Hayagreeva Rao and Henrich R Greve, "Ctrl+Alt+Doubt: Decoding the Language of Online Conspiracy Talk" (Oxford UP, 2026)

Summary

---

Podcast Summary

Episode Overview

Podcast: New Books Network
Episode: Hayagreeva Rao and Henrich R. Greve, "Ctrl+Alt+Doubt: Decoding the Language of Online Conspiracy Talk" (Oxford UP, 2026)
Date: July 5, 2026
Host: Alfred Marcus
Guest: Henrich R. Greve (co-author, with Hayagreeva Rao)

This episode dives into Ctrl+Alt+Doubt, a book that reframes online conspiracy theories as processes of “distributed doubt” and analyzes how conspiracy talk spreads linguistically and socially on platforms like Twitter. The conversation explores why conspiracy narratives gain traction, how they are shaped and shared, the limitations of fact-checking, and what policies or institutional incentives might counteract their harm on democracy and public health.

---

Key Discussion Points & Insights

1. Distributed Doubt: Moving Beyond "Crazy Individuals" and "Bad Algorithms"

• Reframing Conspiracy Theories:
  Greve and Rao shift the focus from blaming “vulnerable individuals” or just algorithms to examining distributed processes of doubt in social media.
  • “Very, very regular, ordinary people can be gifted into these beliefs if they get the right kind of persuasive information.” (Henrich Greve, 04:56)
  • The real driver is the social, collective nature of sharing and reacting, not just isolated mental pathologies or purely technological biases.

2. What Makes Social Media Different – and Dangerous?

• Speed & Selectivity:
  • Social media accelerates the spread of misinformation and enables the formation of insular belief communities.
  • Example: COVID-19 conspiracy theories quickly led to widespread denial and measurable harm.
    • “You’ll have twisted belief systems… that COVID-19 isn’t a real disease… responsible for significant loss of lives.” (Greve, 06:48)

3. AI, Bots, and the Future of Conspiracy Spread

• Bots are “Effective Through Mass, Not Sophistication”:
  • Current bots spread conspiracies via synchronization; AI could bring “mass and sophistication,” posing even bigger challenges. (Greve, 08:52)
• Countering with AI and LLMs?
  Fact-checking by bots or LLMs is mostly ineffective. The spread process itself must be slowed.

4. Why Individual-Focused Interventions Fail

• Fact-Checking, Inoculation, and Prebunking:
  Evidence shows these have little to no effect once doubt has taken hold.
  • “Any intervention oriented towards the individual… those interventions are ineffective... The only lever that has any effect would be influencing the spread.” (Greve, 10:40)

5. Institutional and Economic Incentives

• U.S. vs. European Models:
  • U.S. platforms set rules themselves and have little legal liability.
  • In the EU and Asia, the state can penalize platforms for harmful content (racial incitement, health misinformation).
    • “In Europe… the state sets the rules… and regulated the same way.” (Greve, 12:51)
• But legal frameworks often lag, as seen during COVID-19.

6. Cultural Bricolage and Conspiracy Talk

• Bricolage Theory:
  • Culture is not just internal; it is pieced together from external language and experiences.
    • “People might be in an area of social media where there’s a lot of them, so they start repeating them… to fit in.” (Greve, 20:52)
  • Conspiracy talk spreads as users absorb and repurpose the elements that appeal or help them fit into a group.

7. Methodology: Computational Analysis and “Semantic Cascades”

• NLP Tools and Embeddings:
  • The research uses publicly available NLP tools to map, classify, and track how conspiracies cluster and cascade.
    • “Natural language processing cuts through the variation so that we can say, oh, this person is repeating the same conspiracy theory.” (Greve, 25:30)
• Semantic Cascades vs. Rabbit Holes:
  • People are more likely to jump to similar conspiracy themes—creating “cascades” rather than simple one-way falls.
    • “Once somebody has adopted… a conspiracy theory… they are very likely to fall again… to a similar one… they keep tumbling.” (Greve, 33:35)

8. Conspiracy Structures and Gateways

• Simple, Plausible Conspiracies as Gateways:
  • Milder doubts ("false positives in COVID tests") open the door to more radical ideas ("Bill Gates is implanting microchips").
    • “Once they get into a gateway conspiracy theory… suddenly… more conspiracy theories became plausible.” (Greve, 37:16)

9. Responsibility & Platform Power

• Platforms Could Act—But Won’t (for Now):
  • Platforms have the tools to flag, demote, or ban conspiracy content, but economic incentives (keeping users engaged) outweigh motivation to act.
    • “Detection and prevention… requires some human intervention… but it’s not a big investment.” (Greve, 40:20)

10. Where to Draw the Line? Ethics, Governance, and Harm

• Factually False Content with Implications of Harm:
  • Conspiracies often allege deliberate harm, which can incite violence or dangerous behavior.
    • “Conspiracy theories… facts that in real life are unconnected, they become connected into some grand scheme where there is this evil entity… that’s already problematic.” (Greve, 43:28)
  • Examples: 5G arson, AIDS denial in South Africa, COVID denial leading to non-compliance with public health measures.

11. Policy and Social Change – Slim Prospects

• Only Structural Change Works:
  • “The only effective lever is to prevent the spread.”
  • Individual fact-checking or nonprofit policing is “near trivial in magnitude” or ineffective (51:49, 57:10).
• Effective Regulation:
  Regulation must focus on making platform owners liable for harm.
  • “The effectiveness of regulation depends on how well it controls the platform owners and the platform behavior.” (Greve, 54:14)
• But U.S. legal structures (Section 230) shield platforms from accountability.

12. Pessimism and Historical Parallels

• Cyclicality of Media and Misinformation:
  • “Modern social media is remarkably similar to society in medieval Europe… there wasn’t much of a truth around. There were many, many truths depending on where you were and who you talked to.” (Greve, 61:41)
• New media often amplifies, not enlightens. (Citing Majda Teeter; Marcus & Greve, 63:06–63:40)

---

Notable Quotes & Memorable Moments

• “Our version differs by being more correct than the others.”
  (Henrich Greve, 03:56, on why their distributed doubt theory supersedes individual or algorithmic explanations)

• On bricoleurs:
  “Culture, you find culture outside yourself and you… start using those elements that you want to use, for example, to fit in or because they appeal to you. And this is exactly where conspiracy theory comes in.” (Greve, 20:52)

• On platform capability:
  “A classifier can be trained in a matter of hours if you put enough people on it… then the machine is turned loose on everything and it’s actually more reliable than the person.” (Greve, 30:05)

• On futility of fact-checking:
  “All the evidence we have is that that is ineffective… any intervention oriented towards the individual… those interventions are ineffective.” (Greve, 10:40)

• On economic incentives:
  “It’s mostly a matter of willingness.” (Greve, 42:19, on why platforms don’t suppress conspiracy content)

---

Timestamps for Important Segments

• 02:03 – Introduction and overview of distributed doubt
• 03:21 – Greve explains the shift from individual/algorithm focus
• 06:21 – Historical context & what’s new about social media
• 08:31 – AI, bots, and LLMs in conspiracy spread
• 10:15 – Why fact-checking and individual interventions don’t work
• 12:37 – US vs. EU/Asia legal and incentive structures
• 19:43 – Bricolage, cultural toolkit and the language of conspiracy
• 24:23 – Natural language processing – methodology
• 33:21 – “Semantic cascades” vs. “rabbit holes”
• 36:25 – Structures of common COVID conspiracies
• 40:40 – Detection, corporate incentives, and possible technical responses
• 43:21 – Harm, ethics, why lines are difficult to draw
• 49:04 – If you wanted to counter conspiracy spread, what works?
• 51:49 – On the near-trivial effect of “accurate information” as a counter
• 54:08 – What kind of regulation might be effective
• 57:10 – On policing, nonprofit, and academic attempts
• 61:19 – Social media as medieval Europe redux
• 63:06 – Parallels to the printing press and spread of prejudice
• 64:10 – Closing question: What’s next for Greve’s research?
• 66:41 – Closing remarks and book recaps

---

Conclusion & Tone

Throughout the conversation, both host and guest maintain a thoughtful, empirically grounded, sometimes darkly humorous tone. The mood grows increasingly sober as the challenges of combating conspiracy theory spread are laid bare—especially within US legal and market structures. Greve remains analytical and cautious, offering measured pessimism about the prospects for quick solutions, but underscoring the critical need for structural—not educational—responses. The episode serves as a deep dive for anyone interested in the intersection of computational social science, information ethics, and the future of public trust.

---

Transcript

[Commercial Voice Actor]

When you're seven eleven's Hottest Fire Chicken Sandwich. People think you're intense. Well, yeah, I'm hot. So yeah, intense. Wanna see if you can handle the heat? Then try me with pickles or spicy mayo and jalapenos. Oh, and it's all just $4.99. That's a deal. Almost as hot as I am.

Almost.

Sweets? No spicy. See for yourself. $4.99 and only at $7.11. Follow up through 82546 participating stores only while supplies last. Visit store for full terms.

Chronic migraine is 15 or more headache days a month, each lasting four hours or more. Botox Onobotulinum Toxin A prevents headaches in adults with chronic migraine before they start. It's not for those with 14 or fewer headache days a month. It Prevents on average 8 to 9 headache days a month versus 6 to 7 for placebo.

Prescription Botox is injected by your doctor. Effects of Botox may spread hours to weeks after injection, causing serious symptoms. Alert your doctor right away as difficulty swallowing, speaking, breathing, eye problems or muscle weakness can be signs of a life threatening condition. Pain patients with these conditions before injection are at highest risk. Side effects may include allergic reactions, neck and injection site pain, fatigue and headache. Allergic reactions can include rash, welts, asthma symptoms and dizziness. Don't receive Botox if there's a skin infection. Tell your doctor your medical history, muscle or nerve conditions including als, Lou Gehrig's disease, Myasthenia gravis or Lambord Eaton syndrome and medications including botulinum toxins as these may increase the risk of serious side effects.

Why wait? Ask your doctor. Visit botoxchronicmigraine.com or call 1-844botox to learn more.

If roaches are getting comfortable in your home, it's time for STEM's multi insect spray made with botanical extracts. It's safe to use around people and pets when used as directed, killing ants, roaches and flies without the fuss and with no added dyes, fragrance or harsh chemical odors. You can spray it inside and outside your home. Live life Never bugged with STEM. Visit stemforbugs.com to learn more. Welcome to the New Books Network

[Alfred Marcus]

welcome to the New Books Network. I'm Alfred Marcus and this book is part of my series on the Cusp between Strategy and Ethics. In this series, I talk with authors whose work sits where organizational strategy meets world responsibility. My own work has tried to explore that intersection, most recently in Comeback, my book, which looks at how companies and crises attempt to recover by realigning their strategy, governance and ethics. Today, we're looking at a different kind of crisis, one that runs through our information systems and our democracies. The rise of online conspiracy theories. My guest is Henrik Greg Rev, a co author with Hagi Rao of Control Al Baot from Oxford University Press. The book offers a linguistic and cultural analysis of online conspiracy talk, treating it as a distributional doubt that spreads through social media conversations, rather than just a matter of individual pathology or platform algorithms. Henrik, welcome to on the Cusp.

[Henrik Greg Rev]

Thank you so much. Thank you for inviting.

[Alfred Marcus]

Thank you for being on this podcast. So let me start with your core reframing. You describe online conspiracy theory theorizing as distributed doubt in the wild of social media. What do you mean by that and how does it differ from the more familiar stories about crazy individuals or bad algorithms?

[Henrik Greg Rev]

Well, we mean really doubt about the truth, things that we hold as facts, and also doubt about those that we would normally trust to transmit the truth. Doubts about science, about press, about regular conventional media, as opposed to all the opinions in social media. I mean, those things are no longer taken as factual, necessarily. And it differs from these other accounts? Well, to put it briefly, our version differs by being more correct than the others. But let me make that a bit more precise. There has been a bit of historical overemphasis on people being sort of vulnerable to believing falsehood, believing conspiracy theories. And it is true that there are vulnerable people who can more easily be persuaded. But what's more important is that very, very regular, ordinary people can be gifted into these beliefs if they get the right kind of persuasive information. And so we shouldn't be focusing on the individual so much. It is also true that algorithms spread conspiracy theories, but they don't spread conspiracy theories that much better than other popular content. In fact, people are good at finding conspiracy theories once they've become interested in them, even in the cases where algorithms try to discourage it. So, so really the, the guilty party here is the person, the person who reacts to, to the conspiracy theory, the person who reposts the conspiracy theory, the person who then goes on to look for other conspiracy theories. And, and this is not a, a special crazy person. This is, this is regular person.

[Alfred Marcus]

I think we've always had sensational media. If you look historically back to the 1920s and people even then were swayed by new media, in what way then are the social media of today different? And are they more dangerous, in your opinion, or less dangerous?

[Henrik Greg Rev]

So what makes them. Well, there are multiple dimensions that make them dangerous, but And I would say the speed is an important feature because these conspiracy theories spread so quickly. I mean, any kind of information or misinformation or disinformation will spread very, very fast in social media. That's one. The other one is the selectivity. So you can have these little communities believing certain things and they can suddenly grow. And so you'll have, you know, twisted belief systems, for example, that COVID 19 isn't a real disease. You can have that belief spread very fast and as a result, people are not taking protective measures. I mean, this is a conspiracy theory. We don't have an estimate or we haven't found an estimate of how many people that conspiracy theory killed. But it's a conspiracy theory that's responsible for significant loss of lives, transition towards

[Alfred Marcus]

AI and towards LLMs. They're very fast in what they do, as I've noticed today in particular because I was working with them. And do you think they'll be more objective? I've tried to ask some questions of the LLMs and I think they are more objective. There was actually an article today in the Washington Post about that and they said they differ in their politics and they go anywhere from being a little to the right, no surprise in Grok, to being balanced, to being more on the left. Do you think as people rely on them more, it will be a corrective in any way? I know this is outside the book, but I'm just wondering about your opinion on that.

[Henrik Greg Rev]

So it's unclear. I mean, one of the things we have paid attention to, the bots and the bot farms, because a fair amount of conspiracy theory spread was through bots and we, I mean, bots are slower than people, they are less flexible than people, but they're still very effective in spreading conspiracy theory because there are so many of them and they are so synchronized. But they are effective basically through mass, not through sophistication. Now, if you, if you make them more sophisticated through the use of AI, then you would have mass and sophistication, which is not a situation that I would enjoy being in, frankly.

[Alfred Marcus]

So people would create conspiracy theories using LLMs and then spread them through social media. Is that the channel that would exist

[Henrik Greg Rev]

were that, well, they're already creating them, you know, through, through human means and spreading them through social media. The worthy LLMs would be useful would be that they would be more sophisticated in replying to repostings of their content.

[Alfred Marcus]

They play any role in countering the conspiracy theories. And in general, I know we're getting to your recommendations for how to oppose this trend. But in general, what do you propose is the way that we deal with this problem?

[Henrik Greg Rev]

So what we if you propose that LLMs would take a role in countering, I mean that will be along the lines of fact checking. Which people do? I mean there are many people who volunteer to fact check conspiracy theory and those who spread them and all the evidence we have is that that is ineffective. So because the belief has already reached those who are willing to believe the conspiracy theory and they are now going to engage in spreading it more, it doesn't matter if they receive information to the contrary. In fact, one of the big conclusions from our book is that any intervention oriented towards the individual which includes inoculation where they counter facts, facts contrary to the conspiracy theory, before they get to conspiracy theory, those interventions are ineffective. The, the only lever that has any effect would be influencing the spread. Anything that slows down or reduces the reposting or posting of conspiracy theory can be significantly more effective than any individual level attempts.

[Alfred Marcus]

What about the economic incentives of the social media companies themselves and they, the way the institutions are set up so that there is a big incentive for them to get more people onto the social media so that their advertising revenue goes up. And, and I think there has been a decline in even their attempts to police themselves. And since Trump has been in office. It's really the root of it all those economic incentives and institutionally they're the are defined in the US where they take have no legal responsibility for what they put up. It's the responsibility of those who are putting it on the social media, at least legally as I understand it.

[Henrik Greg Rev]

Well, there is I think more or less unique US model where the rules are set by the owner of the platform as opposed to what you might call pretty unified European model which is similar to most of Asia model where the state sets the rules and those function differently because to the extent that the state sets the rules, harmful conspiracy theories, I mean for example the conspiracy theory that will incite people to into racial riots or to into not protecting themselves against a deadly disease or other forms of harm. I mean the state will generally view that as something to control. Similarly to, you know, having people put on seat belts in cars so they don't get killed controlling or making it harder for them discouraging them from smoking so they don't get killed. I mean these are, it's all the same thing. And the state would think of harmful conspiracy theories the same way and regulated the same way and it wouldn't be, that's not really Controversial, but now we're talking about outside the U.S. outside the

[Alfred Marcus]

U.S. are people receiving less conspiratorial messages? Are those messages being blocked outside the US in any systematic way?

[Henrik Greg Rev]

So at least the legal framework for doing so is in place. And there is, at least in the EU there is the potential for putting strong penalties, economic penalties on the platforms because that complicates things for the platforms because they need to be selective in Europe in ways they're not in the US or they need to hope that those penalties will not be imposed.

[Alfred Marcus]

Is that actually the case? Would it be possible then to conduct an experiment looking at populations in the US and Asia and different countries based on the institutional constraints, in the institutional incentives that social media companies have?

[Henrik Greg Rev]

Well, these are things I would prefer not to experiment on, especially because they would go against research ethics. But in terms of,

[Alfred Marcus]

I mean, with publicly available data, with what we know about opinions in the U.S. i don't mean experiments per se, you know, but

[Henrik Greg Rev]

so what we know from Ovid was that that caught governments by surprise and the legal frameworks to prevent this type of harmful distribution or distribution of harmful conspiracy theories was not in place. And so that one hit worldwide Europe wasn't safe either. Asia wasn't safe. All the legal framework came after political conspiracy theories.

[Alfred Marcus]

Is that being blocked or. And racism in general, Is that less likely to appear in social media? Are there ways of prohibiting it in other countries? And is it less prevalent in other countries than in the US it should be.

[Henrik Greg Rev]

Although I don't actually, I mean, working in Asia, I don't know the details of EU law, but I know that it has rather sharp teeth for violations. I mean, living in Singapore, Singapore law also has sharp teeth. So it, it wouldn't necessarily block, but it will impose penalties. And I believe that there are adaptations to that for the social media companies so that they avoid those penalties. I mean, this, what they do to stay safe outside the US is not something they would necessarily talk about. And in fact, all the security operations of those companies are not public for very good reasons, because they have security operations.

[Alfred Marcus]

Actually, if I originated a social media message in Asia, let's say in Singapore, I would be subject to different constraints than were I to originated here in the United States. But if I originated in the United States, wouldn't it be available for somebody for viewing throughout the world? Or would would it be blocked? Would the social media companies actually block it from you seeing it? Let's say in Singapore, blocking is certainly

[Henrik Greg Rev]

doable if they want to do it.

[Alfred Marcus]

If they want to do it. So they might not be doing it. Can't see other things. Before we actually get further into the book, I'd like to ask, you've such a great record as a researcher in behavioral theory and organization. This is a little bit of a departure for you in terms of your own career. Could you explain a little bit about the motivation and why you moved in this direction?

[Henrik Greg Rev]

It's just social responsibility. The COVID conspiracy theories were deadly. And so let's find out, we thought, let's find out why they spread so quickly and why they seem to be believed so firmly in spite of all the factual evidence to the contrary. I mean, hospitals were busy, people were dying, and people still were spreading, and other people were still spreading those conspiracy theories. So why should we do it, being business school researchers and not, you know, medical researchers? Well, we, we think we have pretty good tools for, for making the right kind of discoveries. Possibly we, we had some of the best tools available for it. And you know, if you combine the best tools with some responsibility in trying to discover the drivers of something harmful, then. And that's a good thing to do.

[Alfred Marcus]

You talk a lot about the idea of brick lodge and cultural toolkit. Could you explain what it means to treat conspiracy theorists as brickalores and how that changes the way we think about the people who create and spread these narratives?

[Henrik Greg Rev]

So it's. Brickleage is a modern way of thinking about culture. The, the, the more, I mean, often how people would think about culture is that you carry it inside your head somehow. It's a mental thing, and it comes from family, from school, from friends, maybe from a church. All of these things are put inside your head and they don't change. But what that overlooks is that culture is expressed, some of it expressed visually, but a big proportion of the cultural expression is through language. I mean, what we say has rich, rich cultural content. And it means that you can learn cultures very easily. I mean, those who are willing to learn a culture can pick it up fast. It's done all the time in the workplace and also when going into new social circles. And this really is bricolage. It means that culture, you find culture outside yourself and you, you incorporate, you start using those elements that you want to use, for example, to fit in or for example, because they appeal to you. And this is exactly where conspiracy theory comes in. People might be in an area of social media where there's a lot of them, so they start repeating them in ways that make them fit in, or the content of the conspiracy theory appeals to them, for example, there is something, an unpleasant fact in the world that they would like to not believe. They would like to engage in some reality denial. Covid is a good example. I mean we're talking about tiny, tiny thing. I mean a virus is almost not life, it's almost a mechanical thing. And this was a pandemic where humans originally, before the vaccines had no defense. You were going to be sick and some of the people who got sick were going to die. I mean that's just the way it was. These are not pleasant facts. And so it's actually appealing for the Brickoller to start thinking about how this all might be a lie or some other grand scheme, a conspiracy of powerful overlords who are who are engaging in deception and other tricks to make you believe things that are not true. This is what the Bricklauer is doing.

[Commercial Voice Actor]

Fourth of July savings are happening now at the Home Depot with select appliances starting at $398 plus get free delivery on appliance purchases of $398 or more. No membership required. Upgrade your kitchen with a modern and sleek GE Profile refrigerator featuring hands free autofill for the perfect pour every time and make laundry day easier with 2 in 1 washer dryer combo innovation that completes laundry in about 90 minutes. Shop Top brand appliances now at the Home Depot. Offer valid June 17th July at the US only C store online for details

Anticipate comfort on the court and after the match in Lululemon tennis gear. Breathable fabrics keep air flowing through long rallies. Dynamic stretch lets you go all out on every sprint, split step and return. And after the final point lands. Lightweight layers mean your day isn't done. Cool down, hit the clubhouse and relive each play. Shop tennis gear in stores and [e-mail]

bonjour compadre. It's the Priceline negotiator. How do I negotiate so many great travel deals? My greatest gadget? The Priceline app. It's got hotel deals, flight deals, rental car deals, all of those deals in a bundle deals, game day deals, concert trip deals. No one deals more deals than Priceline. Hold your horses, there's more. The app lets you filter hotels by neighborhood vibes, star level and amenities like pools and spas and beachfronts and. Wait, I'm not done. Stop cutting me up.

[Alfred Marcus]

Priceline Denial. We want to deny it dangerous this reality. Methodologically, the book is a computational book. Use a word in sentence embeddings to map the semantic space of conspiracy talk around Covid and black lives Matter for listeners who don't live. What do we allow you to see that we would miss just reading tweets one by one?

[Henrik Greg Rev]

So natural language processing can be applied to anything. For example, our conversation right now. There are all these things that we can find out for studying conspiracy theories. One thing it does is that it classifies. So you can look at statements or passages and pick out, is there a conspiracy theory in it? In fact, it's not uncommon that a single statement will have. Will involve more than one conspiracy theory. And that can be picked out too. It's quite sophisticated. The other one, the other feature is that massive amounts of text can be analyzed. And so it's possible to measure at the grand scale how conspiracy theories spread from who to who, how quickly, even if they have variations in how they express the conspiracy theory. Because a conspiracy theory is just a piece of language. But we always say things with some variation. But natural language processing cuts through the variation so that we can say, oh, this person is repeating the same conspiracy theory. The last feature, which is a very powerful one, natural language processing can measure distances between statements. And so it's possible to say that two conspiracy theories are near each other and another pair of conspiracy theories are further apart. So we can start checking. Is it true that you'll be more. A person will be more likely to go short distances in conspiracy theory space, as it were, than long distances. So, for example, to wander among all the COVID conspiracy theories pretty easily, but having greater Difficulty jumping from COVID 19 conspiracy theory to Black Lives Matter conspiracy theory. Because these have different nature. I mean, one is about the pandemic, the other is about how some protests that naturally happened are in fact orchestrated by some. By some entities.

[Alfred Marcus]

The natural language processing programs that you use, these are, I think, are they publicly available for use? They've been developed, I think, by machine learning specialists. I'm not incorrect. I've been working with somebody who's a machine learning expert, and we were thinking of doing. Using one of these. He showed it to me that you could use to analyze, let's say, Fox and the new shows at Fox and MSNBC and cnn. We were thinking about doing the projects. So can you tell me a little bit about what the natural language processing programs that you used were and how you found them and how other researchers might be able to take advantage of them.

[Henrik Greg Rev]

All of them are publicly available, nothing proprietary. And so we, we document that in the book. It's quite easy to. No, it's not easy to use it. They need parameters to be set Correctly and so on. But it's natural language processing now is at the stage where you can do exactly what somebody else is doing. You can replicate what they're doing, but you can also say, okay, what happens if I use slightly different techniques? Will I then get different results? What happens if I get a different data set? Will I then get different results? It's like this is regular science. It functions very well, but you can

[Alfred Marcus]

take any text and you can analyze it and you can characterize it and classify it in various ways if it's big enough. Yeah. So methodologically then it was a. Basically your book's computational book. You use word and sentence embeddings to map the semantic space of conspiracy talk around Covid and Black Lives matters. For listeners who don't know that much about nlp, can you tell us a little bit more about the tools and see what we would miss just reading tweets one by one. And I guess you were, you were reading tweets, right? That. That was that. What else was your subject matter was tweets. And what other social media were you looking at in the book?

[Henrik Greg Rev]

We were, we're purely looking at tweets because that was really the dominant social media during that time and also a very active one for or COVID 19 conspiracy theories. And the way it works is actually it's a bit of an interaction between human and machine effort because these are our classifiers, which means that there has to be a human who classifies collection of statements. So some of them belong to one conspiracy theory. I mean, for example, Bill Gates is using the COVID 19 pandemic to what? There are many versions of it, but the main one was that the purpose was for vaccines to inject a microchip into the bodies of people. I mean, those one example, a variety of the Bill Gates conspiracy theory. And so the human classifies and the NLP software then goes through the ones that are classified by the human and says, okay, I can now by using the word embedding techniques, which really looks at how words appear next to other words in the statement, taking off the trivial words like the and is and so on. And so the NLP software then says, okay, so now let me classify. And then I'm going to show my classification to the human again. And the human then checks the NLP classification and you know, says, okay, this is pretty good, but I can see some problems with it. And then does a reclassification. And so there's a bit of a conversation, human versus machine. And once, once the agreement is fine enough, then the machine is turned loose on everything. And this sounds like a lot of work. It's actually, I mean, on the regular researchers timescale, it is a lot of work because classifying, classifying, classifying. But compared to the effort invested in analyzing such giant data sets, it's tiny. It's very easy because at some point the machine takes over everything and it's actually more reliable than the person.

[Alfred Marcus]

So at the beginning you're setting the parameters the machine doesn't, and you work with the. So it's a little bit like the way AI learns because again, it's because of word being close together and their appearance close together. And it can be used on any text. We used to call this content analysis, I think at one time do it. It's more sophisticated, I would think, isn't it? Or is it not?

[Henrik Greg Rev]

So, I mean, let's keep in mind that what I'm describing is what we used, which is a classifier. There are many, many different things that NLP can be doing, but we use classifiers because that's how you identify conspiracy theories out of text that contains a lot of stuff that isn't conspiracy theories.

[Alfred Marcus]

I think there's some actually programs already set up that you can use to identify these. I think we had, when we were thinking about doing our work, we were thinking of, of using them. A big move that you make is to replace the metaphor of falling down a rabbit hole with what you call semantic cascades. What does a cascade look like in practice? And how does someone move from one conspiracy theory to portfolio them over time?

[Henrik Greg Rev]

So the English word cascade, if you rewind like 2000 years, there was the word castiera, which means repeatedly falling down or tumbling, which is exactly the meaning we have. Once somebody has adopted or started propagating a conspiracy theory because they've been persuaded, then they are very likely to fall again. In other words, if there's another conspiracy theory nearby or nearby means it's similar, then they will also pick that one up and start repeating that one. And if there's yet another one, they will do it again. So they keep falling and falling, they keep tumbling. And of course by doing that, they're spreading these conspiracy theories in their own language through their own account or through reposting what other people are doing. And so then other people are also falling. And so this is a process where, you know, within each person they tumble and they make other people tumble as well. So caschiera falling and falling over again.

[Alfred Marcus]

And there's certain themes that are Being picked up by the natural language processing that you can see being used again and again. They have a set structure. You talked a little bit before about the motivation of people who are receptive to conspiracy theories, but what about the motivation of the people who are creating them? Is it monetary or is it some kind of psychological impulse drive? Have you thought about that? I know it's a little bit outside of what you actually did.

[Henrik Greg Rev]

We don't know anything about the creators and there's not that many conspiracy theories around. I mean, we were studying less than 20 of them. There are actually more, but some of them are just hard to identify because they overlap too much with non conspiracy content. But if you want to hear my suspicion with no evidence behind it, this is easy to do. The human mind is creative. So it's not whether conspiracy theories are hard to create or not, it's whether they get any pickup or not. So what we're doing in studying how they spread is actually the main way that they should be studied. Because it's the spread that's important. It's not really the creation.

[Alfred Marcus]

Some examples of. You said there are 20. You give some examples of the 20. The structure of the 20. I guess it's the structure of how they're put together.

[Henrik Greg Rev]

So it's. Well, one way of thinking about to explain the structure is that some of them were actually quite simple and they're not crazily distant from, from, from reality in some ways. So there was a conspiracy theory that there's lots of false positives so that the reported rates of infection weren't true, and this is something that can actually happen. I mean it didn't happen, but it can happen. And, and that actually was one of the gateway conspiracies. Meaning. Because it sounds plausible. Like actually I have omitted the part about who were the ones conspiring to have that happen because that would vary depending on who was telling it. Right. So it has many variations, but you can conceive of people believing that. But once they get into gateway conspiracy theory, like one of those simple ones, all of a sudden they became more, they started tumbling as we talked about.

[Alfred Marcus]

Right.

[Henrik Greg Rev]

They had more conspiracy theories, became plausible and they, they came into a space where there are some really unusual ones. The, the Bill Gates conspiracy theory is, is definitely freakish big, but it, there is, you know, some sort of grand design of, you know, implanting people with monitoring devices by creating a fake pandemic that will force us to vaccinate ourselves. But the vaccination isn't real. It's Just a device of something else. There was also some. Somewhat of a tunnel like if you know from, from the movie. Conspiracy theory of this was a way of reducing, you know, the human population. Was there too many of us. It's all sorts of very extreme conspiracy theories were, were found. We, we studied some of them, the ones that were most frequent and most easy to. To. To identify. But really there was a fair amount of creativity in, in conspiracy theory creation

[Alfred Marcus]

with the plot of the Chinese. That was one that I, it was very common too. But.

[Henrik Greg Rev]

Oh yeah, and not a complicated one because it's. It happened to start in China and, and so I mean there is. And that idea of the Chinese having some sort of evil plan, I mean it overlaps with some motives that are racist. And so there is a motivation between that one as well. And some of the conspiracy theories, in terms of the people named as the conspirators are obviously anti Semitic.

[Alfred Marcus]

The cascading part of it is also reinforced by the media companies or Twitter because they create these echo chambers and once they know what your interests are, they feed you back your own interests. And so couldn't they, if you're able to do this with Hagi and create a catalog of 20 conspiracy theories, couldn't the social media companies or X right now or any of them, Facebook or Instagram also do this and use it as a way of policing themselves? And yet they are not doing it because of the obvious financial motives of getting where eyeballs are people frequenting their, their, their sites? Or am I being, am I being conspiratorial

[Henrik Greg Rev]

in terms of what they can do? And, and possibly they, they actually do it without talking about it. We don't know any, any detection and prevention of the spread of conspiracy theory requires, requires some human intervention. Because this is not something that, I mean a machine can't look at text and say this is a conspiracy theory that requires human detection. But, but it's, it's. It's not a big investment to have people monitor, say memes that suddenly become popular and check if any of them look problematic. And the same people can do exactly what we did. A classifier can train, can be trained in a matter of hours if you put enough people on it. And once that's done using the same tools that we used, they can label each conspiracy theory message and then do with that message whatever they want, including suppressing it.

[Alfred Marcus]

A warning label. So they could put a warning label on it and say that you are being subject to outrageous ideas that you should be careful about. And this is the source and the truth. Content here is very, very weak. It's fantasy. You may want to see, you may want to subject yourself to fantasy, but be aware of what you're subjecting yourself to.

[Henrik Greg Rev]

Yeah, I mean there's some cost associated with having mechanisms like that, but it's not particularly expensive compared to the revenue that these platforms are getting. So it's mostly a matter of willingness.

[Alfred Marcus]

From an ethics and governance perspective, platforms, they are in a difficult position. Conspiracy talk is harmful, but it's not illegal. So based on your work, where would you draw the line between legitimate doubt? I mean, how did you do that even in the work you did? Citizens questioning authority and issues also I guess of free speech. That's what the media companies tend to justify themselves with the laws around free speech. What is your view on their moral obligation to restrain themselves? It sounds like you think they could do it and they just don't.

[Henrik Greg Rev]

Again, this is the basic fundamental problem with conspiracy theories is these are non factual things and they're non factual in worrying ways because the underlying thinking is that facts that in real life are unconnected, they become connected into some grand scheme where there is this evil entity or grouping that's trying to do something harmful against the believer of the conspiracy theory. Now that's, that's already problematic because sort of intent to harm somebody that usually leads to a defensive response of some kind. I mean just as another example of a conspiracy theory, the COVID 19 was a lie in order to keep people indoors for the installation of 5G machinery. So the 5G stations. And a side effect of 5G stations was actually to make people feel sick. But of course all of these were installed for the purpose of profit and sort of against the interest of the common man. And guess how regular people reacted to that conspiracy theory. Well, they burnt down a few 5G stations, but there was also violence against the people installing them. And this, this is not surprising this. I mean whenever there's intent to harm then there will be some sort of potentially violent reaction. And conspiracy theories. One, one of the things that they typically have is some intent to harm. And then there are the extreme conspiracy theories in terms of. Well, extreme in terms of the danger they can potentially impose. One of them, for example, was that COVID 19 wasn't real. And so you don't need to do any personal protection or protection against of others. There was also conspiracy theory, this is a while ago, but there was a conspiracy theory about hiv, about AIDS not being disease spread in the way we know it to be spread. And that conspiracy theory was connected to some pretty lax rules around AIDS in South Africa, which we know led to a significant increase in deaths. So these can be harmful things at different scales, but we need to consider the harm every time.

[Commercial Voice Actor]

Who cares about your poops? Ollie does. That's why Ollie's science backed Gut Health lineup helps support your family's regularity, from daily probiotics to fiber gummies your kiddos will love. Find it all on olli.com that's o l l y.com these statements have not been evaluated by the Food and Drug Administration. This product is not intended to diagnose, drug, treat, cure or prevent any disease.

Uncovered windows can make your home feel up to 20 degrees hotter. Stay cool and save up to 50% off custom window treatments during the 4th of July mega [e-mail] from outdoor shades to room darkening blinds, finding the perfect fit is easy. Get free samples, expert design help and professional measure and install services or DIY with confidence and support every step of the way. Shop up to 50% off site wide plus huge savings on door busters right now during the 4th of July mega

[e-mail] you want to get your backyard summer ready, but you don't want to break the bank. Wayfair gets it. Planning on dining al fresco or relaxing poolside, Wayfair has everything you need to prep your space. Shop now and save up to 70% off during Wayfair's 4th of July clearance. Score huge deals on outdoor furniture, area rugs and and more. We're talking thousands of products for every style and budget, plus surprise flash deals. July 6th. Don't wait. Shop Wayfair's 4th of July clearance now through July 6th at Wayfair.com Wayfair Every style, every home.

[Alfred Marcus]

I think about this there. It's a system. And the system consists of people who are receptive to conspiracy theories, people motivated to create them. Maybe it's just for attention. Maybe it's not monetary. Maybe it's because or they have evil intent of some kind because of their psychological nature. But then there's the gatekeepers are the companies that stand between one and the other. And I guess part of the system too are governments who have the capacity to regulate the system if they were so inclined. And I guess if we think about levers for changing, for change and for combating it, where do we turn? Where would you turn if you had that capacity to combat it? Right now? Let's say you were advising an organization that was trying to combat or combat something. Let's Say like antisemitism right now, which is being spread as a conspiracy theory. What advice would you give?

[Henrik Greg Rev]

So the mechanisms have to be put in place to prevent the spread. That's the only effective lever. And the manual way of doing it is exactly what I described. You have some detection team that classifies and does something to suppress. There's also other ways that are more automatic. It's there though those accounts that are more popular and who are given a broad spread of all whatever they post, it can be useful to give them reminders that do you really want to spread this content? It's like a reputation warning. Is it good for you to spread this content? Which seems to be problematic. Those types of repetition warnings is actually in modeling we've done is a very effective sort of slowing down of the conspiracy theory spread. But it puts some of the onus on the popular account. But keep in mind that those popular accounts, people who spread conspiracy theories are just people. There's nothing special about them. And so if they see a conspiracy theory and consider spreading it, it's probably not with ill intent. They actually think that this might be true. Which means that the warning can be

[Alfred Marcus]

effective, that trying to harm their reputation doesn't really work because such people want attention and you're just giving them more attention. And even if it's negative attention, they glory in it. And given the laxity in government regulation and the laxity in the willingness of the companies to police themselves, that and given also, at least in the United States, freedom of speech, that the best way to counter conspiracy theories is not to direct your message against those who are the creators of it or even their specific charges, but to put good information on use social media to provide accurate and truthful information. And it's not very strong, but that may be the best way to counter it. Do you think that makes any sense?

[Henrik Greg Rev]

The evidence we know says that that's it's not completely ineffective, but the effect is near trivial in magnitude. The only any effective approach is one that intervenes in the spread.

[Alfred Marcus]

Why somehow either in government or the companies themselves to self regulate to bring about a change. Are you aware of any efforts to do that right now that you think might be workable in the current environment?

[Henrik Greg Rev]

Well, we're hoping that at least some policymakers will read our book because we give a very clear description of what the science says about the different ways that conspiracy theory spread can be reduced. We understand that there is a belief that facts help and that education against them help and various things. It's just that Those beliefs are not really backed by facts.

[Commercial Voice Actor]

The.

[Henrik Greg Rev]

I mean, all of these, These things help, but they help to such a small extent that they're really insufficient. And so if. If the idea that facts help is combined with an ideology of free speech applies to any kind of speech, I'm sorry to say that's a society that's extremely easy to manipulate.

[Alfred Marcus]

I tend to agree with you. But I know that there are people who are very concerned if there are any controls on speech, it will be used for the wrong purposes by the wrong groups. One person's definition of free speech is another person's definition of thought control. But I think you and I share a different perspective. What would your perspective then be on what kind of regulation could really help under these circumstances? Circumstances?

[Henrik Greg Rev]

Regulation is. The effectiveness of regulation depends on how well it controls the platform owners and the platform behavior. Because regulation doesn't really do anything in and of itself. It's the mechanisms that's embedded in the platform that matters. I think it would be overly interventionist to require platforms to have a list of things that platforms need to do. But if you look at, for example, how US Regulation works, it is possible to impose disproportionately large penalties for harm done by any commercial entity. I mean, this is just any. US Regulation has that feature. It is not intended to catch many of the wrongdoers. What. Whoever is caught is penalized disproportionately. And that type of regulation, basically making them liable for harm will mean that the platforms will develop the most effective way of. The most cost effective way of reducing harm.

[Alfred Marcus]

Right now in the United States, as I understand, the legal structure of the way things work is that if a company, if you post on YouTube or if you post on X, or if you post on Instagram or any of the social media outlets that you are responsible for what you've posted, and there's no response. The companies are considered just platforms, so they bear no legal responsibility. They cannot be sued for any damage that's caused on their platform. So that's a huge constraint, I think, right now in terms of policing them. What about if there was a. If we got together and took your 20 themes of conspiracy theories and we started a nonprofit and we began to police and publicize people who were issuing conspiracy theories, and not only the people who were issuing them, but the actual content of them a little bit in academics right now there are websites that try to discover fraud and publicize them. And actually, I think they've been very effective in raising consciousness to this problem. Does something like that currently exist? And do you think it would be a useful addition to the tools that could be created to combat this problem?

[Henrik Greg Rev]

Also ineffective, because I mean, conspiracy theories, the type of conspiracy theory you'll see depends on the situation. I mean the conspiracy theories against the Black Lives Matter movement popped up all of a sudden. I mean they have a right wing origin and so we understand why they, they came about, but they, they were suddenly very powerful and they spread very fast. Any type of sort of academics or other outsiders trying to police them would again fall into the category of fact checking. Basically saying this isn't true. Sort of it's not true that, you know, the big financiers made it easy for people to riot against the police, that they bust them into the cities, that they put loose bricks nearby so they could throw them through windows. I mean, I'm telling you, the conspiracy theories popped out against Black Lives Movement and they, they were all of a sudden all over the place and stopping or slowing it down. Clearly the, the, the responsibility lies on the platform and that means it requires that the platform is held liable for what it spreads because the platform is spreading it. Yes, people are posting it, but the platform is spreading it. It's different from what we do in science because scientists are individuals and groups who have reputations to maintain. And whenever I do something, my name is on it. In social media, it's an account name and that account name may or may not be related to the actual name of that person. There may in fact be no person at all because the account name refers is a bot and part of a bot farm. So individual responsibility isn't really there.

[Alfred Marcus]

So ultimately all we can do is hope that the laws will change and, or the social media companies will see that what they're doing is imminently harmful to themselves in some ways as well. But politicians themselves take advantage of these conspiracy theories. So the likelihood, I think we're going to be subject to this for a very, very long time and it can only get worse. Is your conclusions seem kind of pessimistic and maybe that's the reality. Or do you have any hope for us weathering this storm and combating it in any way?

[Henrik Greg Rev]

I mean, the best known mechanism for changing loss is known as voting. That one has been in place for a while. Not really any other immediate solutions that I'm aware of.

[Alfred Marcus]

I mean, we're in an era that what is true and what isn't true, stretching it and white lies and beyond white lies, actually making factual statements, putting them in various Contexts, it's very hard to return to conventional sense of truth where things are flying left and right out of everybody's mouth and people have vested interests in making other people believe all kinds of things. It's rather depressing, actually.

[Henrik Greg Rev]

I mean, it's a repeat of something that we have seen, at least in Europe, because the really social media, the, the social media structure, if you, if you think about how they, they build up some to be very influential because they have proven popularity. They also group people together because they seem to have shared interest and interest in the same content. It really, the, the modern social media is remarkably similar to, to society in medieval Europe. So because you, you have the priests that many people listen to and then you had the bishops that many people listen to and also priests listen to them. So these would be super spreaders in social media. Then there would be in the countryside, you'd have all the villages and hamlets and people there spoke to each other and didn't speak to strangers much. In the cities and towns, you would have the neighborhoods. You also had the guilds and the societies and people who grouped together, who shared what they thought were the truth with each other. And so there wasn't really much of a truth around. There was many, many truths depending on where you were and who you talked to. Now medieval Europe did have a lot of wars, so it's not a society we generally wish for. But the organization of social media is quite familiar. If you know European history well, there's

[Alfred Marcus]

a book by a scholar by the name of Majda Teeter, she's a Polish scholar, and her basic argument is that the printing press didn't bring enlightenment. With the spread of the printing press, prejudices became much more common throughout the world. And her argument is that every new media which allows people to receive more information quicker, spreads the worst in the human spirit rather than the best.

[Henrik Greg Rev]

Not sure I want to endorse that one, but

[Alfred Marcus]

yeah, this has been a great conversation. Why don't you just. We can end it by saying, what are you working on now? Where does your passion lie at this point in time? You've had such an outstanding scholarly career. I think so many of us are in admiration for what you've done. So what do you, what are you doing right now? It's absorbing your time, attention, your passion.

[Henrik Greg Rev]

I wonder what the workplace will look like, ah, in the near future because I mean, there are many of the usual tasks that regular workers do can be done differently, maybe in good ways. And many of the things that managers do might be done by somebody else automatically. Because the generative AI revolution is something that certainly a lot of companies have high hopes for and they're already starting to react to it. So we're tracing what they actually do. There is a possibility that what a given current company is doing will can be done with significantly fewer people. And then the question is what to do then? Do we work less, do we make more stuff with more companies? Or is there some other way forward? We don't know.

[Alfred Marcus]

I think we don't know. But the conventional argument is that it leads to a decline in demand for people like lawyers, like many white collar workers, people who do programming and so on. But it also, because of the rapidity with which tasks can be done, I think it can create new demands that we currently don't have. And I think that's often missed in the equation because it's very easy, for example, to start a new business, to get the idea for the business. And so it's going to create new demands as well as take away the demands for old types of work.

[Henrik Greg Rev]

We're thinking about how to study that. We know that existing firms are largely using it to shed workers, especially managers. But there's also the other side of the story. What do entrepreneurs use it for? Will it become easier to create new companies? And in two, three years maybe there will be answers to that because we're starting to study Great.

[Alfred Marcus]

This has been a very interesting conversation. Once again, the book is Control Alt Doubt by Hagi Rao and Henrieth Greve, published by Oxford University Press. It's a fascinating and unsettling look at how conspiracy theories are built and spread through language and what it means for platforms, organizations and democratic societies. For listeners who are interested in how questions of trust, governance and responsibility play out inside firms, you might also look at my book Comeback, which examines companies trying to reverse decline by bringing their strategies and ethics back into alignment. Henrik Greva thank you for joining me. And thanks to all of you for listening to on the Cusp on the New Books Network. If you have suggestions for future books or guests or comments, please email [e-mail] amarcusmn.edu thank you for listening to this

[Commercial Voice Actor]

episode of the New Books Network. We are an academic podcast network with the mission of public education. If you liked this episode, please share it with a friend and rate us on your preferred podcast platform. You can browse all of our episodes on our website newbooksnetwork.com Connect with us on Instagram and Bluesky with the handle ew booksnetwork and subscribe to our weekly substack [e-mail] to get episode recommendations straight to your inbox.

[Alfred Marcus]

Hey, it's Ryan Reynolds here from Mint Mobile.

[Henrik Greg Rev]

Now, I was looking for fun ways to tell you that Mint's offer of unlimited premium wireless for $15 a month is back.

[Commercial Voice Actor]

So I thought it would be fun

[Henrik Greg Rev]

if we made fifteen dollar bills, but

[Commercial Voice Actor]

it turns out that's very illegal. So there goes my big idea for the commercial.

[Henrik Greg Rev]

Give it a [e-mail] Switch upfront payment

[Commercial Voice Actor]

of $45 for 3 months, $90 for 6 months or $180 for 12 month Plan required $15 per month equivalent taxes and fees Extra initial plan term only greater than 50 gigabytes. Me slow when network is busy. C terms.