and others creating this new research team at Microsoft. What's your take on it specifically from that safe and ethical AI?

You know, I think we're going to take days, weeks, months, who knows in terms of knowing what's behind that Shakespearean drama? But if you back up a little bit, it's kind of not surprising in that we're in the midst of this huge public conversation about, you know, how do we balance the interests of humanity and the set of companies that are kind of dominating the early wave of, early part of this wave of AI. And so, you know, we heard it all in your headlines, right? It was Meta and Microsoft and OpenAI and Italy, France, Germany. You know, whether it's governments or big companies or small nonprofits or people at home, people are saying, how do we want this to work? How do we keep it in control? How do we make sure our interests are considered in that? And so I think you see that big tableau playing out. OpenAI is really interesting. It started out in 2015 as a nonprofit like Mozilla that was trying to build technology, AI in service of humanity. And I guess at the time it was just a tiny project backed by a bunch of wealthy Silicon Valley people who were like, yeah, I wonder where that goes. Maybe that could be cool. And I guess in 2019, sometime around the pandemic, Sam Altman effectively took it private. I mean, there was a nonprofit at the top to make sure that it followed its mission. But I think given the amount of investment that came in from private parties, you know, $13 billion from Microsoft alone, and people were pretty cynical that that nonprofit piece, which we believe in as a model, would have any, any meaning anymore. And who knows what the weekend really meant. But it does look like it was a conflict over safety and going too fast, the interests of people and, and making a lot of money. Maybe it's not, maybe there's something else to it than, than that, but, but it looks like at the base, it's just a part of this big conversation about how do we balance the interests of, of people and companies.

Yeah, and fast. Right. You mentioned that. Because it seems like everything happening in the AI spaces is too fast. It's literally moving forward, sometimes weeks or months, every single day. And there's obviously the famous letter from now, which seems like a decade ago, but you had kind of all of these big names signing a letter saying, hey, we need to pause on AI development to better understand it, how it impacts us all. Mark, is AI going too fast? And if it is, what problems does that mean for everyday people? You know, people who are using these systems, are they unsafe because everything's going too fast?

You know, it's hard to know too fast. It's certainly going fast. And I guess, you know, the question is not just how fast is it going, but you know, what you, what are the risks and where are the guard rails? And so some of the risks are the real near term ones. Right? We're about to go into a planet, we're about to go into a year with 44 national elections in 2024. Right. And we already have seen what misinformation driven by AI in previous elections has done. And so those are kind of some of the things you got to worry about, right, Is how do we know what the truth is? Like, it's that fundamental. How do we keep democracies if we don't know what the truth is? And so I think on some levels, maybe AI is going too fast. You can deep fake somebody's voice, but we've seen that coming for years. On the other hand, we're not going fast enough. How are we using AI to watch for misinformation in more sophisticated ways than we have in the past? How do we use AI to help us see maybe through a browser, what's real content and what's fake? A lot of it to me is actually what are we using AI for? And some of it's too fast, and some of it's not fast enough.

It's an interesting point. And it seems like a lot of times when we talk about, you know, even. Even regulating AI, right, people say, hey, the. The best way to regulate AI is to use generative AI systems, right? Is that also problematic? You know, and you've seen even the biggest companies say, yes, we are going to, you know, regulate our AI with humans, but also through use of AI. How, Mark, how do you find that balance? You know, even of like, hey, how do we go faster? How do you find that balance of still being responsible about it when you do have to go fast now to keep up? How do people and companies do that?

Well, again, I think you need guardrails, and AI is going to be actually a part of the guardrails. And really the question is, who's holding them? So governments, for example, need to, and I think are quickly trying to get their, their act together in terms of creating accountability. You saw a really ambitious executive order come out of the White House a few weeks ago that talked about all kinds of guardrails, and it talked about testing big AI systems. And you're only going to test them by using other AI systems to pressure test them. Right. I mean, it's the same thing we know from cybersecurity, from the last, whatever that is, 10, 20 years is we need to use the technology to test the technology to make it more robust. But unless you have people incented and paid to do the testing, to do the regulation, to have the oversight, which we don't have enough investment in that, and incented to invest in the safety research and the guardrails inside the core companies, which as we saw with Meta, you know, doing whatever they did, but looking like they kind of shut down their responsible AI team, there's clearly not enough incentive there, or they don't care, I don't know. So I think technology has to be a part of how we build those guardrails. There's no question. It's true in other industries, but it's about there being enough authority and expertise in the hands of the government, in the hand of public watchdogs and researchers, and enough incentive and accountability on the part of the big companies that they can actually invest and do the right things. Car companies, we know if you go back 50, 60, 70 years, left to their own devices, wouldn't invest in safety. And it really took in the 60s and 70s a whole lot of pressure from the public to say, you gotta invest in safety. It didn't stop car companies from making cars or getting rich. And in fact, it didn't stop safety innovation, it sped it up. So we just need that balance between, you know, watching what's going on and the requirement to do things in the interest of the public while still running a company.

Are you still running in circles trying to figure out how to actually grow your business with AI? Maybe your company has been tinkering with large language models for a year or more, but can't really get traction to find ROI on gen AI. Hey, this is Jordan Wilson, host of this very podcast. Companies like Adobe, Microsoft and Nvidia have partnered with us because they trust our expertise in educating the masses around generative AI to get ahead. And some of the most innovative companies in the country hire us to help with their AI strategy and to train hundreds of their employees on how to use Gen AI. So whether you're looking for ChatGPT training for thousands of or just need help building your front end AI strategy, you can partner with us too.

Just like some of the biggest companies in the world do.

Go to your everydayai.com partner to get in contact with our team or you can just click on the partner section of our website. We'll help you stop running in those AI circles and help get your team ahead and build a straight path to ROI on gen.

Couple, couple of things I want to dive deeper into there. Mark, from, from, from that response we talked about or you know, you just talked a little bit there about some of the news there with, you know, mata, the Facebook parent company kind of disbanding its internal responsible AI team. Do you see this maybe setting the tone for other big companies? Because it seems like most big companies, you know, that are developing the generative AI systems that we all use and even just straight up art artificial intelligence systems that don't have the generative component. It seems like they all have those internal teams and they're there to, you know, quote unquote set up the guardrails. Do you see maybe this being a trend where maybe companies are saying, hey, because of this internal team that we set up, it's actually causing us to go too slow and we want to go faster. Do you see that as, as maybe something that might continue to happen.

Well, it's a thing to worry about, right? There is a market pressure to go faster. And then you do hear in that famous or now infamous Marc Andreessen letter, his manifesto on whatever the hell it was, where he said, any ethical AI, any people who are trying to set up guardrails, they're the enemy of innovation, the enemy of progress, the enemy of society, which is boohocky. You see a set of people who see the need to find this balance as, you know, as friction. And it's actually friction that's needed. Right? Friction is actually a part of a good system used in the right way. So it may be that there are people who feel like it's slowing them down and they may try to shut those things down. That might be a trend of a kind. We're also in a moment where governments are saying, sorry, slow down, we need guardrails. We're in a moment where the public, like, look at how many people are listening to your podcast, are saying, slow down, we need guardrails. Like, we were not having this conversation about AI at that level as our society a year ago. So I, I, as I say, I think this is a kind of a grand moment we're in and a kind of grand wrestling with the balance between the public's interest and the interest of, of a few big companies, hopefully. What has happened in moments like this in the past, as I talked about auto safety, oil monopolies, I mean, this list go on and on. Our societies actually do have a history when we get to this kind of hot moment, of reigning things in, and that's what we need to do. And certainly I call on our, our government leaders, whatever country you're in, to, to keep leaning in on that.

And hey, as, as a reminder, if you're just catching up now, or maybe you need a refresher, we have Mark Sermon, the president of Mozilla foundation joining us to talk about safer AI. So actually, Mark, a great question from Tanya here in the live stream. Tanya, thank you for the question. So we just talked about government involvement and, you know, some of the pros and cons of that. But Tanya's questions here. How do we regulate the government's involvement so that it benefits society without the holdbacks and politically instigated regulations? Tough question, Tanya, but I'm sure Mark can, can handle this one.

Well, I think one of the key things is really transparency on the part of the AI companies and transparency, of course, on the part of the government. I mean, There's a whole reason that we have been an advocate for open source in AI. We signed a letter with a bunch of very famous AI scientists, as well as activists and policy people a couple weeks ago around the UK AI Summit calling for open source AI to be protected in regulation because openness and transparency are so key to being able to look into the black box of what's going on. So they're key if you want to regulate, but they're also key if what you want to do is even keep regulators in check. We need to understand what's going on here in order for us to regulate effectively and hold people accountable, whatever party they are in the system. I would just really lean on transparency as key.

And hey, speaking of that, let's, let's get back to what's important here. Let's talk a little bit about Mozilla AI. So kind of explain now that we have actually a great foundation in this podcast so far of some recent event trends and where we're going. How does Mozilla AI play into this equation?

I'm kind of embarrassed by this website because it came out in March when we launched and we did the, I don't know, the perfect or the silliest thing ever, which was we announced what we're doing and then kind of went into stealth mode. But what we're doing is still what this, this website said in March, which is building a startup and a community, building trustworthy and open source AI. And what that means is we really believe, as has been the case with the, you know, the last 30 years of the Internet, that you need a set of open building blocks. Whether those open building blocks in the past have been Linux or the web stack, HTML, JavaScript, all those things, or now things like open source large language models. If you want a broad set of people to be able to innovate, if you want things to be transparent enough for people to be able to keep things safe. And so that's what we're in the business of doing right now. It's a kind of core team of about 20 engineers really focused on taking the wave of open source large language models that have come out. Yes, llama, but not primarily llama. You know, also things like Eleuther, things like the Allen Institute is working on, things like Falcon, which has come out of the uae. You know, you've seen this growth of basically open source clones of things like OpenAI or of GPT X. And what we're working on is how you make them safe and usable so you can easily download those things, but there's not that much you can use them for out of the box. And so just like 20 years ago, as more and more people turned to Linux, Linux distributions emerged as a way for you to quickly get Linux on your system, configure it to do things, configure it to add other functions around the edge. We're working in Mozilla AI to do that, helping people take open source large language models, train them on their own data, fine tune them so they're super accurate and they're not hallucinating and they're not lying, and evaluate them to make sure that they're safe.

You know Mark, I'm super curious because from, from my vantage point, right, I've always been kind of a tech enthusiast, right? I've been building websites for like 25 plus years. So I've always been a little bit of a geek and you know, been, been, you know, putting my, putting my foot in the AI waters for a couple of years now. But it seems like right now it is easier than ever for an individual, an entrepreneur, a small business owner to leverage AI, right? So even some of the things that we're talking about, you know, open source large language models like Meta's llama, you don't, you don't have to be a developer or you know, an AI expert to necessarily use these tools, right? Which I think is both exciting and it brings I think a lot of optimism for, for business growth. But at the same time, a lot of, I think a lot of times in the past because of those kind of higher hurdles to clear, you would generally have someone who's maybe more quote unquote of an expert using and leveraging these tools. So are there actual downsides specifically when it comes to trustworthiness and you know, safe and effective tools, that it's actually almost so easy that anyone can use it, that maybe people are just making mistakes when it comes to their data, when it comes to creating trustworthy models.

Well that's exactly the question and I think that that question about are the risks that come from how easy this is to use and anybody can use it for, for anything emerge with both the closed models and the open source ones. I mean you can bend OpenAI or BARD or you know, GPT or Bard to do lots of different things as well as bending the open source LLMs to do lots of different things. And that's why we're focused on trustworthy AI. Possibly that's why people, you know, got rid of Simon Altman at OpenAI over the weekend, right. Is there are big questions around trustworthiness and safety. That's a space for innovation. And so we're trying to kind of build that safety and usability layer on top of open source large language models again, so they're not only accurate, like let's say I'm doing cancer research with an AI research assistant. I want to know that it's not hallucinating. Right? We know that these LLMs hallucinate, that it's, you know, what it's giving me back from looking at this whole body of cancer research data that I've collected is something that I can actually trust to go design my next experiment. But also if I start using that in social circumstances where I'm judging people's loans or housing, any of these things where discrimination emerges, I want to be able to double check, evaluate that it's not discriminating against people. So all of those things we've been talking about that broadly people talk about as AI safety or trustworthiness are, are real areas where we need innovation as well as guardrails, but we need innovation. So that's what Mozilla AI is really focused on, is making sure this stuff that's easy to use now is also easy to use in a way that is trustworthy and safe.

Great question here that I'll bring up. And I was also curious about this myself because Mark, we were talking a little bit earlier about even the Biden White House with their executive order kind of on AI regulation, which I'll have to save my thoughts on that for another day because there's. There's some things in there that are great, but there are some things in there that are very, very broad and a lot of, a lot of. Right. But split Lana here is asking, are there any resources or reports Mozilla AI has published on safer AI best practices?

The thing to look at now, and you can, can pretty easily find it if you just search for this. You can, you could Bing it or Google it or whatever you search in. I mean, it's an interesting thing. I've been trying a Bing to see how they're using generative AI. But anyways, the thing you can get now is there's a paper from 2020 called Creating Trustworthy AI by Mozilla that myself and a number of other people wrote. And it was an early take, before we were in this frenzy on some of the things you need to do, which are, you know, shift industry practices and norms in terms of trustworthiness, build different building blocks, help consumers to be more aware of what they're asking for or getting in AI and then help regulators make better choices. So you can take that paper. I actually think a lot of what's in there is really durable, even though it's three years old. We're coming out with a new version of that paper or a progress report on that paper, I guess, in January. And then Mozilla AI itself, because we're building some pretty deep tech, has been quieter and in the background they've been hiring engineers, setting up infrastructure, running experiments. But I think you can expect more from them also in the new year. So, you know, after we kind of rest from a year of AI over Christmas for a second or two, I think you'll see more reports and also some, some software early from, from us early next year.

All right, fantastic. And if you're listening on the podcast, don't worry, check the show notes. We'll, we'll make sure to include that kind of paper that we have up on the screen now from Mozilla. And, and when it comes out, I'll, I'll hound down Mark and make sure that we can send out that. Send out that new version to you as well.

Absolutely.

You know, Mark, something else that I'm, I'm very curious about, because when we talk about, you know, you know, trustworthy AI and ethical AI tools, it almost implies. Right. Without saying that maybe some of the tools out there aren't safe or maybe aren't ethical, what are some of the biggest concerns that you have that, you know, most of us everyday people who aren't studying this should be aware of? When it's maybe like, hey, it's not safe or maybe it's not ethical to do A, B and C. What are those things that we should be looking out for?

Well, there's, there's, I mean, it's a long list.

How much time do we have here? How much time?

Some of them are really at the societal level, but then they impact us individually. And I kind of go with this at the individual level because that's what you're asking, you know, so at the societal level, we have all these questions of misinformation and it's impacted huge numbers of things from health issues to democracies. And I guess at an individual level, just being hypercritical, especially if you're going to be sharing content about the source of the content. I mean, it sounds boring, but that's an AI issue and it's an AI issue at a huge scale. So being critical about the authenticity of media material that you're sharing, basic but important. Then I think when you think about more the generative AI tools that are emerging, just be conscious of what data, especially if it's not yours, it belongs to your company, it belongs to your community, it belongs to your friends, what data you're putting in there. That's one of the reasons we're interested in open source large language models is you can set them up on your own infrastructure or on some infrastructure you control and really control where the data goes. So I think just be conscious that if you're using data that's not yours to use and you're throwing it into AI systems, you're basically offering that into the training data stream of those systems. And potentially, although I think this is not as common right now, somebody's going to surface that data and find that actual fact. That might be a secret.

Yeah. And something that I'm thinking a lot about now, Mark, is how our everyday usage of generative AI is probably going to change. Because I think for the last year, or three to four years, depending on how early of an adopter you are, you were logging into a system, right? You were opening a browser, logging into a system to use generative AI. But now, as we see with systems like Microsoft 365 Copilot, whatever, Apple's forthcoming Ajax GPT, whatever you want to call it, whatever their system may be, but it seems like generative AI is, is going to creep onto our devices, onto our operating systems. You know, we have the, the humane pin, right, which is going to be following us around everywhere. How do you, how do you prepare for that type of future? Well, in some cases it's not the future because, you know, 365copilot has already started to roll out to some enterprises. But how do you prepare for it then when it's just everywhere?

I think there's the same set of questions, right, is in who's in control and if it's somebody other than you, you know, in whose interest is it being run? And so that's where I actually think in the long run, I don't think this is the focus of humane. I have a real interest in on device personal AI that you control that is sort of sovereign in some sense. And open source lets us get to that future faster. So if, you know, if there's really something totally on my phone or on my PIN or whatever, and I could control, you know, the relationships it has with all the thousands, millions of other AIs out there in the world where it's kind of my agent. Like that could be a good world, but where that is something that is run by a cloud services company that's vertically integrated, that is in everything from social media to backend to blah, blah, blah, blah, blah, that sells advertising. Probably the way that that thing is running isn't actually designed for my benefit. So I think a lot of this is who's in control and who benefits as we move into this era. And I think there is a possibility and those of us working on the kind of open source, hybrid, public interest, commercial kind of space are trying to drive that possibility that we could be in more control, but it's not foregone. In fact, you know, things are trending in the opposite direction.

All right, so, so Mark, we've, we've already talked about a lot. We, we've talked about how some of the recent news, like what's going on at OpenAI and you know, Meta disbanding, their kind of responsible AI team, how that affects us. We've talked a little bit about Mozilla AI and you know, kind of even the future of generative AI and how we can, you know, have safer and more ethical systems. But to, to kind of wrap this up, what is the one takeaway that, that you hope, whether it's it's a business leader helping to make decisions on generative AI usage at, you know, a Fortune 500 company, or maybe a solopreneur entrepreneur trying to figure out how to use it to, to grow their new startup. What is that one piece of advice or that, that most important takeaway that you can give to people to, to make sure that they have more safety in their generative AI kind of tool stack and just tools that they can trust.

I don't know, I guess it feels like we've talked about this in different ways, but it is really make sure that you're super critical about in whose interest this set of tools you're buying are operating. In particular, keep control of your data or know how your data is being used. Because we're moving really fast into a place where we're kind of signing a lot away and it needn't be that way. And then start to look for providers. We will become one of them. And I think you're starting to see more and more of them who bring open source as an option that you can use as a way to keep more control over your own infrastructure and your own data.

So much good advice in this episode about how we can have safer and more ethical AI tools that we use every single day. Can't thank you enough for joining the show, Mark. We really appreciate your time.

Thanks so much, Jordan. It's been a blast.

Hey, and as a reminder. Yeah. There's a lot more AI news we couldn't get to today. A lot more from from Mark and the Mozilla foundation and what they're doing even at Mozilla AI. So make sure you go to your everyday AI dot com. Sign up for that free daily news newsletter. It's in the show notes every day as well. For more on that, hope you can join us tomorrow too. We're gonna have a dedicated conversation about what the heck is going on with this Shakespearean tragedy. And fast forward at OpenAI for joining us today. And we hope to see you back tomorrow and every day for more everyday AI. Thanks, Y'. All.

Thanks.

AAAnd that's a wrap for today's edition of Everyday AI. Thanks for joining us. If you enjoyed this episode, please subscribe and leave us a rating. It helps keep us going. For a little more AI magic. Visit your everyday AI.com and sign up to our daily newsletter so you don't get left behind. Go break some barriers and we'll see you next time.