How AI Swarms Weaponize Disinformation

Jonas Kunst: It's an arms race, but we have not joined the war. We need to be better than the adversaries that use them. But to do that, we need to join this arms race, whether we like it or not.

How AI swarms fabricate reality

Michael Krigsman: AI swarms are the most dangerous influence weapons ever built. Daniel Thilo Schroeder and Jonas Kunst co-led a 22-author study published in Science to map this threat. Daniel, what are AI swarms, and how do they fabricate reality?

Daniel Thilo Schroeder: The new thing here is that we see coordinated behavior. There's a whole bunch of studies, in particular from 2024, 2025, that show that large language models, for example, are super convincing even though they lack capabilities to reason. They are very good with talking you into something or talking you out of something. There's, for example, a big study that shows that, they can kind of drag you out of conspiracy beliefs.

And AI swarms go one step further, so they would kind of combine the power of these AI agents, and, just start a coordinated behavior, in order to convince the body of society.

Jonas Kunst: Given that you suddenly are not just communicating with one agent but whether you're exposed to a swarm of agents, that really provides a new arsenal of weaponry, so you can, fabricate agreement by, seeding narratives and make them, and create an illusion of majority support for these opinions. So it's, it's really a next level of, information warfare.

Why cheap inference changes everything

Michael Krigsman: Why does all of this matter today? So please dive back in and then Jonas, you can, please add your thoughts to this.

Daniel Thilo Schroeder: The real new aspect here is the coordinated behavior and, the fact that, access to these AI models and AI inf-inference became cheaper and cheaper, meaning that, these agents can be longer and more sustainable over extended periods of time in online social networks compared to older campaigns where they were there for special, for a special purpose, let's say an election.

Meaning these agents can collaborate to, work on influencing opinions, over extended periods of time and not as obvious as they used to do that before. On top of that, they would be able to understand local environments, and adjust process narratives in a way that they are observed by the local environments in an optimal manner, so to speak. I think these are, these are really the new aspects here.

We have this coordination, kind of strategy development, combined with the convincing capabilities of large language models. I've just said they might not be good with reasoning, but they're extremely convincing. This is what research from the past 2 years has shown. Plus this agentic AI aspect, meaning capabilities to, observe, to understand structure, network structure, for example.

An agent of a AI swarm could, for example, decide to attack a particular central network or a central node in a network. So network centrality in the language of those that do complex network research means a very kind of connected node, So it might be easier to convince a person, that's particularly vulnerable and has a million followers than convincing those million followers.

Jonas Kunst: And I just want to add equally you could aim to inject these type of nodes yourself to get to position your agents in a way that are central to the network, thereby, boosting, the pathways to which you can influence entire groups. And, in terms of what Daniel said, I just want to add that in terms of this coordination, you hinted at it, Daniel.

It's very, just very important to highlight that they are increasingly able to coordinate autonomously. So there is, there is, decreasingly needs for human oversight. So we're kind of moving from, a central command that usually, was at play in traditional bot networks to kind of emergent hive behavior. I think that is a very important, aspect, to keep in mind.

And, in addition, the swarms, while they coordinate, they will be, self-optimizing, meaning that they can run a lot of small experiments, testing out different messages, learning from the feedback these messages produce and thereby can propagate the winners that are most efficient in influencing others. So that really gives them an edge in terms of, boosting their chances of success.

Michael Krigsman: Would it be accurate to say that the key point here is the communication among these many tens, hundreds, thousands, even millions of AI bots and the autonomous action that arises from that? Is that a correct statement?

Daniel Thilo Schroeder: I think it's a very important aspect, Communication and maybe the very central data structure that represents the outside world, which is changed by this multitude of agents and used for coordination, is probably very central from a technical perspective.

However, I think the convincing capabilities of, large language models, in combination with cheap inference and, the extended set of tools, So an agent somehow can, these days control, and we have seen this with frameworks like OpenClaw or several online social networks at the same time, So, we go more towards a persona-centric, p-p-p-persona-centric, approach than, one in which individual online social network accounts are, simulated.

The persona-centric agent

Michael Krigsman: Jonas, can you elaborate on this idea of a, persona-centric approach? It's very interesting.

Jonas Kunst: Previously, you had AI agents that were, q-quite crude and simple in their behavior. They had certain decision rules, on how to behave, but now you've kind of infused them with the cognitive abilities, of large language models. So you suddenly have a much more, capable, AI agent. And as this cognitive ability is basically reflecting human ability to a certain extent, you also have a much more capable collective that can talk to each other.

So I think, that is really the qualitative shift, from previous, influence operations.

Daniel Thilo Schroeder: The idea that one agent is, not representing an individual account in, let's say an online social network, but as a multitude of those, so there could be an Agent Michael that is very persistent and mimics a persona, in a way that it has an email account, it has a Twitter account or X account, a Bluesky account, and a Facebook account.

And this also means that, for example, detection has to be something that is, working platform independent.

Threats to democratic discourse

Michael Krigsman: How significant is this shift from traditional bot networks to AI swarms in terms of the nature of the threat and the impact?

Jonas Kunst: We are very concerned in terms of the threats this has for, democracy. And, unfortunately, some of the capabilities that these swarms have are targeting the pain points of, democratic discourse. First of all, the. If you can, simulate human collectives to a very high degree of authenticity, you can manufacture synthetic consensus on social media platforms.

People are generally conformist, even though we don't want to say this about ourselves, but if people believe something, we assign some credibility to it. This social, heuristic is hijacked by these swarms. But at the same time, you could also see scenarios where, these AI swarms seed fragmentation. They could, bolster different narratives and different opposing communities and thereby polarize societies even more.

In addition, we are also concerned about, their potential to orchestrate targeted harassment to exclude or suppress critical voices. These could be, politicians, journalists, whistleblowers, or even their families. Think about that, and because of this, targeted harassment might, withdraw from, public discourse.

Something we also discuss is that if these swarms become very dominant on social media platforms, people might start to distrust any consensus or individual, opinion, and that might induce epistemic, vertigo and possibly drive users to gated channels.

People disagree on whether that's a good or a bad thing, but at least it reduces, discourse across, political, affiliations, And finally, related to this point about being able to run, thousands of different experiments, tests, or different messages, you could see scenarios where these AI swarms, micro-target different social groups to, a, suppression of, voting intentions or even, mobilize other groups to vote for a certain candidate. So I think the breadth of these threats are enormous and. And we talked about social media platforms. You could see this happening in other channels as well.

Real-time adaptation and human control

Michael Krigsman: What about disinformation that spreads by accident as content gets slowly distorted through reproduction, like a copy of a copy of a copy? This is a question from Anthony Scriffignano, who's a data scientist who's been studying these problems and has been a guest on CXOTalk a number of times. What about that, disinformation spreading by accident as the content gets slowly distorted?

Jonas Kunst: That's really, something that-An AI swarm can monitor and adapt to continuously, something would be much harder with a traditional, farm where you still would need a human to track how the narrative evolves, an AI swarm might evaluate continuously, whether the way this information is presented is beneficial for the swarm's, goals and accordingly react, Maybe try to steer it in a different direction or support this direction if it seems positive.

I think that is probably where, these adaptive capabilities, give an advantage.

Daniel Thilo Schroeder: I would assume that, the capability to constantly monitor this narrative, and the omnipresence of that swarm would allow to very fine granular change this evolution of the narrative, while it's spreading even more than it's happening

Michael Krigsman: Today. What about the aspect of AI agents and humans working together? How does that intersection happens? You were talking about autonomous agents, but at some point, there has to be an intersection with, people trying to drive propaganda.

Daniel Thilo Schroeder: I think this is a spectrum. On the very left side, we would have a swarm that is entirely human controlled, While on the very right side, we would have a completely autonomous swarm that doesn't need any human to action we would define an AI swarm as something that is occasionally, instructed by a human with very high level, commands, or high level agendas from there to completely autonomous.

And this is a spectrum, and I think this will advance with the technology.

Jonas Kunst: The better these feedback loops get and the learning from the feedback, the less influence you probably need by humans.

Poisoning data across platforms

Michael Krigsman: And what about LLM grooming and the corruption of AI training data as a result of AI swarms?

Jonas Kunst: AI swarms can flood the internet, flood social networks, but also create a lot of different, pages on the internet that are fabricated, shattered for, and optimized for machine consumption, through certain, schemas. And when, large language models are trained on this data, these, synthetic narratives, calcify within, their model weights during retraining. And that poisons the internet's epistemic substrate, because few. It might in a worse case force future AI tools to, output skewed facts as objective reality.

Michael Krigsman: We have, an interesting question from Twitter, from X, and this is from Chris Petersen who says, "Are AI swarms on social media largely text at the moment, or are we already seeing multimodal or multi-platform swarms coordinating?"

Daniel Thilo Schroeder: It currently lacks the methods, to detect the coordination signals of these swarms, there's a lot of AI-generated content on social media. And when you speak to experts, in the platforms or that operate these platforms, they recognize an increase in automation of coordinated behavior. You'd still have to differentiate between human coordinated behavior and completely automated coordinated behavior, and this is very hard to detect. We're lacking the mechanisms.

But we would assume that there are already, multi-platform coordinated attempts, to influence.

Jonas Kunst: In terms of multimodal stimuli, that would not be a technical leap. That's feasible. There are many, bots that have been detected that create avatars, that are photorealistic, that can evolve even, changing clothing with the same person, over time. Both in terms of producing content using different types of, media and also in terms of representing themselves, very likely.

Michael Krigsman: Folks, I want to say that you can ask your questions. If you're watching on LinkedIn, pop your question into the chat. If you're watching on Twitter, X, use the hashtag CXO, CXOtalk or reference me, @mkrigsman. But take advantage of this opportunity to ask 2 world experts on the subject of AI swarms and disinformation whatever you want. Use this opportunity. Gentlemen, have.

Detecting coordination, not content

This is another question from Anthony Scriffignano, "Have you considered how to measure the accidental versus intentional manipulation of the truth by AI swarms?" you were, alluding to this a moment ago, but how do you make the distinction between the intentional manipulation of social media versus, people talking and the organic spread of information? That's

Jonas Kunst: The real challenge here, Because at the message level, it is not very distinguishable anymore, from, human, outrage or some type of social movement. Our main argument is that we need to have access to, what happens under the network itself in the background. We need to be able to detect unusual coordination.

That's the only way to do it because the individual accounts are so believable, the messages won't reveal an AI agent, and by far not an AI swarm anymore. We need to look at the group level, the coordination level.

Michael Krigsman: You're saying that the agentic AI development has reached a point where it's indistinguishable. Each message is indistinguishable from a real person. Is that, is that. Am I hearing that correctly?

Jonas Kunst: Exactly. We are very close at least to that level. There are several studies that look at the message level that generated text is often perceived as even more human than, human generated text. That is not a, choke point anymore. Ne- we need to, you zoom out and look, try to look at how accounts connected, how their narratives connected, how they sync, systematically. These are the patterns we are interested in.

Michael Krigsman: You're looking at, content patterns, the same types of messages. Is that basic. Same type of content across accounts?

Jonas Kunst: Yes. That is one way. What is the distance between the different type of messages? What are. When is something posted, when the responses appear, there regularities in terms of who responds to what type of, content, these are the way to try to identify whether people work together or agents work together.

Michael Krigsman: It is extraordinary. I was looking at, TikTok the other day, and I came upon a TikTok post. Had something to do with Elon Musk and saying. Describing all of his properties in Texas. And there were, 600 or more at that time messages, most of which were supporting Elon Musk. And I was astonished because it was obvious that these are fake messages because just the volume and the consistency of the message.

But the use of language, the use of idioms, it was. It all came up, came across so naturally that each. If you were to look at any one message, they all looked real.

Daniel Thilo Schroeder: This is what we observe. This large language model and this is in particular true for online social networks, would pass Turing test. In an online social network, communication is sparse, So it takes a while until somebody reads a comment to a tweet, for example, and comments on that again. It's almost indistinguishable, at least this is what we think.

Jonas Kunst: Interesting simulation papers that show that you can use AI agents to simulate, societies. Yeah, it is getting very close to how, humans mobilize, how express themselves, and that is the challenge. But probably the coordination is still imperfect and that's where we need to, focus our research agenda.

Daniel Thilo Schroeder: It's not only the coordination. You could imagine that a swarm, while it's living, is assessing this social environment, So there particular attack surfaces, areas in a social network that are more prone to be targeted, this could also help to detect those, being aware of this target threat landscape, and identifying those areas, where swarms potentially attack to find if several accounts follow a similar strategy or show coordinated behavior, is what we want to go for.

That being said, Michael, these things do not yet exist. So attempts to look into behavior general, but not at coordinated behavior at a large scale, and this is what we want to change with our research. And we would like to simulate to understand these attack surfaces and these coordination patterns that would emerge from AI models automatically coordinating these attacks to develop detection that is appropriate or even active countermeasures if detection doesn't work anymore.

Jonas Kunst: It's increasingly difficult for us to look under the hood of, platforms because, flying blind since tech companies guard their, API data or make it very expensive. That is a very, difficult situation for us because we lack access to the data that would allow us to develop these type of detection systems.

Platform incentives and the detection gap

Michael Krigsman: We have potentially millions of coordinated accounts making agentic decisions, autonomous decisions among themselves as they go Each particular message that they're sending appears indistinguishable from that of a real person, and we do not yet have effective methods for identifying in a, i-in a, in a repeatable way, identifying these swarm accounts. And the social media c-platforms do not easily give up that data, so it's therefore hard to simulate. Seems we're screwed.

Daniel Thilo Schroeder: I think this is not the case. Are researchers, and as such, we have to point to risk and why we, observe that there's probably an increase in automation of this coordinated disinformation or influence campaigns. We also see, a lot of opportunities for platforms for, the very building blocks of democracies in AI. We see a lot of opportunities in AI for active countermeasures, also for detection. I think we are not lost here.

We just think we need to point to the fact that there's not that much money, for example, AI research and AI security research, and we would like to understand this phenomena to use AI for the good, to detect, to prevent, to, create a situation where we're not screwed.

Jonas Kunst: In conversations we have had with, people working at these different platforms, it appears as if, they work on such detection systems. But the question of how much resources they are willing to assign to these tasks probably depends on, to what extent, AI swarms benefit or threaten their business model.

Because y-you could make the case that, if they generate massive volumes of posts, likes, and shares, to artificially boost content, and elicit real emotional responses from organic users, from humans, that might prolong the platform use of these, users and at the end inflate the daily, active user count the platforms report to advertisers and shareholders.

I think it's a question of, whether and when AI swarms become dangerous for these, platform business models that we will see a shift in resource allocations.

Michael Krigsman: You're fighting the, business models of the platform companies such as Facebook or Twitter, X, and their, incentives to enable and even reward swarms of fake accounts because of the advertising revenue, the engagement numbers that these swarms create. Is that correct?

Jonas Kunst: That covers this well to platforms, if you reach a state where AI swarms are so prevalent that we reach this, status quo of an epistemic vertigo, that's not beneficial for the models. There will be an upper limit to, how beneficial it is to them. But right now, we don't see that. Time will tell.

Michael Krigsman: Especially since the AI swarms are so effective at creating social media posts that look indistinguishable from real, so they can therefore turn a blind eye. They have, plausible deniability.

Jonas Kunst: They do. And also, algorithms, as we all know, engaging content, high arousal emotions, outrage, tribal identity, to maximize the watch time of users and keep them scrolling. And AI swarms can, run experiments to test and generate content that perfectly is calibrated to trigger, those exact human reactions. And there is this other layer to some of the platforms, where you have direct monetization, programs.

So if creator funds and ad revenue, are shared with people creating content, this also allows swarms to convert their, engagement and content directly into cash. And as you say, because the system cannot, distinguish, or not with a high accuracy distinguish between a genuine human outrage and algorithmic mimicry, these platforms are, effectively paying malicious, actors to pollute their own information ecosystem.

Michael Krigsman: We have, an interesting question from Thalyann O. On, LinkedIn. And Daniel, maybe I'll direct this one to you, and you can jump in and, Jonas, add your thoughts. And Thalyann says, "Could there be a way to authenticate human-made data by introducing random artifacts in the metadata, or is that something AI swarms could figure out a workaround?" For example, she says, using something only humans can produce, misspellings, varied typing speeds, and so forth.

Daniel, thoughts on that?

Daniel Thilo Schroeder: I think in or these unique features can be replicated by AI, as far as I know.

And when it comes to features like typing speeds, these messages can be from everywhere, you're not necessarily forced to use, a keyboard and a browser to tweet, It can also happen on a phone, or I can copy-paste, and this is what happens quite a bit a tweet or a social media post from a large language model into a chat window or into a message window, And I'm not a big fan of these authentication methods, and I would assume that they can be, faked to a large degree.

Jonas Kunst: I agree with that. And in terms of these human imperfections, a model could be trained to emulate them. That doesn't help much.

Toward a global early warning system

Michael Krigsman: What do we do? Here in the US, we have elections that are coming up, and it seems we are at tremendous risk. So what should. What do we do?

Jonas Kunst: Obviously, these capabilities are of, massive interest to, political campaigns that seek to gain an electoral edge or, foreign adversaries that want to other, countries. We have a few visions on how we could, address these developments, and one of them is to build an, early warning network of some type in our paper, we argued for establishing a distributed AI influence observatory.

Instead of relying on a single tech giant or a government agency, this could create a decentralized early warning system where academic institutions, researchers like us, NGOs, and civil society can share threat intelligence in real time. That's a very important first step to start to document, these AI swarms and to share that information openly.

I already talked about the necessity to, encourage social media platforms to open their hood and at least let independent researchers get free access to the data. That is a necessity. And at the same time, we also need to stop fighting, the last war. So we have to, look into the future.

We need to invest in, agent-based simulations, which essentially means using defensive AI to, red team our own networks and see what works, what doesn't, and understand the capabilities of AI swarms. In other words, you would, simulate, how future swarms might attack, during an election period within a safe, isolated sandbox so that we can, stress test and train our detection systems before, malicious actors deploy these, new tactics in the wild.

Daniel Thilo Schroeder: I think the first thing we need to do, Michael, is create awareness. There's very little money, that allows us to understand this phenomenon.

In the Science-Policy Forum paper in which we came together with a lot of expert in order to inform policy that these things, are going to happen, we argued that automation and AI technology will accelerate, what current influence, operations already could do, The first step here is to realize and to create awareness, and understand the phenomenon, understand the degree of automation, understand what the capabilities of large language models and AI models in combination with agentic AI in influence operations.

And for that, we need what Jonas said, data access to platforms. We need a communication infrastructure that allows to, communicate or allows communication, inter-platform communication, as well as communication to policymakers and researchers. And then we must simulate this phenomenon order to. We had that already with this threat landscape, We need to understand what type of strategies emerge when these swarms, operate, and then build detection from that.

And after that, it's important to, also look into things like active countermeasures. This idea that you can't really prevent each and every campaign or AI swarms campaign, but that you can prepare people to not believe in the single. Being more, this is probably a good strategy. And last but not least, it's important to inform those that were targeted. Countermeasures very often do not reach those that, actually believed in the false narrative.

Realizing that you can inform people after they were targeted, and correct is probably a very good strategy as well.

Michael Krigsman: Where does responsibility for all of this lie? Is it the government? Is it private. Corporations? How, part of the problem it seems is the impact is diffuse across society, and therefore the responsibility right now is also unfocused. So who's, who should be doing this, taking the lead on this?

Daniel Thilo Schroeder: This is what makes so complicated, Responsibility is diffuse. It's first of all each and every of us that, has to prepare for not believing everything we read, for questioning things we read on the Internet, but also platforms, governments, and this is really complicated here, there is no single entity with responsibility for all of that.

Jonas Kunst: It's critical to have collaborations between policymakers, regulators, tech companies, to work together. It cannot be solved by one party, by. On its own.

Michael Krigsman: And on this point, Chris Petersen on, Twitter, X, says, "Are there additional dangers or opportunities that arise by having direct ties between the social media providers and the AI companies? For example, Meta, Facebook, Twitter." So when you start having these ties, there are additional risks potentially created.

Jonas Kunst: We are very in terms of collaborating with industry because there will be some conflict of interest, that needs to be managed. That is definitely something that we need to be very aware of.

As I said, it is debatable whether tech companies, maybe not so much the large language model developers, but, the social media platforms, to what extent that is, AI swarms are their current it might be more of the interest maybe to the business models of large language models because it really undermines their training data. Conflicts of interest are real, whenever you work with, tech companies on these issues.

But on the other hand, you suddenly, get possibilities, access to data that would not be possible before. So it's a, it's a trade-off.

When swarms target corporations

Michael Krigsman: What about the impact on corporations, on businesses? Do they face similar kinds of influence operations that individuals face when it comes to things like, elections or in general public opinion?

Jonas Kunst: We have so far focused very much on the of political aspects of AI swarms, but, the tactics of, manufacturers can be weaponized against, corporate brands just as easily as against political figures. A competitor or some malicious actor might deploy a swarm to, simulate a massive grassroots boycott of a certain product, or fabricate, safety claims, in, at a large, volume. And generally consumers naturally err on the side of caution, when they see even a few isolated complaints.

And now imagine what happens if they see a majority, an apparent, majority warning about a safety issue. That can cause, in a worst case, consumer panic and, brand abandonment. And, in addition I talked previously about this potential to harass, certain, political figures, academics, journalists, or even their, families to suppress their, voices or to make them withdraw from certain platforms.

That could also be, turned against executives, corporate whistle-blowers, or board members that might, try to force them to resign from their position or, try to support a new corporate policy shift, for example. Yes, the spectrum of, domains where you can maliciously utilize the potential of AI swarms is broad. It's not just limited to politics.

Who is launching AI swarms

Michael Krigsman: Have you been able to identify w-where these swarms are coming from? In or, i-in other words, who's organizing these things? Daniel, have you seen patterns?

Daniel Thilo Schroeder: We think that organizing these things in particular became very easy. We currently working, with a colleague of ours, Lucas, who has programmed such simulations that can even run from within your bedroom, So it's really everybody that can access these technologies. Furthermore, you have all these coding tools these days that allow people with a very low threshold to build their own campaigns.

I would assume that state actors, for example, or malicious actors that have, geopolitical interests are probably way more sophisticated, campaigns and access to more resources. However, and this is part of our assumption and our statement here, everybody is capable of, building at least small scale AI swarms these days.

Jonas Kunst: And I just add that we have some, examples from very related, influence operations, something we call cyber propaganda, which uses the same structure as an AI swarm, but at the very end has real humans copy-paste the message, tap posts, and then, disseminate this content to their social networks.

Is that is a clear reality that, according to colleagues that we have at the, Canadian National, Observer, has been used, I believe, in Israel, in Portugal, and in Georgia elections. And these are applications that are ready to use. You can, you can buy access to them. So the reality is here. It's not only limited to a powerful, foreign state that has the resources to do something like this anymore. It's very accessible.

Michael Krigsman: I'm assuming that you need a way to automate the creation of accounts on social media. You need to generate AI, headshots, and then you need an LLM to, describe the text, the messages that you want to convey. It's more to it than that, but it seems that's the basics. Is that correct?

Daniel Thilo Schroeder: Without going too much into detail because we do not want to give the instructions here, But each of these agents within a social network can't be detected, so there's something like a cost function for every action an agent can perform. And this also limits the environment. So an agent embedded in a social network can only see a certain area of that social network and monitors it.

And then it's a lot about, adjusting higher level strategies that are built in a hub, to local environments. And this is a basic mechanism here. Plus we need capabilities, and this is probably, beyond what you just said, to reflect and assess the success of, commands, plus, the capability to assess psychological profiles, based on social media data.

But there's way more, We just talked about businesses, there's an entire ecosystem for data aggregation that allows to facilitate marketing, this is all data and data sources, that can be used, and there are no problems to be used, with agentic AI these days anymore.

Jonas Kunst: You for me, the gist of this is that, developing, thousands of agents that look realistic and, pass as real, that's not a challenge. The challenge is really creating efficient coordination systems that are autonomous to a certain extent, that self-optimize. These are, computationally more complex tasks. But also, as Daniel said, white coding, existing multi-agent frameworks that are out there, have really lowered the barrier, of access.

An arms race we haven't joined

Michael Krigsman: Correct me if I'm wrong, what you seem to be saying is the technology for creating AI swarms has reached a relatively. Has a v- a relatively low bar, at least for the basics. And on the other hand, the technology for evaluating that AI swarms exist and combating the problem has not yet reached a point where we can do this consistently and effectively.

Jonas Kunst: Exactly. It's an arms race, but we have not joined the war. So that's, that's really where we at, are at right now. And as Daniel also said earlier, it's hard to prevent AI swarms fully. We need to compete. We need to be better than the adversaries that use them. But for that, we need to join this arm-arms race, whether we like it or not.

Michael Krigsman: What happens if we don't join this arms race? What happens if the status quo, continues as it is, meaning influence campaigns can evade detection? What goes on? What happens?

Daniel Thilo Schroeder: I would assume the clear bottleneck here is AI inference. And, being the one that has, more sophisticated models and capabilities, and this, unfortunately, in today's world, just a few. So, my assumption is that the power and the power to control masses will probably, be in the hands of a few, namely those that have the resources.

Michael Krigsman: What about the impact in areas su-such as we've been discussing, impact on elections, impacts on, corporate reputation, things like that. What, where does. Logically, where does this go over the next number of years if we don't find, better technologies for addressing the problem?

Jonas Kunst: I think it, this links directly up to what you just said, Daniel, that, influence is going to be very asymmetrically distributed. So, some people who have resources are able to influence all types of domains, be it elections, corporations, business.

So, first, of course, we also need to keep in mind then, while this influence increases, the power of real collective action, comprised by real humans who care about an issue gets watered out, because it's going to be hard to compete with such a machinery.

Michael Krigsman: I just think about this. Here in the States, we've had demonstrations recently on a pretty large scale, and I just think if the people trying to organize, collective action are fighting, swarms, AI swarms, one, do they crowd each other out? Do the, do the people crowd out the swarms? Do the swarms crowd out the people? Do they operate totally independently? Although you said that they don't because they. The s- the swarms adapt based on the behavior and the psychol- psychological profile and so forth. So thoughts on this?

Daniel Thilo Schroeder: I don't really know if this is about crowding out, So, our. This, my main concern here is that you can, sustainably change, minds over extended periods of time. I think this is not too much about having one crowd that is larger than the other.

I would even argue that you do not need these millions of agents, Maybe it's enough to have a few that are very convincing, It really depends on the operation and what you would like to achieve. And I don't know if, crowding each other out is really the race here.

Jonas Kunst: I do think that conformist tendencies are very strong and, cultural evolutionary models show that who is the majority or the seeming majority really matters for how people, what they consider even within their overtone window or what they, consider to maybe adopt as their own beliefs. So I do think that there's strength in numbers in terms of AI swarms. I also agree with what Daniel says that there's. That's not the only factor.

How you do it, there might be many ways of, doing it more efficiently.

However, in terms of what you just said, with people, protesting on the streets, that's a really interesting scenario because, even if we, if you then suddenly, initiate a counter or a, an AI swarm that pushes some type of, opposite narrative, that will be very obvious, and might actually be a pr- a way to reveal, such AI swarms when social media, behavior, large scale behavior doesn't match up with what people see happening on the streets. So that might really help people to detect that something is off.

However, th- this might also again, lead to some type of epistemic vertigo where people just distrust social media, and social media can be very beneficial in terms of organizing, protests in the first place.

Advice for leaders and policymakers

Michael Krigsman: As we finish up, let me ask each of you for your advice to people in business and especially to policymakers. Jonas, maybe I'll start with you and then Daniel, you can add your thoughts as we finish up.

Jonas Kunst: People need to wake up to this reality. They need to place this threat at the top of their, let's say, boardroom agenda and, really acknowledge that traditional digital engagement metrics are, fundamentally compromised. I think that's, that's a first starting point, in terms of, developing strategies for the future. I think that would be one of the first things.

And, in terms of corporations, I would probably say that, but maybe also leader- leaders generally, they need to recalibrate their crisis response protocols so they don't panic or, re- reflexively, alter corporate strategy just because there's a sudden wave of, coordinated online outrage. So people need to be aware of that and factor that in their, crisis response.

Michael Krigsman: Daniel, advice for leaders, in the private sector as well as in the government. Any thoughts on that?

Daniel Thilo Schroeder: We first of all have to understand we need access to data, and we need to create infrastructure that allows us to communicate and collectively understand, these phenomena and also the influence in general of AI on, influence operation and cognitive warfare. This is super-duper important.

Jonas Kunst: I think it's very important to, for people to reach out to, other, sectors of society. We talked about, private companies, policymakers, to work together and, utilize. To use synergies that might exist in order to, build protections against this threat.

Can AI swarms serve the public good

Michael Krigsman: And we have actually one last question that's come in under the wire here, on LinkedIn from Greg Walters, who says, "Can AI swarms be used beneficially in a positive way for society?"

Daniel Thilo Schroeder: Yes, they can also, good intents, they can, fact-check, they can collaborate and just, build, so let's say digital twins of humans in order to process information in a way this particular human would understand. There's a lot of opportunities and a lot of chances, and this is part of the arm race, Jonas has described before, Developing these technologies, and understanding, what, let's say, a good swarm would do, when a, when there's competition with a bad swarm.

Jonas Kunst: This is of course not without risks because whenever you use, let's say, positive AI swarm as Danny said, that's, that also to some extent normalizes the use of such swarms for certain goals. Let's say political goals. So it's, there's a trade-off, and a decision has to, has to really be weighted, against possible drawbacks.

Michael Krigsman: Okay. And with that, we are out of time. A huge thank you to our guests, Daniel Thilo Schroeder and Jonas Kunst, who have written a very interesting and important article published in Science Magazine, and you can look it up. Gentlemen, thank you so much for taking your time to be here. This is such a crucial topic. I'm very grateful to you.

Jonas Kunst: Thank you so much.

Daniel Thilo Schroeder: It

Jonas Kunst: Was a pleasure. Pleasure

Daniel Thilo Schroeder: To

Jonas Kunst: Be here.

Daniel Thilo Schroeder: Thank you.

Michael Krigsman: And to everybody who watched, thank you for joining us and for your great questions.

Now, before you go, subscribe to the CXOTalk newsletter. Go to cxotalk.com and hit Subscribe so we can notify you about upcoming shows. We have great shows that are scheduled.

Right now we have shows that are scheduled through July. So everybody, thank you so much for watching, and we'll see you again next time. Take care.