Talking Tech with Roy Austin & Alex Givens on The Future of AI Governance
Artificial intelligence is rapidly transforming how we work, communicate, and make decisions, but its impact is not just a question of technological innovation. It also raises critical questions about fairness, accountability, civil rights, and how we ensure these systems serve everyone.<br><br>As policymakers, researchers, and communities grapple with these challenges, leaders across sectors are working to shape the future of AI in ways that reflect our values and protect the public interest.<br><br>In this episode, we're exploring some of the biggest questions facing AI today — from responsible development and governance to ensuring that communities historically impacted by technology are included in shaping its future.<br><br>Joining us for this conversation is Roy L. Austin, Jr. the Inaugural Director of Howard University School of Law’s almost 1 year old AI initiative and Alex Givens, President and CEO for the Center for Democracy & Technology.
Top Keywords
- need 0.008
- tech 0.007
- work 0.007
- howard 0.006
- tools 0.006
- civil rights 0.005
- make 0.005
- domestic 0.005
- build 0.005
- frontier 0.005
- domestic violence 0.004
- alex 0.004
Transcript
Welcome to CDT's Tech Talks, where we dish on tech and Internet policy while also exploring what these policies mean to our daily lives. I'm Jamal Magby, and it's time to talk tech.
Welcome to Tech Talk. Bye. CT.
Artificial intelligence is rapidly transforming how we work, communicate, and make decisions. But its impact is not just a question of technological innovation. It also requires critical questions about fairness, accountability, civil rights, and how we ensure these systems serve everyone. As policymakers, researchers, and communities grapple with these challenges, leaders across sectors are working to shape the future of AI in ways that reflect our values and protect public interest. In this episode, we're exploring some of the biggest questions facing AI today, from responsible development and governance to ensuring that communities historically impacted by technology are included in shaping its future. Joining us for this conversation is Roy l Austin junior, the inaugural director of Howard University School of Law's almost one year old AI initiative, and Alex Givens, president and CEO for the Center for Democracy and Technology. With that, I'll turn it over to Alex to begin the conversation.
Roy, it is such a pleasure having you on Tech Talk. So happy to have you here today. I am so happy to be here, Alex. Thank you so much for the invitation. Alright. So you spent much of your career working at the intersection of civil rights and public policy, including in the justice department civil rights division, the White House domestic policy council, and later standing up Meta's civil rights team, which came after that big civil rights audit led by former CDT board member Laura Murphy, and was a really big deal for those of us at the time. Now you're at Howard launching their AI initiative. What's motivating the work? What issues are you focused on? You know,
Alex, first of all, you you jump into the serious questions. Can can we talk World Cup just for a moment? I I was not an England fan before, and I was so proud of how your, your boy showed up. It was pretty amazing.
That's so funny. So you don't know that I'm from Cape Friday now?
Well, I may move there. I'm thinking.
Cultural appropriation, I suppose I shouldn't claim it. I am indeed from England for those of you who know my roots, and I was proud of them through the end. They tried. Yep. Yep. One of the most moving scenes was the the crowd singing, hey, Jude to Jude Bellingham,
after that Norway match. It was just it was just powerful, but it's actually relevant to AI. And and so that's also part of why I I I raise it as they did something that that we really need to do around the world, and that is that they gave every team, all 48 teams, as I understand it, access to the same quality AI, information. So they were measuring everything about the players, about the touches, about everything, and they share that equally across everybody. And so I I raise that because that's really why I'm at Howard Law School, with this AI initiative and what I hope to do. And that is AI is an amazing tool, and what we need to do is make sure that it's available to everyone so everyone can can use it to advance their work, can use it to advance their their personal lives. Just really, how do we make it something that is shared by by all? I love that goal. The way to,
the strategies for it are challenging right now. So I'm curious. Tell me a little bit more about some of the specific things you're working on, how you're structuring the initiative to to make progress in that direction.
Yeah. So I I kind of look at it in two buckets. One is, you know, the bias discrimination pieces that it's just real world, and and and it's showing up in in our AI. And so how do we tackle those? But, also, we have the wealth inequality pieces, which again is is real world, but is, really exponential in the AI world. What we're seeing is, you know, five I call them basically the five dudes who have all of the wealth and are really using that wealth both to build more, but also to influence our politics and influence whether or not we have regulation and everything else. So the ANI initiative is broken up into three pillars. The first one we call the Impact Foundry, which is led by Sandia Rao, and that's an accelerator to help people who are building, founders, who are who are building AI tools. And we're gonna start that off hopefully this fall, with some seed funding for for folks in an RFP. The second one is the justice lab, and the justice lab is led by Malika Sadasar. And that's looking at the issue of our nonprofits, our NGOs, civil and human rights organizations, some of them being really reticent about using AI. And as many problems as AI has and comes with, it's also an amazing tool for advocacy. So bringing together the early adopters with those who are reticent to build tools. And and we actually just recently saw them come together, and and they built a a tool. One of the tools they built was something to work with domestic violence survivors. And then the third pillar, we call next gen leaders, and that's led by Melanca Clark. And that's the idea that we really want to have leaders in AI that are centered in human and civil rights. So how do we give them the skills?
How do we give them the information that they need to become the next set of leaders in artificial intelligence? I wanna pick up on all of those threads, but starting, obviously, as a nonprofit leader myself, I'm thinking a lot about AI integration, how to do it, how to do it in an ethical way, how to stay true to our values, but also how as an applied matter to get smarter in this tech so we can have more informed conversations about regulation and governance. I'd love to hear about the the DV survivors example. So how how are you thinking about the work there? Yeah. I mean, so the work was is is let's bring together technologists
with subject matter experts. And so, you know, what do the subject matter experts need? What does someone who is a survivor of domestic violence need to improve her or in in rare cases, his life? And how can AI be that support? How can AI be that thing that directs them to the resources that they need, quickly when you can't necessarily get to a counselor when you need emergency information? And so that's the kind of tool that was built, in a in a few hours. So it's, you know, it's not ready to market, but what can happen now is now the organization that works with domestic violence survivors can now take this next level and can, you know, over a longer period of time,
build something that actually could go to market that could change lives for the better. That work feels really important for two reasons. Like, one, there's obviously the service component of it. So you are literally helping a really important nonprofit do their work more efficiently. But the angle that we've been thinking about at CDT is that when you do that and it is the right people helping build those tools, you also can live our principles of how to build and prove that it is possible to build these tools responsibly as well. So, like, what is it to do with trauma informed practices? What is it to do with meaningful privacy practices? What is it to do with effective content moderation that thinks about guardrails and the appropriate ways to respond? What is it to do it with, like, multilingual, multicultural competency? And I'm curious how you think about that intersection. It's coming up a lot for us in the civic tech space, so there's a lot of energy around AI for good. And one of the things that we're trying to say is, like, yes. That is amazing energy, and please don't forget that this is the moment to model how to build these tools responsibly and well and show that it's possible by having that expertise at the table as well. So I'm curious kind of if you're thinking about that integrated approach as you do it. A thousand percent, Alex. I mean, the, you know, the the cliche is if you're not at the table, you're on the menu.
And what we have increasingly here is we have these tech people who don't have the subject matter expertise who are building tools for people. But what you really need is people building tools for themselves. So if you're doing something for domestic violence survivors, you need domestic violence survivors building it. If you're doing something for people who are reentering from incarceration, you need them at the table building because because they know what they need. They know how it feels. They're going to see the problems. So, you know, when you're talking about civic tech, we're talking about democracy. And so we need the people to be weighing in to the tools that are built to tell you not just whether they work, but whether they matter, whether the things that actually make them more interested, more willing to be engaged in our civic lives, in our democracy. So, yes, you need technologists. You need people who understand how the tools work, what they're capable of doing, what issues could come up. But when you are building it, if you don't have the people who they're going to be impacting at the table, you're wasting your time. You obviously have a deep background in the criminal justice system from prior aspects of your career and even your work at the domestic policy council focused on this.
Where are the some of the opportunities
you see there for this type of approach, new tooling or interventions that could be useful while still being deeply respectful and protective of the vulnerable status of people that would be using those tools? Oh my. And the criminal justice, it it's it's everywhere. And so it's, you know, starting off, let's take legal aid, for example, and and legal aid is inundated with people who need help with everything from receiving benefits to rentals and and things like that in housing. The ability to provide people with quick answers and accurate answers. So you really need a tool. Like, maybe this is as simple as a chatbot, but it's it's not just like gathering information from everywhere. As we source it, if we build it as a a rag on top of a frontier model, frontier model, as we source it, making sure that the information is accurate, making sure that the people who work at legal aid are the ones who are, in fact, the ones who are training it, testing it, red teaming it, making sure that it's providing accurate information. But you can serve so many more people. So the average person can go and and say, you know what? My landlord is is not giving me the services that I need. File a complaint in moments, asking the right questions, getting it set up, getting it teed up for a proper complaint to what we're seeing in the surveillance world right now, which is really kind of awful in the way it's used, but it it doesn't have to be. It's really again, this is user error. This is human beings being awful, not AI being awful. And so take something as simple as a a license plate reader. In and of itself, that's fine. Like, it'd be great if we had, you know, if if someone were to steal our car, we would want it back. If that's what it's being used for, that's one thing. But if it's being used for immigration enforcement, if it's being used as they have something right now called signal trace, which can actually download anything that you have on Bluetooth into a system and follow you in that way, being used by law enforcement in ways that completely reverse what we thought of as a fourth amendment right, that's problematic. You know, when we're deciding on fairness across systems, I I wanna say it was it's California that has removed race from charging documents, from sentencing documents in a way that you you can't really do human just by by human beings alone to try to ensure fairness in how they are asking for sentences and how they're asking for probation and and things like that. There is nothing in the criminal justice. Training inmates, providing people who are incarcerated with job training is something which AI can do better than trying to have mismatched individuals coming in institutions, leaving institutions, security concerns, and all that when you can actually have AI train people so that when they are released and and the vast majority will be released, they're ready to go and get a job. So there are so many things across the criminal justice system. But, again, it's it's the human being that matters and someone who's using it for its proper purposes as opposed to something that is problematic.
Yeah. I mean, what you're pointing to is really there are two things that are at their highest point right now. Right? One is the promise and the opportunity of these tools, and the other is that skepticism and awareness of the risks has never been higher, and it is not tinfoil heading. It is legitimate concern based on proven evidence of when people's data has been improperly shared, misused. You share it with a government agency so you can access SNAP benefits, and suddenly DHS is trying to snap it up for immigration enforcement. Yep. So one of the things that we're really focused on at CDT, and I hear in what you're saying too, is, like, how do you hold those two truths at once and not give up on the promise of the tech to deliver, but build it right the first time to try and overcome that trust gap and try and create this new paradigm of saying, we can have that service, and we can have it with legal and technical and business model protections built in as well. Because that's the only way that we're actually gonna have people use these and get the benefit in the types of impacted communities that we're talking about. This isn't the big bucks deployment. This is about genuinely using these tools in service of communities and people who need them most. Yeah. We have a real problem right now because we we're dealing in a in a time when the tech companies are are leading everything and and the people aren't leading everything. And and and so, you know, they're looking for,
I hate to say it, but it seems real, a quick buck. The tech companies are. They're not looking at it as far as like, okay. What are the guardrails that we actually have to put on this? They they they talk broadly about, yeah, we we need some regulation. But if they truly want a regulation, they could come up with the things that they could be regulated on and that would be appropriate and and that would actually allow them to to do their work at the same time. So that's kind of a joke that they actually want to be seriously regulated. I don't know how we get this technology with the speed that it is moving into the hands of people to do the good work that needs to be done with it. And and so I I should note an another piece of of what we're trying to do at at at Howard Law through the AI initiative is we we kinda have to start building our own. We have to start building our own tools instead of waiting for them to build tools for us. And if we build our own tools with our own data, we might actually create something that that people are just naturally going to go to what is good, not what is expedient. And it may not be instant, but I honestly think at some point here, people are gonna look at AI and say, wait. Wait. I have a choice here, and I have a choice of using a product that is in fact good. Maybe I'm gonna go to that product instead because it does the you know, we're we'll do the vacation stuff, and we'll do the recipe stuff, but we'll also do the, you know, how do we deal with poverty and and homelessness and discrimination and hate and, you know, environmental damage and all of the other pieces of it.
So we can use AI for that, and we could also use it for the fun things that we wanna use it for. So that sounds like some of the underpinnings for the AI Impact Foundry. Is that how you're thinking about it and thinking, hey. This can be an entrepreneurial play. Like, it it's not all philanthropy. Like, this there's actually a business model here. Look. You know, I I've become a huge fan of crowdsourcing,
and there are just so many ideas out there. You know? I I think I have lots of great ideas, but they aren't going anywhere. What we really have to do is just create opportunities for people to to to come up with things. And so if we give people some seed funding, you know, they come in with a a really general idea and we say, you know what? That's an interesting idea. It wasn't something we thought I thought of. Let's give you some seed funding and let you develop that. I think we're gonna be amazed at what people come up with. Because because people know what's in their lives. They know what's gonna make a difference in their lives and the lives of people close to them. And so let's give them the opportunity to build those things out. So that is exactly, what the impact boundary is. And and we wanna deal with workforce first given kind of and this is what the tech community did is they said efficiency, efficiency, efficiency. We're gonna get rid of your jobs. We're gonna make it so that all you need is a a a universal basic income, and and you'll have fun. And everybody's like, well, wait. I actually like working, and I actually enjoy that. And I and that's a part of my life. But so how do we use AI to fix AI to make sure that it's actually leading to people having productive
work lives? I wanna turn to kind of the first pillar that you started with, which is the anti discrimination work, thinking about how to make sure that systems work for everyone. And, obviously, that is a fundamental issue that our field, CDT, many others have grappled with for some time with and and you certainly worked on when you were leading up, meta civil rights team. I'm curious how you're seeing those issues play out as AI is increasingly integrated into decisions that affect people's employment, education, housing, government services. Are there lessons from what we've gotten wrong before? How are you thinking about the next frontier of that debate?
Well, there there's certainly lessons from what we had before, and some people are calling AI kind of the the next civil rights movement, because it impacts everything that you just listed, Alex. I mean, it it it impacts what who can buy a home, who can get a mortgage, what kind of education people can receive, their interaction with the criminal justice. So it it impacts all of those things. I'm not I I mean, this is really where we have to start thinking harder and harder about building our own because the you know, what is what is AI built on? It's built on Wikipedia and Reddit largely. I mean, there there some other stuff as well, obviously, some stuff that's been stolen, the whole copyright question. What's not in there are the papers of many historically black colleges and universities like Howard that is sitting on the largest collection, that exists, of black organizational papers and personal papers, not digitized. It's not digitized, so that means it's not part of what is training AI. Think about, Hispanic serving institutions. Think about Native American institutions. Think about women's colleges and universities. Think about, you know, the papers and the works of people with disabilities and and and the LGBTQ plus community. There's a ton of data out there that could be training the models to answer these hard questions that is not part of the current training material. How do we get that not to be part of what the Frontier Labs control, but that we control? We being the people control it. The communities control it. And I think that we're moving increasingly in that direction as we're seeing, basically, your your big tech companies are pricing people out of the work or or out of using their tools. So if that keeps happening, people are going to find cheaper alternatives. And those cheaper alternatives, I think we can build in a way using what we have open source, putting this material on top of it to train a model or train many models that really speak to our particular communities. And that's how I think we can people who actually care about bias and discrimination and recognize where it exists and have thought about for a long time, how do we how do we limit the amount of bias, how do we limit the discrimination,
if those are the people that are building the models, we will have better model. Yeah. It it feels really important, and it's an interesting expansion because I think in my mind, there are categories. Right? So when AI is being used to make a decision about a person, like, there we need legal safeguards, verifiability, right of appeal, etcetera. But there's also the entire other picture of is a model actually culturally competent? Is it linguistically competent? Does it have a sense of context? Representational bias, which of course has been an issue in search for a long time and has not gone away as we now think about what this next frontier looks like. I'm curious if there are places you're looking to for inspiration. You know, the Maori very famously were kind of at the front of this curve in terms of asserting sovereignty over long held recordings of their audio transcription. They were very careful about kind of how that was used and captured for training LLMs. Are there places that you're looking to or places where you see kind of affinity in what this move could look like for data repositories like what Howard is holding.
Yeah. So Howard did one quick project that that really kind of speaks to this, and this is they they went around the country and and looked at the African American vernacular. And they said, you know what's wrong here is that when African Americans speak to AI, AI doesn't understand them, doesn't understand their culture, doesn't understand the language. And so how do we actually train it on the hundreds, if not thousands, of different types of English that is spoken here in The United States alone? So building on that, you know, it's incredibly important that, you know, we we kind of have pretty basic English training all of our models, and you lose nuance. You lose culture. You lose everything with that. So how do we grab more people with their native language in the way they speak it and and make that a piece of it? But the problem is that we can't give that all to the big tech companies because they're just gonna use it and and make a mismatch of it because what we really want is we want people to be able to live in the communities that they wanna live in. So how do we take native American languages and not exploit it, but actually use it as part of, a model that that native American tribe can use to better themselves? And so, you know, those are the kinds of examples of the difference it makes to get this thing right and to bring in the right people.
Yeah. I love what you're talking about really hits because I think there are many use cases where there are communities and this could be both in terms of language, culture, also a lot of knowledge generating institutions that have a nonprofit purpose. And the natural instinct should not be to put it behind a paywall because you actually want the models to use it. And yet the notion of it just being suckable by a data grab without compensation with, in some cases, like Wikipedia, overwhelming burden on the channel on the website itself because there are so many calls going to their web traffic every single day by scraping bots. CGT has been doing more work on kind of this conundrum of, like, what is it if you actually want your material to be scraped and used by AI models and it doesn't work to to simply say this goes away for free and you can scrape anytime that you want. What are some of the other interventions? We've seen obviously some of the science journals and medical journals have just gone paywall all the way, personhood credentials to be able to access the site. For a bunch of folks, I think museums, cultural institutions, you know, some science communities, that's not gonna be the solution space. So it'd be interesting to follow-up with you because I do think that that's a really important third path for how we do that and what is the role of open source, how are there ways to do this, and what are the alternative compensation models that do recognize the value
of that inherent knowledge that is being collected and used. Yeah. We're we're working with a team from, called Axum, a x m. And what they have found is they what they have built is something that follows your IP across the frontier models. So at a minimum, you are able to say to your big frontier companies. You're able to say to them, hey. Wait. Wait. I didn't give you permission to download my book, but it's clear that you did. I didn't give you permission to take a photo of my piece of artwork, and you're putting it on your platform and so you can move to cease and desist letters. This is a real, you know, the alternative funding models are what we really probably need to look at here. Because what you have is they've scraped what they can, and they're making all the money. So the person who put in the labor to to write that book, to paint that painting, to make that movie, to do whatever the creative piece of it gets nothing. And they're profiting enormously from that. Look. They're they're looking at now paying their way for their environmental costs. Well, you know what? I suspect that if push, they they will find a way to pay their way. I mean, look. You can take almost like a Spotify model to pay for what they're taking from creatives. And a team like Axsome is working on ways to to make that happen. I'd love to end with a final question. So you're at Howard. It's a learning institution. So you were thinking a lot, not just about the current scenario, but what are the future leaders? What do they need to be
ready for? So when you are talking to students about the future of AI, the world in which we want them to live in five, ten years, how they need to show up to help build that world, What are you talking about? What are you prioritizing as you talk to them? And we can let that wrap us up. Yeah.
Well, first of all, I'm talking about not AI. You know, the first thing is, you know, we we have to learn to to to reason. We have to learn to think. We have to learn to discern all of those things. And so I'm seeing what is happening at a number of law schools right now where they're saying it, like, first year core courses. No computers, no phones, come in with your pen and paper, and you're going to learn to think, and you're gonna think in class. And so a piece of it has to be that. And then once you've learned that because the thing is is we are of a generation, I think, where we've we've had to do that. So for AI, we can actually look at AI and and we can say that's not right. But for people who simply all their information comes from AI, they've lost, I think, the ability to discern. So the first thing is, and it it's kinda like long division. Why do we all learn long division in in elementary school and what's not when we can just punch it into a calculator? Well, there's something about going through the steps of learning these things, which makes it so so we are better and more capable of being smart at the end. And then once you learn the basics and you understand how the world operates without AI, let's add AI to that. And let's look at what you can do to improve certain aspects, but not everything. Because, really, where we're ending where we're where we're headed right now is your AI is gonna fight my AI, and we're still gonna need a human being to make the decision. And so in the educational spectrum and educational institutions, we need to have people who know how to make good decisions regardless of the existence of AI. And I think that's something that we can train people for. I think that's something that we need to be doing in our educational institutions across the world.
Wonderful. I diverged from you on long division only because I spent yesterday evening teaching it to my youngest daughter. I'm ready to be done with it. But that aside She said, mom, why can't I just put this in AI? It'll give me the answer. They're gonna do it. Also, I had to get a little refresher for myself on what the most useful tips are. Math is different these days. They've reinvented it, people. Roy, this was such a pleasure as we knew it would be. Thank you so much for your time and all of the great work and leadership that you're doing. We're excited to stay in touch and keep talking about what you're up to. And, Alex, I love what CDT does. It's an amazing organization
run by an amazing leader, and really trying to keep the human, in all of the work that we're we're doing here. So so thank you for this opportunity, and thank you for what you do.
Thank you for listening to Tech Talks presented by the Center for Democracy and Technology. I've been your host, Jamal Magdi. Tech Talks is edited by Jacob Kaufman and produced by Drew Courtney. Check out more of CDT's work by visiting us online at cdt.org and on various social media at sendem tech. That's c e n d e m tech. Thanks for talking tech.
