JASON NABI: Good evening, everyone. I'm Jason Nabi. I'm the project manager for the UVA Futures Initiative. And on behalf of the Futures Initiative, I'd like to welcome you to a conversation with Dr. Alondra Nelson. 

[CHEERING] 

[SCATTERED APPLAUSE] 

ALONDRA NELSON: Thank you, my two friends. 

MELODY BARNES: It's the fan club. 

JASON NABI: I have a bold prediction about the very near future. Tonight will certainly be one of the most pleasant and edifying Tuesday evenings for many weeks to come. Thanks, y'all. And well beyond that, of course, tonight's illuminating conversation is sure to inspire and sustain us for many futures to come. 

A quick word about the Futures Initiative, and then we're on to it. We were launched in January under the auspices of the Provost's Office. We scan the higher education horizon in search of ways to proactively position UVA to thrive in a radically evolving world. 

Toward that end, the members of the Futures Initiative working group, which is a task force made up of representatives from all 12 of UVA schools, several of its administrative and academic divisions, and four of its Pan-University Institutes, members of the working group have been asking far reaching questions about what UVA might do to achieve its strategic goals in striking and innovative ways. And in doing so, to become the University of the future. 

Part of that process, through the Futures Initiative speaker series, involves bringing thought leaders from a variety of sectors to grounds to share their visions of the future in their respective fields. We are excited to continue this series with today's event in which we focus on a future that will require us to better harmonize the great of scientific and technological advances with the good of societal well-being. 

For this, it is our great pleasure to be hosting Alondra Nelson, joined by our own Melody Barnes and Yael Grushka-Cockayne. Please join me in warmly welcoming them. 

[APPLAUSE] 

YAEL GRUSHKA-COCKAYNE: Hello and good evening. I guess I could say good evening. It's wonderful that you all join us on this beautiful day. And I'm honored to be sitting here on this stage. And I'm excited about the conversation ahead of us.

My name is Yael Grushka-Cockayne, as was introduced. I'm a professor at the Darden School. My area and expertise is in data analytics and decision sciences. And I think one of the reasons I'm here tonight, although I haven't actually confirmed this, is because we just announced a new LaCross Institute for ethical AI in business. 

And I am honored to be one of the co-directors, academic directors of our new Institute. And I'm excited about the conversation tonight. 

MELODY BARNES: And I'm Melody Barnes. I'm executive director of the Karsh Institute of Democracy. And I am thrilled to be here with Yael and to be in conversation with our friend, Dr. Alondra Nelson. 

And you all heard the screams and cheers when Alondra's name was mentioned. I'm going to tell you why. I get the pleasure of giving you a brief overview of her bio. And normally, you just say, oh, refer to the printed materials.

But this gives me real pleasure because Alondra is, and I'm just going to say it, a badass. And here's what I mean by that. We're going to have a conversation this evening about the transformative changes that technology will bring to our society and to the University and to our lives. 

And there is no better person to have this conversation with than someone who has the expertise that Alondra has, and who also has a bio that is as mind blowing as that technology itself. 

She is a scholar at the intersection of science and technology and policy and society. And she is currently the Harold F. Linder Professor at the Institute for Advanced Study, which is a Research Center in Princeton, New Jersey. 

She's also a distinguished fellow at the Center for American Progress, which is a think tank action tank in Washington, DC. Alondra served in the Biden administration. And her final role was that of Deputy Assistant to the President and Director of the Office of Science and Technology Policy, or OSTP. 

She was there from the beginning of the administration until the fall of 2022. She is the first African-American and the first woman of color to set science and technology policy for the country as in ever.

AUDIENCE: Woohoo! 

[CHEERS, APPLAUSE] 

MELODY BARNES: And in that role, Alondra spearheaded the development of the blueprint for the AI Bill of Rights. She provided guidance to expand taxpayer support for federally-funded research. She served as inaugural member of the Biden Cancer Cabinet, and strengthened evidence-based policymaking, as well as galvanized a multi-sector strategy to advance equity and excellence in STEM. 

She's also had a distinguished career in the nonprofit and academic fields as well. She was the 14th president and CEO of the Social Science Research Council, and led academic research strategy at Columbia University, where she was the inaugural Dean of Social Science. 

You won't be surprised to know that she's the author of several award-winning books and essays and articles that have been translated into multiple languages. She's currently doing research on a book about Science and Technology Policy in the Biden administration, an essay collection, Society After the Pandemic, which I can't wait to read. That sounds really fascinating, and engage in research on the social power of the platform and governance of AI. 

As you can imagine, Alondra holds many honorary degrees and awards. And she's also was named to the inaugural Time 100 list of the most influential people in the field of AI. So trust me when I tell you, I could go on and on. Literally, I could go on and on. 

But I will close by saying that I'm thrilled that this very, very busy and accomplished woman is also on the Karsh Institute of Democracy advisory board. So thank you so much. And join me again in welcoming Alondra.

[APPLAUSE] 

So I want to start with the first question. And I think the first question should be the obvious first question, which is how did you get interested in and how did you start your work in science and technology. 

ALONDRA NELSON: Yeah So thank you for having me. Thank you for that incredible introduction. I'm both here and on the Karsh advisory board because when Melody Barnes calls, you just say yes. You don't wait. You don't wait to find out what the ask is. You just go like, yes. And then you figure it out and hope you don't get yourself in trouble. I'm delighted to be here with both of you. 

I think like probably many looking at students in the room, I was a STEM kid. But I was never fully satisfied or fully-- like that was never the kind of world I wanted to be in fully. I grew up with parents who work in technology and in science. I grew up in San Diego, California's biotech, science place. So my childhood was like going to the Salk Institute for after school programs and going to the Scripps Institute. So that's like a kind of San Diego kid like that and surfing, or going to the beach. 

Then I got to college. So I was supposed to go to medical school like so many of us. You get tracked, you're supposed to go to medical school. And then I got to college. I went to UC San Diego, where I grew up. And I realized I was much more interested in, people and problem solving and other sorts of things. 

And I found myself, luckily, at an institution that had what's still called, but is rare these days, four fields anthropology, which means undergraduate degrees in anthropology, but which means to finish my degree, I had to do physical anthropology, so which we now call biological anthropology. So I was doing biology. I also had to do courses in archaeology. So that was like soil chemistry and geology in addition to a linguistic and sociocultural anthropology.

So the undergraduate degree that I did had both science and social science together. And that perspective, which sort of old-fashioned on anthropology was sort of you can't understand human societies without understanding all of these things. So not the science and its isolation and not the people in their isolation. You need to understand them together. 

And that felt so right to me. So I kind of immediately went from being on a science track to studying that. But I still was at an institution that allowed me to take like amazing physics classes and chemistry classes and the like. So that, I think, is kind of where that sort of begins. And so I have always thought about those things together. 

I also should say, my mother was an army cryptographer, if you can imagine all things. I mean, she was like a whack or a wave, whatever the thing was, before women could actually be full members of the army. 

But as a child, I grew up with her working on sort of big IBM mainframe computers. My mother was a computer programmer and systems engineer. And so there was also a kind of childhood in which I didn't think that women and computation was like an opposite thing. When you're in it back of the Chevy Vega with punch cards that mom was just using in the IBM, you don't think like, God, if only women could work in computing. 

You're just kind of like computing is like--

MELODY BARNES: This is normal.

ALONDRA NELSON: It's like detritus in the back of the station wagon. And so I think I just had the benefit of just also having this extraordinary path-breaking mother, I think, who made things that I now know as a teacher and as a mentor and as a policy advisor, extraordinary, seem very commonplace. And so that just gave me, I think, a different perspective. 

And then I think more immediately, thinking about my work in the Biden administration, it was-- I had been working on a book about the Obama administration. I had been working on-- I started in 2016 interviewing people who had worked in the Obama administration because I was fascinated by the ability of this administration that you worked in, that was clearly trying to take us back to this kind of bold era of big science. 

So it was like under the Obama administration, that you had the Precision Medicine Initiative, which was this initiative to get a lot of American data in a database and think about how we can do a genomic analysis. It was under the Obama administration that you had the BRAIN Initiative, which was trying to map all the neurons of the brain, much like the Human Genome Project. 

And so I was kind of fascinated in the way-- while also trying, that administration was also trying as this administration, I think, the Biden administration does, is also thinking about how science and technology and innovation are important drivers for the economy and for education and other facets of the social world that we care about. 

So I was absolutely fascinating. And what I saw as like a shift in how we did science and technology policy. And it was through that work that I came to work in the administration.

MELODY BARNES: And so you're now in the administration. The blueprint for the AI Bill of Rights, tell us how that got started and what you wanted to accomplish with that.

ALONDRA NELSON: Yeah. So as a student of the Obama administration, I knew that administration, over the course of 2012 to 2016, had published some very smart that you can still find online white papers around, first, what we call big data, and then by 2016, AI. 

And these were sort of broad think pieces and sort of guidance. And you call it the formerly guidance, you'd call it in government, that were about what is the United States need to do if it's going to be ready for really leveraging and interoperating data sets. And how do we need to think about the privacy implications and what are the implications for work and for health care, et cetera. 

And then by the time you get to 2016, there are white papers on AI and civil rights, AI and job opportunities and employment and the like. So there had already been all this prior thinking. 

And then in the Trump administration, which didn't do a lot of Science and Technology Policy. I mean, we can talk about that in the Q&A if you like. But the OSTP that I arrived to on the first day of the Obama, I mean, the Biden-Harris administration, had only 30 people in it. I mean, we had-- 

MELODY BARNES: Contextualize that for people.

ALONDRA NELSON: Yeah. So the Obama administration had about 150 people working in that office. And when I was working in the Biden-Harris administration, we had also about the same. OSTP does a lot of things, including just the grunt work of congressional mandate.

So every time there's a piece of legislation, some of it says something like and OSTP, every year, will submit data on the thing or the report on the thing. And so we came in after four years with Congress yelling because you didn't have enough people, actually, to fulfill all these mandates and all of this law had said. So the office had just shrunk.

But what the Trump administration did quite aggressively and well was around AI policy. So in the last NDAA-- National Defense Authorization Act-- of that administration, there was something called the National AI Act. And it sort of stood up a few pieces of infrastructure that then we were able to move with. And the AI Bill of Rights idea I think is both comes really emerges from work that the Obama administration was doing. 

But also, I think, some of the work that the Trump administration was doing that we felt in the Biden administration was kind of under-realized. So you had an executive order or pieces of guidance that said things like AI should abide Democratic values or whatever. It's like, what does that mean in practice? Like, how do we begin to implement and think through what that means? 

And so the AI Bill of Rights was an attempt to make that granular. What does it mean to say that we've got shared values as society, bipartisan? And how do we make those real. And the way that we are beginning to slowly, in the United States, advance AI governance? 

MELODY BARNES: I'm going to turn it over to Yael and come back later. I have another question on that.

YAEL GRUSHKA-COCKAYNE: So we're going to tag team it. And we also threatened that we might go rogue at some point. So as I mentioned, I am a professor at the business school. And we often think about various needs for change and strategy. We think about it as either top-down or bottom-up.

Top-down, it comes from the leadership. It's a vision. And it's kind of been shared with an organization that then they have to follow suit. Bottom-up, meaning pressures from either employees or customers or consumers. It comes from the bottom. And therefore, there's some kind of an adaption or a change that occurs because of a reaction to a pressure coming from the bottom-up.

When we think about AI governance, is your sense in the United States or even globally, is your sense that is this going to be more of an evolving top-down approach or a bottom-up approach? Is it something that is going to come from users, from corporations, or is it come from, for instance, The White House? 

ALONDRA NELSON: So I love that you use the term "AI governance" and not AI regulation. Because I think AI governance is both. And that if we are going to get the use and applications of AI to a place that it's mitigating risk and maximally beneficial for the most people, we need to have a suite of tools and levers. 

And so those include things that are, hopefully, if we can get some laws out of Congress, top-down, which would be formal regulation, new regulatory authorities and regulatory agencies, actual laws and the like.

But there's also kind of new-- there's standards. So those are technological standards. What are the ways that we should be thinking about how technologies are built and used? What is a high capability AI model? What is low? How do we think about those things? How is everyone using the same language both in the United States and abroad? These are international standards.

And so those are not quite laws. Those are agreed upon kind of definitions in ways that we think we're going to move in the world. And then there's norms. We've got these new tools and systems. So how are we going to use them. No one's to say like we might talk since we're at a University, one of the world's finest universities, how should AI tools and models be used in the classroom.

I mean, that's not-- I mean, so you can create a kind of University law, but it's not a law per se. It's actually a norm that we're kind of slowly creating. So I think we need the bottom-up and the top-down a whole kind of suite of things. I will say, the United States needs a bit more top-down right now.

I mean, we have not been able to have any kind of systematic regulation around AI in the United States, although certainly, the president's executive order on AI is extraordinary in a lot of ways, and in part because, as colleagues and former colleagues in DC say, that the president said we should pull every lever. And so I think you've got federal agency, you kind of using all of the tools at their disposal to try to make sure that these tools are used appropriately and beneficially.

But I think we need a whole kind of panoply of things. And so I don't want us to get overindexed on regulation and understand that there's lots of other things we can do as well. 

YAEL GRUSHKA-COCKAYNE: And so maybe if I dig deeper, just if I will, like what happens when the movement that is coming from the ground, from the industry, from the users, what happens with it clashes with the regulation that eventually gets introduced?

I think it's actually a moment where those things are converging. So we hear, both in political theater and performance, from big tech executives when they go to Capitol Hill saying, please regulate us. My God. 

MELODY BARNES: It's what we dream of. 

ALONDRA NELSON: If you watched the hearing last summer or two summers ago. But they do kind of mean it. I mean, if you talk to folks in industry, they do feel like things are a little bit out of control. And I think it creates a lot of both organizational risk and financial risk for companies when you don't know the kind of basic terms of engagement for the field that you're engaging on. So there's that. 

I also think, as AI has moved from in March and November of 2022, shot goes around the world. It's like those of us who had worked on AI for many, many years, people thought we were the most boring people in the world. And then all of a sudden, at the Hanukkah table, at the Christmas table, everybody was like, AI. And everyone wants to talk to you about it, and thinks you're very, very interesting.

So that moment--

YAEL GRUSHKA-COCKAYNE: I know everybody in this room has been doing it for a long, long time.

ALONDRA NELSON: But all of you can remember those moments and you're just like, I'm working on this, like AI, whatever. And your whole family was, oh, my God. Shut it down. But that moment of these sort of behind the scenes technologies becoming consumer facing meant that companies want consumers, whether those consumers are the federal government or individual consumers, how do you get adoption. 

And adoption means that also consumers of the technologies, whether or not it's the University of Virginia procurement office or the federal government procurement office or us, as individuals, that kind of adoption that all of these companies want requires there also to be rules of the road. 

And so there is, if you look at things like the Edelman Trust Barometer, the trust of Americans in artificial intelligence tools and systems is like low, like some of the very lowest in the entire world. And so if this is the market you're counting on to live out your valuation of 5 billion or whatever, you're kind of multi-- it's probably a trillion valuation. It's not going to happen unless people feel that the tools are safe, that they're responsibly used, that they're not putting their privacy or their families at risk in engaging in the use of them. 

YAEL GRUSHKA-COCKAYNE: OK. Fantastic. So thank you for that vision and understanding how hopefully these pressures can coincide. 

ALONDRA NELSON: Yeah. I think there are some-- we have this kind of rare moment of like overlapping incentive structures for a little bit of time. 

YAEL GRUSHKA-COCKAYNE: I'm going to change topics just a tad and talk a little bit about the fact that in the Bill of Rights, you mentioned one big chapter or some of the discussion there is around algorithm, discrimination, protection to prevent biases from occurring, and protecting various individuals. 

There is conversation around race, but it's also about color, ethnicity, sex, religion, and so on. Is this a new tension around the concern of AI and technology with regard to those dimension, or is this something that has been there all along with other technologies, historically? You've studied this for a while.

ALONDRA NELSON: Yeah. Well, thank you for raising-- you raised, I think, two of the issues in the AI Bill of Rights. Certainly, well, it has five sort of prongs. And protections against algorithmic discrimination is one. And it's significant because it's a through line in the whole policy document. But it also is about the fundamental issue that AI tools and systems should be safe and effective.

I mean, that's just fundamental consumer standards. Looking at Professor Citron, they should have some modicum of data privacy. I mean, like shocking, that you should have notice when an AI tool or system is used for a consequential decision about your life. And that you should have some sort of fallback if a decision is made using an AI tool or system, and you have a question about how that decision was reached.

YAEL GRUSHKA-COCKAYNE: And an ability to opt out.

ALONDRA NELSON: And an ability to opt out. And those things were all distilled from almost a year of engagement that we did with academic researchers, industry researchers, folks in civil society, and just regular folks. I think the opportunity and challenge that artificial intelligence presents us-- let's make it take it down a level, let's say generative AI, is that it is a tool that I think brings together lots of different dynamics that makes it different from past technologies.

So it's dynamic, it's iterative. It uses historical data often. The data is not transparent. We don't have a lot of accountability around the tools and systems. People who make the systems tell us we don't understand them. We can come back to that. I call that algorithmic agnotology. And to me, it's kind of a learned ignorance that one wants to have around their systems that should be, I think, pushed back against. 

So it does present a different challenge for us. And I think some of the big challenges that we've seen immediately, we can talk about both near-term and sort of farther term risks and harms. But what we are seeing already are harms to people with dark skin, the use of facial recognition technology, who are being falsely identified, falsely arrested, falsely convicted in some instances, that can't be. That can't be how we want it. That's not how the society that we want to live in. 

We know that artificial intelligence use of generative AI, that makes it easier to do a cut and paste job and to create, basically, to do cyberstalking or sexual violence is becoming a challenge with generative AI.

So it increases, I think, concerns that we already have the sort of scale and velocity of doing them. So I think it's different from other technologies and that you don't worry about this with the toaster or the car. You might have worried about it with the computer, the introduction of the computer. And obviously, everyone, let's be very clear and not prudish about this. The very first technology innovation with any new technology is pornography.

I mean, when people are like the sex AI chatbots and all that. It's like, of course. Of course. Every sort of wave, if you could create pornography with it, that is what it's been. And so that's always been a special risk in those spaces. But I think this technology in particular presents those problems. I also think we learn more and know more. 

So I think in 1989 or '94, when we were all beginning to get the personal computers kind of trickling through our lives, I don't think we even knew how to think about what that might mean. And that transition was pretty slow, actually, if we think about it. I mean, it happened over the course of maybe a decade or eight years or something like that. 

We woke up on November 22 or whatever day it was, and chatbots were a consumer-facing product, and everyone in the world had access to it for free. And that was kind of overnight transition. And that makes all of these risks in some way more acute and make it a little bit different from, I think, prior kinds of technology paradigm shifts. 

YAEL GRUSHKA-COCKAYNE: Yeah. So maybe in the past, we've had an opportunity to think about it, a just, the academic world, that moves a little bit slower, had time to develop some thinking around it. And in this case, we don't have time because it's moving so fast. I'm going to pass it over.

MELODY BARNES: And I want to come back, given the conversation that we've been having, to something you said a few minutes ago about the AI Bill of Rights and our shared Democratic values and putting, these are my words, meat on those bones. 

And I'm curious about the struggle and the challenge of doing that at a moment when it feels as though we don't all share those same values that one group, one person's values may not look like another person's values and how you go about the process or when about the process of policymaking given that challenge.

ALONDRA NELSON: Yeah. The day one kind of a crew of us that were day one in the Obama-Harris administration came into a really crazy scene. I mean, I'll leave it for others and historians to write this book, but we did a presidential transition, not only with contestation and political violence. 

So when I first came to DC. Like every building in DC near the White House and near the Capitol was like double fenced. I mean, every store was closed because of the pandemic and everything was just fenced. Like it was unbelievable, actually. 

And we did the transition during the height of a pandemic. And so before the sort of White House Council, the National Security apparatus, had approved Zoom for use in the White House. So when I started in the Biden administration, we were still doing conference calls. And the only thing that we could use with video was Skype. For some security reason, I didn't know.

YAEL GRUSHKA-COCKAYNE: I thought you were going to say Webex. 

ALONDRA NELSON: No. No. We couldn't even use Webex. I mean, that's new. That's newfangled. You're getting crazy, Yael. I mean, it's crazy, crazy talk.

MELODY BARNES: I can empathize. I remember going in in '08 and '09. And we were in a big meeting and they were like, and you can't take your laptop home. I was like, what? We don't have the security for they got it. But yeah, I mean, technology comes late.

ALONDRA NELSON: We can take them home now but also, you can't use the Google Suite. Like they're just like oh, no. So we were trying to figure out, in that context, which was already different from any other, I think, transition context. 

And moreover, the question that we were facing, particularly in OSTP is like, how do you do science and technology policy in a moment where American trust and Democratic institutions and science and our ability to wrangle this pandemic is low-- low, low, low. 

And we came in, the leadership with a philosophy that was like, we're going to try to do it differently. So we had, by the time I left, science communicators on the staff for the first time, people whose expertise was to translate science to the public. 

So instead of OSTP, as it traditionally done, would pad out a policy document and we wonky people would write it. And if people could read it or not, shrug. And the job is really, we turned it back upon ourselves. And it's just like actually, our job, as people who work for the American public, is to make these documents clear to them. 

So part of what you see in the AI Bill of Rights is a commitment to that. What is clear communication to people beyond ourselves about the stakes of these issues? But I'll say about the process, so the AI Bill of Rights is a curious document because it's guidance. It's not formal policy. But we did a formal policy process. 

But we also engaged the public. So it is announced in an op-ed in wired that has an email address that goes to the White House at the bottom. So it was like, if you have anything to say about this, write to us at OSTP about this thing. 

We took a page from the FDA and did a series of just town halls. So if you've ever gone to an FDA hearing, you get a two-minute timer. And there's a facilitator and they say, anyone can speak. And so we did that. We had several of those. We did them different times of day. So people in different time zones could come. 

We had panels on topical issues like AI and health care AI and the workforce. And we had weekly standing meetings that all of the staff working on this workstream had to set aside in their calendar. And we just met with anybody who would want to meet with us, many of whom we met from this email address that was in this op-ed and wired. 

I mean, we met with high school students. We met with rabbis. We met with other kinds of clergy. We met with just regular folks. So the typical, I'm new to DC. And so it felt a little weird to me that you had the kind of big civil society organizations and the big lobby organizations. Those are the ones that you talk to. Like, is that how it works? Why? Why do we do that? So we really did have a broad swath of people that we engaged. 

And the AI Bill of Rights is really a distillation of all of those conversations. It's not breaking any new ground. It says nothing that I think any document, including a Trump administration document, a lot of that work was led by Michael Kratsios, who's quite good. 

People want their systems to be safe and effective. I mean, these are very, I think, common sense claims. We also do talk about issues around discrimination. We talk about vulnerable and marginalized communities. And that's very much the imprimatur of the Biden-Harris administration in thinking about these, but at a high level. 

I think that we felt comfortable by saying, you should know when AI is being used. It should be safe and effective. If it's being used for resume screening at a job or to give a health diagnostic tool, that you should have a fair shot at. You should not be discriminated against in the use of those tools if you're trying to rent a house or get a mortgage or get a job. 

And you should have some sort of fallback. And so I think we felt-- it took almost a year of that process. But we did try very hard to get to a place that felt that most people would, I think, find it quite common sense, and moreover obvious, and in many instances a kind of resuscitation of what they had said to us about what they thought should happen. 

MELODY BARNES: I want to ask a question. We've been talking a lot about AI, thinking about science and technology, generally, as they affect other aspects of people's lives. 

So thinking climate change, thinking public health, and how you see, over the course of the next decade or so, the use of science and technology evolving in the policymaking process to try and tackle those big issues, particularly at a moment where sometimes, there's a struggle with data, with facts, and what that means for policymakers as they are also listening to hearing from their constituents? 

ALONDRA NELSON: Yeah. That's a lot of hard, I think, issues kind of clustered in that really good question. I mean, I think one of the things that, at least in my time at OSTP and working in the White House, we were committed to is continuing, I think, from the Obama administration in particular, that started things like the Open Government Partnership, that started the first White House GitHub account. 

Hello, did you know the White House had a GitHub account? That's all the data nerds in here. There was an attempt to make data available to the public. And I think that we also wanted to do that and understanding that people might take that data and do crazy things with it. 

But you wanted to be able-- I think as a policymaker, you wanted to be able to say, we've provided you the facts. We can also provide some interpretation. But I think as we're trying to the extent that we can restore trust in the work of government, part of that is just giving the data to people. 

And you can't control how they interpret it. And all of us have been on social media. You see the diagrams with the strings of all the one White House memo and another, and leads to this kind of conspiracy. So you can't control that. But what you can control is a policymaker, and as a leader, is giving people high quality data. 

We also had support for that and things like Paul Ryan was a big sponsor of the Evidence Act, which President Trump signed into law, which also has other obligations for evidence-based policymaking, for data that's supposed to be provided to the public and the like.

So there's been, in the last decade, maybe not clear to I think people have not worked in government, a real sea change in how government thinks about its obligations around data to the broader public. And I think that was really important. 

I will say, the challenge that we-- the other challenge is that facts alone, or the science alone, don't solve the thing. And that's the work of policymaking. And that's the hard stuff. So coming into government during the pandemic, the science and the engineering was like miraculous on the level of the miracle. 

We had the genome of SARS-CoV-2 decoded in less than a month. We had an operational viable vaccine in less than a year. I mean, that has never happened in the history of the world. But then we had all sorts of challenges. So how do you keep it cold enough? Then there were all this kind of infrastructure challenges. And then how do you get people to take it? 

And those are questions of social science, behavioral science. Those are questions of whether or not people trust the government, whether or not they trust the research because it was done so quickly. I mean, part of why we engage folks with science expert communication expertise is that it was clear that government had to be a lot better at saying things like, we got it done more quickly because we did all of these other X things more quickly than before, not because we cut any corners or like risking people's health to get this virus done very quickly. 

So it's just a kind of different philosophy. And then the other thing I would add, I think, the amazing work that you did in government and the Domestic Policy Council, I think all of that work are now science and technology policy issues.

So not just climate change, but how we're thinking about DHS and immigration, which is using the CBP One app, that you have to have a smartphone to be able to use, which probably many refugees and asylees don't have, if they have a phone at all, and requires that you have very good lighting, and if you're a dark-skinned person seeking asylum. All of those technical questions.

Ditto, the health portfolio, ditto, the education portfolio. So in retrospect, I think my interest also in the Obama OSTP was that I think it was a real kind of awakening that all significant domestic and international policy issues were also science and technology policy issues. And I think that really is where we are now and where we're going to be sort of heading. 

MELODY BARNES: Yeah. And in listening to you and your bio, this next question, and your reflection on the fact that science is not just a collection of facts, but a deeply social process. And your background and interest, also in the humanities and that combination of the sciences, and I'm wondering if you could talk a little bit more about that, about the social process. 

ALONDRA NELSON: Sure. I mean, let me talk-- maybe, offer a couple of examples. And we can think about gender. And all of this are just as a more granular way. But you should think of it as a metaphor for training data and AI systems. And we can talk about those two, but I want to be less abstract before getting more abstract and talking about AI. 

For example, if you think about clinical research that we do and sort of protocols that we have for clinical research and for creating new drugs, for example, we've had, traditionally, pretty much up to the present, all clinical research subjects have almost been exclusively male. 

And we've created whole drug ecosystem and diagnostic ecosystem around men and male biology. And so we can put-- we can problematize men and women, all of that. We can have that conversation. But that means that we have created things that we say work for all without actually probing that as an empirical question. 

So those are early design choices, in part, made by the scientists who were working-- like, I'm going to test it on my guy, and see if it works or whatever. We think about 18th century science or something, and see if it works. 

So that's kind of one example in that if we think about science or biomedical science or something, not just being the drug, but that process through which you achieved it, then if we have pharmaceuticals that work for some people and not for others, we shouldn't be surprised if you're able to think that through. 

On the kind of engineering side, if you think about, our crash-test dummies, for example, some of you might know this data, are men, or it's like male physiology. And there was a kind of headline two or three years ago. It was like Swedish scientists create a woman crash-test dummy. 

And so you think about that design choice and that somehow, that was supposed to sort of sit-in for everybody else. And obviously, there's all sorts of reasons why that doesn't work. And you can take the male crash-test dummy, make it smaller or whatever, but it's still not quite kind of mainstream or normative, or whatever phrase you want to use, kind of woman's physiology. 

I share those as more like obvious examples. But if you take it to the space of AI and you think about historical training data sets around employment or around housing. Computer science, if you're going to do a higher end computer science, who traditionally in the data set?

So if you look at all the resumes of all the successful computer scientists in the history of the world up until 2010, what is all that data tell you? It tells you they're male, tells you they went to Caltech or Berkeley. There's a few data points. 

So that kind of data is being used, we know for certain in like resume screening in 2024, depending on the company, there are companies that are much-- I'm a big fan of Indeed and their CEO, who's actually very engaged in these conversations about bias and the training data. 

But we're pulling all of this kind of historical constraints and limitations in our design choices around how we're doing AI in the contemporary moment. And so part our mission, if we want to do it better, and do it in a way that benefits more people, is to be willing to ask those social questions, those philosophical questions about how we're getting to the data and the decisions that we're making using these tools and systems, whatever that is. 

MELODY BARNES: That's great. Thank you.

YAEL GRUSHKA-COCKAYNE: We're here with the Futures Initiative. And a few weeks ago, we had a talk in the same capacity, related to the future of higher Ed, a very stimulating talk. And obviously, AI and generative AI and STEM, more broadly, kind of play a key role in the future of higher Ed. 

In your mind, what are some risks and some challenges, and maybe even some hopes related to how AI affects higher education?

ALONDRA NELSON: Yes. This is a fascinating space to talk about these issues. I mean, first of all, I have a wish that I have said to Sam Altman. So he's well aware of that, that they had actually taken-- that they were not racing to market and they had taken another day, another week, another month to talk to teachers about this tool before they released it. 

The freak out that happened with schools banning it and penalizing students and analyzing their papers with bad AI detection tools that do not work. I mean, like this whole thing just did not have to happen. I think even if you had said-- even if you had given them a little bit of a heads up.

So I think whatever market incentive or market desire was behind that, that's just irresponsible. And I think that we need to be able to say that. We could have imagined a rollout of ChatGPT even a month later. I mean, people had seen ChatGPT 3. We'd seen 3.5 and 2. It wasn't like this thing just came into the world and people didn't know that there were increasingly capable tools. 

But you could have imagined a partnership with teachers, where they had different kinds of tools, teaching modules, like things that brought it into the world in a way that was less, I think, adversarial and less traumatic, I think, for the classroom.

YAEL GRUSHKA-COCKAYNE: Or at least wait until after winter break. 

ALONDRA NELSON: There's that. Right. Exactly. Let the parents deal with it. Go back home. So let me just say that at the beginning, I mean, I think we have to teach differently. And I can say this with some liberty because I don't have students right now at the Institute. And I think that's OK. 

I mean, I think one of the things I've been interested about around in the last couple of weeks is Google's introduction of NotebookLM, which has all sorts of problems, including that it hallucinates. So if you're using that, don't think it's giving you-- you need it for something factual, do not use it for something factual. Because there is a degree of hallucination.

But I think one of the takeaways of how people have been responding to that is that we learn differently. And that we are-- I mean, go back to biblical times, where oral cultures. So should we be surprised that students are like, oh, my God, this essay that I'm supposed to read, and it's supposed to have some virtue, my teacher said it's supposed to be more virtuous to read it, has come to life for me with this bizarre chatbot podcast thing, which is how I learn or how I take in information, ditto video and YouTube and TikTok, and all the things that people who are younger than us are much better at and much more interested in. 

I think we need to be open to those kinds of conversations. And also, remember that all of our cultures start as oral culture. I mean, the reason we know anything about anything is because we used to talk about it. And that's how people took in information. 

And that should be OK. I also think we don't know exactly how these tools can, will be used. I mean, we're still to figure it out to the extent that they should be used. And we can talk about that. I mean, I have concerns, particularly around K to 12 education, the tracking and the data surveillance, particularly for young people in schools around that that we should, I think, really be wary about. 

But I think also, students that are going to show us how they're useful. And we can call that plagiarism or we can call it something else and give them other ways of learning and other sorts of tools in the classroom, other kinds of assignments that are different, and that incorporate the fact that this technology exists in their lives. And they're going to use it. 

YAEL GRUSHKA-COCKAYNE: So that's very optimistic and that's the hope and the vision. Do you have any concerns, or are you-- do you want to some risks that are important--