Latest / Quantum Potential / SPECIAL EPISODE: AI, Propaganda and Democracy: Inside a Groundbreaking Discovery by Vanderbilt researchers
Transcript
- 0:00I think individuals need to be vigilant about what it is that they're putting into the public domain and vigilant about what it is that they're consuming.
- 0:07Like, there's a saying,"If you're using something for free, you're the product." What does it mean to give away all of your information?
- 0:15Artificial intelligence is transforming our world in ways we can't have imagined a decade ago.
- 0:21But what happens when these tools are weaponized— not to solve problems, but to manipulate public opinion in ways that threaten democracy?
- 0:29Today, I'm joined by Vanderbilt researchers Brett Benson and Brett Goldstein, whose groundbreaking discovery uncovers just how serious this challenge is.
- 0:50Welcome to 'Quantum Potential,' a podcast dedicated to curiosity and discovery.
- 0:55My name is Cybele Raver, and I serve as provost of Vanderbilt University.
- 0:59As a leader in higher education and as a social scientist, I believe that we can transform higher education for the 21st century.
- 1:06At my core, I am wildly curious about other people's work, whether it's in science, the social sciences, the humanities or the arts.
- 1:14This podcast gives us the opportunity to break open new ideas through discovery and creative expression.
- 1:22Before we dive in, let me introduce our incredible guests, and I'm so grateful to both of you for being here today.
- 1:29Brett Goldstein has one of the wildest resumes you'll ever hear.
- 1:32He helped build OpenTable into a global powerhouse, became Chicago's first-ever chief data officer led digital innovation at the Department of Defense, and now spends his time helping governments and organizations understand the power and the risks of AI.
- 1:48Brett Benson is one of our nation's most renowned political scientists, who uses AI to model some of the world's most complex security challenges, especially in East Asia.
- 1:58His work dives deep into China- Taiwan dynamics, shifting global alliances and how technology shapes international relations.
- 2:06And the best part?
- 2:07Both Bretts are professors right here at Vanderbilt.
- 2:10But most importantly, they bring their extraordinary expertise to the next generation of researchers and innovators.
- 2:16Before we dive in, could you both just take us to that moment when you first broke open this story, this case, this, this moment?
- 2:24What went through your minds?
- 2:26Well, it all started with a text message.
- 2:30So I got a text message from a researcher friend, and it included a link, and it was a dump of documents.
- 2:38It was a series of PDFs, roughly 400 pages, and I started to comb through it.
- 2:45It was mostly in Mandarin, but included technical specs, pictures of U.S.
- 2:51congresspeople, and I knew something wasn't right.
- 2:55And I did the basic research and said,"There's something super interesting here.
- 3:00So I better call my friend Brett." He said, "There's something that I want you to take a look at." And, he said, "I sent you the, the link to this secure site" where you dumped everything.
- 3:11So I looked at it like the following day, all in Mandarin, 399 pages.
- 3:16399 pages of Mandarin is like 600 pages translated to English.
- 3:22And so I have, I have some language skills in Mandarin, so I started to poke through it that evening.
- 3:28And I was pretty shocked at what I was reading, and also really, really skeptical.
- 3:32I just thought,"You know, there's there's no way that this is what it purports to be." And, and, and then as we started to go through a little bit more, you know, I was still very skeptical, but I was pretty, pretty blown away at what it was that GoLaxy claimed to be and what they claim to be able to do. Incredible.
- 3:53And so, I was really suspicious until, until I read one of the particular documents that, was really well put together.
- 4:02And at that point, and that was maybe two weeks later, at that point I had that surge of both excitement but also dread.
- 4:10Yeah, intense.
- 4:11So for those who haven't read the New York Times essay, can you give us the short version, the, you know, film trailer version of what you found and why it's so significant?
- 4:23Why don't I give the the quick tech...
- 4:25Yeah.
- 4:25...and then you can talk the bigger picture.
- 4:28So imagine a system that collects a massive OSINT.
- 4:32So that's open source intelligence.
- 4:34So for example, if we're talking about you, if you have a YouTube account, Facebook, Twitter, any of those things, the system goes out and it collects everything that's on the internet. Yeah.
- 4:45I just know that Spotify can publish your playlist.
- 4:49All of these things, everything there is to know about Cybele is collected into a system with the intent of crafting a message to persuade you to do something.
- 5:01So this is a bit of a jump from what we've seen in the past where you would create cohorts.
- 5:08So what is a typical cohort for Cybele?
- 5:12And it jumped from the cohort to the individual tailoring with intent to persuade at an enormous automated scale.
- 5:22Yeah, so the documents— there are documents that are about a company in China called GoLaxy.
- 5:28It was a leak by a disgruntled employee.
- 5:31And GoLaxy is a company.
- 5:33They have a website.
- 5:34And on their website, they claim to be an AI driven data analysis company that ostensibly does data collection, big data collection for companies and other clients for the purpose of gathering information on public sentiment.
- 5:51Which is a huge marketing tool.
- 5:53Right.
- 5:54And and so on their website, they're claiming to be the number one company in China, or they aspire to be the number one company in China that does this.
- 6:01Political candidates who want to determine how to craft political messages and campaign ads.
- 6:06This would be this would be a real effective technology to gain a market edge and to respond to market demand.
- 6:12And so on their website, their, their website presence, is about delivering this type of good and service to their clients.
- 6:21The documents make them out to be a bit more than that.
- 6:24In the documents they are... you know, they claim to be founded by the Chinese Academy of Sciences, which is the top national research academy in the in the Chinese government, run by the State Council, which is the premier decision- making body in the Chinese government.
- 6:41And they also claim to have links to and and have state-aligned goals for the purpose of shaping ideology and shaping messaging and basically to deliver propaganda for the Chinese government.
- 6:55And that's the part that really shocked me.
- 6:57And the and the documents are just full of this information about how they do it, how they have done it, a claim to have done it effectively in Hong Kong and Taiwan.
- 7:09And so using the technology that Brett Goldstein just described, they, they claim to have these state- aligned goals to manipulate public opinion in places like Taiwan, Hong Kong, but also now the United States.
- 7:22Incredible.
- 7:23And so that was the, images the, the rather that was the lead that you had that they were pursuing U.S.
- 7:31congressional leaders or other, key opinion leaders in the United States— that this was not simply historic data on something that might have happened two or three years ago in Taiwan or, in Hong Kong, but actually was could potentially happen in the future here.
- 7:47So it gave a lot of history, but then it gave a look ahead.
- 7:52So looking at the West and looking at the role this could play and looking at the congresspeople.
- 7:59But then when you looked at their website, we started to get a glimpse of what they were doing.
- 8:04So we had years and years of history of more traditional AI neural networks.
- 8:09But on the website they talked about integration with DeepSeek.
- 8:13And obviously DeepSeek has had a lot of news, and we saw their move into gen AI, which gave a lot of signal for intent.
- 8:21But there's one little...
- 8:23it's a little technical, but there's a nuance we found in the docs where it talked about protecting personas.
- 8:30So they would create accounts, you know, just an imaginary synthetic account at John Doe, say.
- 8:37And they talked about how to protect it, so won't be burned down by Twitter or Facebook or whomever.
- 8:45And they talk about it at length where it, "Let's make sure it's hidden and protected and seems real." It is atypical to see in a business intelligence context, a company saying we're going to make personas that we're going to protect from being turned off.
- 9:06And it was that intent in there that told you there's more to this story.
- 9:11And so this sense that this was not and I mean, obviously it would have been big news if it had been a more, you know, mass- produced chat bot that might be focusing on misinformation or, persuasion.
- 9:25But this idea that these specific personas would be crafted for, for specific individuals, whether in China or in the U.S., that would be basically undetectable or could evade, standard security protocols.
- 9:39Well, you're exactly right.
- 9:41Because when you say chat bot, you know it's a chat bot. Right.
- 9:45But there's an intent here to create a persona where I don't know I'm interacting with an AI.
- 9:53So, because of that, you enter this murkiness of what is real and what is not real.
- 10:00And and that both shows intent and what's coming.
- 10:04So incredible. Brett.
- 10:06And you've said, actually the other Brett, Brett Benson, you said this discovery proves the limits on propaganda no longer exist, right?
- 10:15This idea that that we can't detect what's propaganda and what's fact or what's true.
- 10:22So can you say a little bit more about this idea that propaganda no longer exists. Yeah.
- 10:27So I don't have Brett Goldstein's technical background, so our conversations together have been really helpful to help me put this into context.
- 10:36In my research, I have looked at propaganda in China, you know, going back to the 1950s.
- 10:43And, China has been engaged in sort of ground- level propaganda messaging in, in recent years and been collecting a lot of data on citizens.
- 10:53Even so, I had no idea what AI-driven propaganda could do.
- 10:58It just hadn't really clicked and registered in my head.
- 11:00And so these conversations with Brett have sort of helped me put that into perspective now.
- 11:06And can I just ask, you know, when I think about that, I think about print media or I think about a message.
- 11:12It's very hard for me to imagine that message coming from an individual because I think of individuals as trustworthy, right?
- 11:19So suddenly, does that mean that if it's not happening in person, our conversations are corruptible or, you know, essentially, like, you know, this is the only secure...
- 11:30Yeah.
- 11:31...mode, is the analog mode, right?
- 11:32What does that mean in terms of trusting information if I were to get an email from a colleague?
- 11:38How do I know that that email isn't a very carefully individualized form of propaganda sent as if it were that colleague.
- 11:46Let me put this in context of something that I do understand because I had to I had to I had to really think through what the scale and capability of this new type of AI-driven propaganda could do.
- 11:56So years ago, I had, I had a, a project with a colleague in China, and we were studying—it stopped since because of obvious reasons.
- 12:05But we were studying the effectiveness of Chinese propaganda on public opinion and on beliefs and just among regular Chinese.
- 12:13So we had a we filled in a survey and, and we had collected some data.
- 12:17And what we knew about Chinese propaganda at that time is that it was really effective, but mostly effective among a vulnerable population— a population that even though they knew it was propaganda, they didn't have opportunity to check it against other sources of information.
- 12:35And so my takeaway from that project is that propaganda can be effective, but it's limited because first of all, it's pretty expensive.
- 12:42It's hard to, to deliver.
- 12:44And most of the recipients are pretty suspicious.
- 12:48Even in the 2016 election here, most people who were targeted by Russian propaganda and messaging on, you know, on social media, a lot of people were pretty suspicious.
- 12:58They could identify it.
- 13:00This is completely different because it's increasingly not very expensive.
- 13:04It's easy to deliver.
- 13:06And then the adaptability of the personas means that there isn't just a segmented, vulnerable population.
- 13:12I'm vulnerable.
- 13:13Everybody's vulnerable.
- 13:14Because even if you're not vulnerable to the persona itself, the persona can infiltrate a network, and then the message can be spread by individuals in that, in that network. Right, and so it's to very high-influence individuals as well as to a random person on the street.
- 13:31It's not just my power as a voter, it's my power as a leader or your power as a leader, or any of us who make decisions on behalf of institutions.
- 13:39It's incredible.
- 13:41It's it's gotten much easier.
- 13:43So my example earlier where we create cohorts, right?
- 13:48Say I'm targeting Benson, a middle-aged white male, right?
- 13:52You create a cohort, and then you hope you hit it.
- 13:55And hopefully he clicks a button.
- 13:58That's old news now.
- 14:00So we did an experiment a year ago where we pull LinkedIn profile info and create a message tailored for someone, and it knew enough about the person using gen AI to get them to click.
- 14:12Like you're talking like 70 to 80% click through. Just LinkedIn.
- 14:18Now with these amazing tools we've created, you can create individual persuasive messaging for each individual, which honestly, you're going to click and you're going to engage.
- 14:33That's how personalized it is at this point.
- 14:35And it's technically easy.
- 14:37You don't need a big data center.
- 14:39Like I could pull out my laptop, bang it out now and have it done in a couple hours.
- 14:44And then it adapts, right?
- 14:46So it adapts with the interactions in a much the same way that TikTok adapts. To the extent that users are engaging with the persona, it's adapt...
- 14:54I mean, like ChatGPT, you can interact with it directly.
- 14:57I mean, that's just so fascinating, right?
- 14:59Because I think we like that portion of the adaptability or the adaptation, the capability of the machine to understand our tastes and preferences and to give us more of what we recently liked, right?
- 15:11Can you say a little bit about the way in which we understand this as dangerous to democracy?
- 15:16Well, I think you'll get another two-part answer.
- 15:20So the challenge here is we have this amazing innovation, yet we don't know how to detect it.
- 15:28And we're in this pattern we've seen before, where, like, anthropic and OpenAI are doing and a bunch of companies are doing amazing things.
- 15:37And every week there's a new advance in this.
- 15:41But right now, if I were to generate a short-form AI-generated target toward Benson, we can't, we don't know if it's an AI.
- 15:52Like it's now gotten beyond that... the detection capabilities.
- 15:56So what does democracy look like when you can't tell the difference between AI and human in a world which is very much shaped by social media, Twitter, all these different pieces.
- 16:10At a time when we are no longer, or going less regularly to political forums.
- 16:15I mean, it immediately makes me think of the value then of universities or the value of town halls, or the values of those moments of civic engagement where you actually are standing in present in un...
- 16:27you know, incorruptible analog moments because of the way in which you at least know that the people on stage are saying the things that they're saying.
- 16:34Well, you've made a really good point twice now, on the analog side.
- 16:38Because, okay, we know we're having this conversation.
- 16:42Like, really high probability this is real.
- 16:45We do this in a digital way, it's trivial to change that.
- 16:52And we're charging ahead, and what are the mechanisms in place for any of us to know that we're really having a conversation.
- 17:02And the fact that we don't...
- 17:03we can't come up with a bunch of ideas of what's being done is a little scary.
- 17:08Yeah. Okay.
- 17:10So that's ...
- 17:11we're going to talk about the Institute of National Security and the ways in which both computer science and political science can address that.
- 17:18But Brett Benson, just on that point, you know, the the moments where U.S.
- 17:24elections or foreign elections have been thrown, can you give us a sense of the magnitude of the power of misinformation, or disinformation, in an election system or in a political system? Yeah. So maybe two questions there.
- 17:40One is in the election system itself, but the other is in a political system that is a democracy.
- 17:45I have two concerns.
- 17:47One is the effect on elections.
- 17:49An overwhelming percentage of people and voters in this country and in other democratic countries get their news from social media.
- 17:56And I think that that's increasingly the case with younger people.
- 18:00And and so it doesn't take, you know, it doesn't take a lot to think through the implications of, polluted ... polluted and corrupted information in social media targeted to particular individuals and networks and the potential effect it can have on elections.
- 18:16There are attempts in the works to try to shape public opinion in advance of elections.
- 18:22And one of the concerns that I have with respect to elections is not just that that really effective, AI-driven propaganda strategies can shape public opinion, but if a lot of such organizations get into this space and it creates chaos, I start to worry about civic disengagement because people don't know what they can trust, or they just go to their corners.
- 18:43That's already happening in our country, and they go to their corners and they only listen to siloed information.
- 18:50And you don't sort of get this cross-fertilization of information that I think you need for a healthy democracy.
- 18:55Yeah.
- 18:56Interesting intangibles that are incredibly important— trust, morale, faith in the institutions that we believe are robust, like voting. Yeah.
- 19:07But democracies are vulnerable on the national security front, too, because so much of what we do in a democracy when it comes to making national security decisions depends on public opinion.
- 19:19And also, you know, it depends on on the validity of information that's coming into the system, which also could be manipulated for decision-makers.
- 19:28And so I'm concerned about the effect of AI-driven propaganda on public resolve when it comes to big security challenges, when it comes to manipulating decision-makers by corrupting information, polluting information when they are making important national security decisions.
- 19:48I worry about deterrence.
- 19:49These are some of the other concerns that I have that are just separate from the concern that I have about elections.
- 19:54Yeah, you've you've totally freaked me out.
- 19:57I'm now very much clear about the significance of the problem.
- 20:01I'd love to learn a little bit more about the solution, because I think that's what universities can really be about, right?
- 20:08Not just lifting up the problem, but actually really, really moving quickly towards an accelerated set of feasible solutions.
- 20:15Tell me a little bit about, the ways in which you had this opportunity to do this work here at Vanderbilt.
- 20:22Can you say a little bit about how Discovery Vanderbilt funding made this possible, and how you were able to move swiftly to be able to really identify this is a key issue that we should work on.
- 20:33So one of the things that was amazing about this was the partnership.
- 20:39Like often we operate in silos, and I should say in my computer science bucket, and Bentsen should be in his poli sci world.
- 20:48And I think through Discovery Vanderbilt, it encourages this radical collaboration, where I like this happened at a speed that I don't know that we see in academia.
- 21:02I got a link, I made it safe.
- 21:04I realized we had something.
- 21:07I pulled in Benson, and we surged and got through 399 pages of fun, got it out to the New York Times, and this all happened in a few months.
- 21:20And now we're briefing Congress.
- 21:23We're working on a longer journal piece, and then a body of research on how do we actually get at the problem.
- 21:30And this is a few months out.
- 21:32Yeah. Yeah.
- 21:33I don't think that Goldstein and I would have met had it not been for the opportunities that are provided here, because we live in two different worlds.
- 21:42And, but the collaboration has been pretty exciting.
- 21:46So I love this example of the two of you meeting in that moment and then figuring out how to work together.
- 21:52Was there any, skepticism like, "Wait, you don't get my...
- 21:55you don't get my world?" No.
- 21:58It just kind of it worked, both technically and from the poli sci way.
- 22:03The fact that I came from the ops side and Bentsen came from the academic side, and I was able to bring some of my, my sort of national security operational knowledge and Bentsen giving context. Like it brought all the pieces together.
- 22:24And it also, what's really nice is it's really grounded, like the work we're doing.
- 22:29Yeah. It matters.
- 22:30It matters to Congress.
- 22:31It matters to the Nat Sec community.
- 22:34It matters like, honestly, we're at the front of something that is going to be a wicked problem coming to the national security community.
- 22:43This partnership brought together the pieces to allow us to operate at a different speed.
- 22:49Especially on this particular project, there's a role for for public actors in the public space, in government to develop solutions.
- 22:59There's certainly a role for people in the private sector, in social media platforms.
- 23:05But the individuals who have an opportunity to work on this who have the most credibility are researchers because we're not tied to a commercial agenda.
- 23:14We don't have a political dog in the fight.
- 23:18We're just trying to solve the problem. Yeah.
- 23:20So establishing that you're the real deal and establishing that you are politically neutral, you're neutral in terms of products and commercialization.
- 23:28You're neutral in terms of any special interest.
- 23:31Tell me a little bit about what it was like as you thought about giving congressional testimony and the, the direction and guidance that you could provide, you know, national public figures.
- 23:42So it was ... we were surprised at how many congressional requests we got.
- 23:49And it was from multiple committees, multiple jurisdictions.
- 23:54And and it was both giving foundational education, but also a little crystal ball on what is to come.
- 24:03Yeah.
- 24:04And doing it reinforced the importance of the work, because having the back and forth with these folks who are insanely busy, who have to face all these challenges and there was a lot of like the "oh crap," like, "this is a big problem." And it, it, it reinforced that academia needs to play a role here because a traditional reactive response to this problem set?
- 24:32That won't go well.
- 24:33We're always reacting to a problem and not actually getting ahead of the threat.
- 24:39In this case, what happens if we're flooded with one-on-one models that understand us better than we do, and we're constantly trying to figure out, how do I identify them?
- 24:53That isn't going to work.
- 24:55We're going to be in a very complicated world where things are trying to influence us in ways that are probably not great.
- 25:02So academia needs to figure out, and in the Nat Sec community, we call it"left of boom," getting ahead of the big incident.
- 25:13Left of boom. Yep.
- 25:14That's that's, the the Monday morning fun here.
- 25:18Is that a sailing metaphor where the booms can swing around and hit you in the head if, say, you tack?
- 25:23So that would be a good one.
- 25:25That's more gentle than the kinetic.
- 25:28Kaboom. Okay. Yeah.
- 25:31So this is one where we if you have the AI companies moving at this amazing speed, and we don't want to stop them.
- 25:41But if the landscape continues to change, what is the academic innovation that helps us identify AI versus human in a sustainable way, where everyone wins so that we can understand who we're talking to and who's talking to us?
- 25:59And I think academia is really the one to solve that proposition.
- 26:03Yeah.
- 26:04And I'm curious about what it was like to speak with, you know, U.S. political figures or congressional leaders and have to tell them that they themselves are being targeted, whether individually or collectively, right?
- 26:18So we talked to a lot of committees, and the first question that all of them had was,"Are we on the list?" Yeah, they all wanted to know.
- 26:25Yeah.
- 26:26And, we released the documents so they can go check to verify that for themselves.
- 26:31We don't really know the the extent to which people are on the list, because in the documents it was just a snapshot to show, like proof of concept.
- 26:39Yeah. You could be on the list.
- 26:41You could be on the list.
- 26:42And it was also three years ago, and the way that this technology has accelerated, the pace, the list is...
- 26:50dwarfs whatever we saw on the document by now Yeah.
- 26:53It's pretty reasonable at this point to think everyone's on the list.
- 26:56I mean, so that within those five minutes, in providing that insight to these congressional groups, the key that you highlighted earlier is be prepared, be ready.
- 27:10What else did you feel that you could say to them in terms of policy solutions?
- 27:14They're asking for policy solutions after they get over the shock and dread of, being potentially their offices, you know, themselves being targeted.
- 27:23It's hard to know in a, in a 30-minute conversation what intelligent questions to start to ask about crafting policy. Yeah.
- 27:31So they asked for some suggestions, but the biggest thing is that they really want to know what the next step is.
- 27:37And in having a conversation, who should be involved in that conversation, what the topic of that conversation should be.
- 27:44Should it be about detection?
- 27:45Should it be about other types of things?
- 27:48And so a lot of the conversations, at least my impression was, we don't really know what to do next.
- 27:53Maybe we circle back and have a roundtable.
- 27:56Which is really precisely what the national summit that happens here at Vanderbilt every year is really about, right?
- 28:01The chance to get those and get those voices all at the table there.
- 28:06There's an education piece that is required here.
- 28:09It's it's a complicated topic.
- 28:11We will, as a teaser, be talking about this at summit...
- 28:15Great. ...this year.
- 28:18I think it behooves us to help Congress help whomever get smarter on it.
- 28:25But that does have to parlay into action.
- 28:28And I think there's really a three- legged stool here— private sector has a role, government has a role and academia has a really big role.
- 28:36My fear is if we don't get those pieces in place, it's going to get complicated.
- 28:43And say more, Brett Goldstein, about the ways in which different teams or different interdisciplinary groups might be really leveraged there. You know, is it that, I should triple the investment we're making in our new College of Connected Computing?
- 29:00Do we need computer scientists, or would your colleague Brett Benson be lobbying for a lot more funding for political scientists?
- 29:07Or do we need people in completely different areas, like, you know, folks in the arts or folks in the humanities or psychology who understand the power of the message in communication?
- 29:17Where do we source the talent to come up with a solution?
- 29:21I think she directed that to you because she already knows what I'm going to say. Yeah, I think so.
- 29:26I'm sure Benson needs lots of help.
- 29:29So I think it's like right now we're interdisciplinary.
- 29:33I think we need to be more interdisciplinary.
- 29:36And so Benson and I are working on a project now where we work on two parts.
- 29:42That's three parts.
- 29:42One is creating short-form AI messages, the very best we can come up with as a sample. Two, detection methods.
- 29:56And then three understanding networks, so if you find a node that is a potential synthetic, synthetic, message, then you can then trace it back and maybe learn a little more about the network.
- 30:13But if you're operating with, say, 200 characters, that's just not a normal computer science problem.
- 30:21Like what are the right disciplines to bring in here to get at it?
- 30:27But that is a concise problem that eventually gov, Congress, everyone needs us to solve.
- 30:34But it brings in all of those different pieces.
- 30:36Yeah.
- 30:36So I hear you saying that detection is the first solution, right?
- 30:40Detect detection and tracing back to be able to identify the source and to shut it down or just to turn the spigot down, right?
- 30:47So, you know, if someone's breaking into your home, there are many things you could do.
- 30:52But the first thing is to make sure that the locks are strong, right?
- 30:54Well, I think part of it is we don't have a great sample of what is AI-created content out there.
- 31:03So it's detection, but we're also taking the stance that we need to create the best because there's a sampling issue.
- 31:11Because who knows what's out there.
- 31:14Yeah.
- 31:15So it's a weird role of creating this potential content, detecting it, but then where it gets even harder is what happens when your creation methods change every week.
- 31:29Imagine when you have hundreds of different models created.
- 31:32I can then create all of these artificially intelligent personas.
- 31:38From there they change every few days, and I need to be able to detect them.
- 31:44Yeah.
- 31:45Constant morphing.
- 31:46Constant morphing, and if you get behind in the wrong way...
- 31:49It makes me think of the Philip K. Dick novel, where he's ... where, where the protagonist is wearing the scramble suit, right?
- 31:57Where you can appear to be anything to anybody. Right.
- 32:00And that, from a national security perspective, when you have a gap in detection, that's an enormous window of weakness.
- 32:11And we have to figure out how to not have that gap.
- 32:16And no one knows how to solve this right now. Fascinating.
- 32:19So it's a computer science problem for sure, and then Brett's like queued up to tell me how it's all these other disciplines.
- 32:26Well, so so the reason why we're working on this project together is because we have...
- 32:34we're approaching it from two different angles.
- 32:36So Brett Goldstein wants to develop the technology to detect, and what I'm interested in is being able to trace the source and then the impact.
- 32:48I want to know how effective these messages are.
- 32:50We can't determine how effective they are until we can detect them.
- 32:54If they're not that effective, then the problem is not so much about influence as it is about creating information chaos and and then decreasing trust.
- 33:04And that would be the overall effect on democracy.
- 33:07So there are multiple different angles here when it comes to detection.
- 33:14From my perspective, catching one of these in the wild, or a couple of them, enables us to start to analyze the network.
- 33:21And then we can trace it back to the source, because another problem, I'm frankly very suspicious about our long-term ability to engage in systematic detection.
- 33:30But if we can figure out the source, then we have, you know, we might have the ability to frustrate some of this at the, you know, at the spigot. Yeah.
- 33:42But the other the other thing I, I'm hoping that we can do is once we can detect a handful of personas, then we might be able to trace the effectiveness.
- 33:52And we're thinking about through ways that we might be able to do that, to determine how much these types of messages, you know, the extent to which they spread, the extent to which they actually might change public opinion.
- 34:02And we're starting to think about ways that we can do that.
- 34:05But detection is just one side.
- 34:08There are other things that, you know, there are other policy solutions that we need to be thinking about.
- 34:13We need to be thinking about hardening the types of information that populate our AI models.
- 34:21Because if, if our AI models are corrupted at the source, then that's another place where, where information can be manipulated.
- 34:30Increasingly, national security decisions are reliant on AI models.
- 34:35And the hope, the long-term hope is that AI was supposed to increase deterrence and and peace because it gives us more information, it gives our adversaries more information.
- 34:44It decreases the chances that we might have conflicts through misperception.
- 34:48But if all of this is subject to misperception and and information manipulation, then that kind of undoes the hope of... Totally.
- 34:54You're going to go back to handwritten notes, right? Yeah.
- 34:57So your point about analog is a real good one.
- 35:00Maybe one of the maybe one of the solutions here is that we start to think a little bit more about interpersonal interactions that we can trust and that we know, but then, you know, AI's here.
- 35:12It's not going to stop.
- 35:13And what does political science have to say about galvanizing or mobilizing public trust?
- 35:19You know, what are the key factors that lead to an increase rather than a decrease in trust?
- 35:25Certainly, you know, if I sit in the stands of a Vanderbilt football game, I feel a massive surge of pride, engagement, belonging, trust, right?
- 35:35But, you know, what are the key features that would be activated in addition to deterrence?
- 35:42Yeah, this is a hard question.
- 35:43And some of my colleagues in the political science department are better equipped to answer this particular question.
- 35:48Already we're suspicious of public opinion data, and public opinion increasingly is manipulated.
- 35:53Then we'll become more suspicious.
- 35:55To increase trust— I can only tell you anecdotally because it's really not my area—but, we have to start to put more investment into interpersonal interactions.
- 36:07In an era when when politicians are becoming more disconnected from people and where more interactions are digital, it seems to be going in the opposite direction.
- 36:17Yeah, stealing, standing at the train stop and shaking everybody's hand, right?
- 36:22There's this, means by which working, working the crowd, individual by individual, at least, is is uncorrupted in that regard. Yeah.
- 36:31Be fascinating to hear that as a next step in this conversation.
- 36:37You know, as you have those radically collaborative, engagements in discussions and roundtables.
- 36:43You know, certainly, I think a lot about, where you want us to go as national leaders in this conversation, and you've both called for that kind of urgent collaboration.
- 36:56What gives each of you hope?
- 36:57Brett Goldstein, what gives you hope?
- 37:01Oh, that's a hard question.
- 37:03I know. Hope.
- 37:04We've just spent such a long time talking about what places dread in our hearts, but what places hope? So so I, I do have hope in this window, like AI is doing amazing things.
- 37:20There is a immediate recognition when we published on this that something had to be done.
- 37:27And I'm seeing us now move at a speed which I haven't seen before.
- 37:34There's speed, motivation, support.
- 37:38And there is a way. And getting back to one of your earlier questions— I haven't heard from any of the Congress folks, students, faculty, saying"Oh, we need to slow AI down." No one's saying that.
- 37:51Instead, we all need to work to a solution.
- 37:54And in my experience, when you have the right people at the table with the right motivation, we can get at it.
- 38:02And I think we're moving at speed.
- 38:04Yeah, and it's up to us.
- 38:06I love the "get at it" attitude, right?
- 38:09That with, you know, enough recognition of the challenge, we can solve anything in a very fast-paced and ingenious way.
- 38:18And something I love about Vandy where in, in my prior academic life it was a little more focused on you published, and it was great, and we had Nobel Prizes, and love it.
- 38:32But here there's that support for"and we're going to make a real solution." Yeah.
- 38:37"And we're going to have impact." And I think at the institute we found— especially when people come to summit— they want to hear what we're working on.
- 38:46They want help.
- 38:48And I'm confident we as Vanderbilt can be that partner that gets at some of the hardest problems that is, that are facing us.
- 38:58It's great. How about for you, Brett Benson? What gives me hope?
- 39:00Yeah, what gives you hope? So I think that the same things that make democracies vulnerable are the same things that make democracies able to overcome these types of challenges.
- 39:13For now, any democratic society will be a cycle behind, because a sort of a top-driven organization like GoLaxy, that's incentivized by the Chinese Academy of Sciences, it's it's going to have an advantage and there is an advantage.
- 39:32But time and again in, in democracies when people are responding to incentives, if trust is low enough, then there is an incentive to try to solve that problem because there's a demand for for credible information.
- 39:47Corporations want credible information.
- 39:50People in the marketplace want credible market signals.
- 39:55Congress, you know, eventually should be in a situation where it wants to protect and increase national security.
- 40:03Our national security institutions should, I mean, we have lots and lots of different individuals who are incentivized in the right direction, even if we're maybe a step behind right now, I think that the the creativity that can spread across a democracy and across institutions, academic institutions and elsewhere, I think that's what really gives me hope, because over and over again, we have been able to address big challenges because of an open society where information can proliferate and where there are incentives to respond to these types of signals.
- 40:37Yeah, I love that. I love that idea. Yeah.
- 40:40The sort of, you know, it breeds greater creativity and more entrepreneurial, you know, diversity of out of solution that might actually yield a more powerful outcome.
- 40:51I didn't mean to speak over you, Brett, but I was just my brain was immediately triggered to think about, how constraining fear is to coming up with good solutions.
- 41:03Right? Yeah.
- 41:03If you're afraid, you can't really think as clearly as if you are inspired, right?
- 41:08Yeah, yeah.
- 41:13Super exciting.
- 41:14If you had to say two things or one thing or three things to close, what would you tell our listeners?
- 41:20What would you suggest to them or recommend, or how would you advise us going forward?
- 41:26Oh, your first on that one.
- 41:28First on that one.
- 41:29So I think that public awareness is really important. Yeah.
- 41:33And I guess that's the reason why we're doing this. Absolutely.
- 41:37We ... our message to to Congress and to public officials has been the same on this front.
- 41:42I think individuals need to be vigilant about what it is that they're putting into the public domain, and vigilant about what it is that they're consuming.
- 41:51So public awareness education is key on this.
- 41:54Yeah. Great.
- 41:55Yeah. Vigilant works. Be suspicious.
- 42:00Like we're we're in an age where we don't really know what's real.
- 42:05And you know, think about whether it's a deep fake or whether it's content.
- 42:11Is it real?
- 42:12Is it not?
- 42:13You need to think about as you, like, there's a saying,"If you're using something for free, you're the product." Yeah.
- 42:20And what does it mean to give away all of your information? Yeah.
- 42:25And think about that while you're using tools.
- 42:28I'd say that, but then I'd also say to other people who are listening, if you see content that's suspicious, send it to us.
- 42:36We are very interested, and we're hearing from a lot of researchers now.
- 42:41We're interested in finding personas, things that are just suspicious.
- 42:47Catching them in the wild.
- 42:48Yes, we want wild samples.
- 42:50And then yeah.
- 42:51And also if you have ideas for the research, would love— because we are at our heart very interdisciplinary and trying to figure out and angle together a wicked problem.
- 43:02I love it, I mean, how profoundly democratic, right, to democratize the catching of the things in the wild.
- 43:08I love that it's very much crowdsourcing of science.
- 43:11I think it's a great way for us to end.
- 43:14Thank you both so much.
- 43:16This has been incredibly elucidating, obviously a little terrifying, but also really exciting.
- 43:22At Vanderbilt, we believe universities have a critical role to play not just in advancing discovery, but in protecting the foundations of democracy.
- 43:30Brett and Brett, thank you so much for sharing your work with us today.
- 43:34And to our listeners, stay tuned.
- 43:35This is only the beginning.
- 43:37Thank you so much for being with me today.
- 43:38And thanks to our listeners for tuning in.
- 43:40We'll be back next time with another guest who is pushing the boundaries of discovery.
- 43:44Until then, I'm Cybele Raver, and this is 'Quantum Potential.'