← Question Everything
Question Everything
Can a Chatbot Convince Conspiracy Theorists of the Truth?
00:26:49
A psychologist whose brother believes in conspiracies runs an experiment.
Loading summary
Transcript104 lines
- [00:01]
A
Something I'm always thinking about is the way that algorithms are often self reinforcing, giving us more of the content and news that we want to see, which isn't the most healthy way to take in information. Which is why I'm excited to be partnering with Ground News. Ground News shows you how the same story is being covered across the political spectrum so you can actually get the full picture and not just the one version that's being given to you. If you want to check it out, go to groundnews.com/ to get 40% off their unlimited vantage plan. Again, again, that's groundnews.comquestion. make sure you use our link so they know we sent you.
- [00:33]
B
On the broad side, we take you into the heart of the south with stories that'll surprise you. Bigfoot apparently loves glow sticks. He likes to party.
- [00:43]
A
Exactly.
- [00:43]
B
Exactly. He's a raver. And topics that dig into the muddy margins of history. Right.
- [00:49]
A
The good, the bad, the ugly. It's not clean at all. It's so messy. Wait a second. This. This is actually real.
- [00:55]
B
Listen to the broadside. One story every week exploring the rich traditions of the South.
- [01:09]
A
A few years ago, Gordon Penny Cook was having a conversation with his older brother that a lot of people would have tried to politely escape. His brother had leaped down the rabbit hole of right wing conspiracy theories like QAnon and become pretty consumed by them. But Gordon's a psychologist and a good brother, so he listened.
- [01:27]
C
We were at a campground in rural Saskatchewan, and he's telling me this story about how he thinks Republicans are gonna, like, take up arms. He was pretty sure that during the Biden years, what the government was doing is so egregious that eventually it'll be the last straw and, like, regular Republicans are gonna take up arms and it's gonna be civil war in the streets.
- [01:45]
A
He, like, very sincerely believed that.
- [01:47]
C
He sincerely believed that. When I first told him that I was moving to New York State, he thought I said meant New York City. His eyes got big and I was like, well, I'll be upstate, so that's good because you're not going to want to be in New York.
- [02:03]
A
QAnon at the time included stories about some event called the storm that was supposedly imminent and how a majority of Americans would go through a great awakening and realize a secret cabal was running things and determined that they have to go fight. On this camping trip, Gordon says he tried to gingerly nudge his brother away from these kinds of conspiratorial beliefs.
- [02:23]
C
He said, okay, well, what percent of people do you think would be willing to literally, like, take up arms and start shooting people because of their, like, political grievances? And he's like, I'm not sure, but it's more than 50%.
- [02:35]
A
Whoa.
- [02:36]
C
Yeah. So he. That was his. But that was the world he was in. Like, it was a very kind of like, angry, violent world. Yeah.
- [02:41]
A
I find just every one of these conspiracies, each one is almost like a Lord of the Rings volume in a way. Like, each conspiracy has an entire world, a cast of characters, a history, a mythology, languages.
- [02:54]
C
Yeah, they're steeped in it. Their information ecosystem is so stilted that they can form a kind of coherent view of the world that is totally disjointed from reality. It takes a lot of effort to break into that.
- [03:08]
A
Gordon is uniquely qualified to try and persuade his brother out of conspiracy theories. He's an associate professor of psychology at Cornell University who studies specifically why people believe things that aren't true.
- [03:19]
C
Or as he calls it, the Science of Human Stupidity. That's the course I teach.
- [03:25]
A
That's the official title.
- [03:26]
C
That's the official title of the course here at Cornell.
- [03:28]
A
Okay.
- [03:29]
C
Yeah. They let you just name it whatever you want.
- [03:32]
A
It's a title that's meant to get students to stop as they're browsing the course catalog. But Gordon doesn't intend it to be mean spirited towards his brother or to anyone in particular. What he's trying to investigate in his work is, why do humans make choices that aren't good for us?
- [03:46]
C
I want to understand why, like, I'm stupid and why everybody is stupid. Like, what are the mistakes that we all make? If you think about all the, like, biggest problems that we face as a species, they're problems that we've, like, made for ourselves, like decision making problems, climate change, all these things. So if you want to, like, fix these things, you have to understand the underlying causes, which is human stupidity.
- [04:05]
A
And that includes believing unfounded conspiracy theories. Whether it's election denial or Pizzagate or theories about the assassination attempts on Donald Trump being inside Jobs or QAnon, conspiracies are infecting our lives and society in all sorts of toxic ways. Gordon wasn't able to change his brother's mind, and he kind of figured that's just the way things are in general. People don't change their beliefs much. And you can imagine conspiracy theorists being even more rigid. Certainly as a journalist, that's been my experience. It can be very frustrating and demoralizing to try and get someone who buys into conspiracy theory to accept facts that undermine their belief. But Gordon, bless him, wanted to check that assumption scientifically.
- [04:49]
C
We wanted to test the basic idea that I think most people hold, which is that people who believe conspiracies will never change their mind, no matter what.
- [04:58]
A
And did you have a hypothesis?
- [04:59]
C
Yeah, our hypothesis is that they wouldn't.
- [05:02]
A
So Gordon and his colleagues came up with an experiment, and what they discovered ended up changing Gordon's mind and mind, frankly, about how we should think about conspiracy theories and the people who believe them. From Placement theory and kcrw, I'm Brian Reed. This is question Everything we investigate how the truth gets buried, distorted and denied and the ways people are fighting to make it matter again. Stick around.
- [05:29]
D
Artificial intelligence is moving very, very fast, and it's raising new questions just about every day about what it is, what it isn't. When all is said and done, what is the end game? I'm Chris Hayes, and as part of my podcast, why Is this Happening? I'm speaking with leading experts each week to help ground that conversation.
- [05:44]
A
We're right now in a situation where it's very difficult to understand what is real and what's not real.
- [05:50]
D
Why is this happening? The AI Endgame, a special miniseries from Ms. Now. Start listening today, wherever you get your podcasts.
- [06:01]
A
Gordon wanted to run an experiment to see if it's true that people who believe conspiracy theories rarely change their minds about them. So he and his team recruited more than 2,000 people from around the country. Everyone was paid to take part. They knew the experiment had something to do with conspiracies, but not the specific purpose. They did it from their computer. During the survey, they were shown a prompt. From your perspective, what is a significant conspiracy theory that you find credible and compelling?
- [06:28]
C
People come in and they can describe a conspiracy theory that they believe in in their own words. And then what they do is they give us reasons why they believe it.
- [06:37]
A
Some people said they didn't believe in a conspiracy theory. Others said they did. Some believed conspiracies about the JFK assassination, alien visitations, Covid, the 2020 election being stolen, Walt Disney being cryogenically frozen. He wasn't. Gordon and his team then fed each participant's theory and the reasons they gave for believing it into an AI model, ChatGPT4 Turbo.
- [06:58]
C
And then what the AI is told to do in the back end by us is to debunk the conspiracy, to provide strong facts and evidence, to counteract the view that they're stating and to give them stuff to think about.
- [07:10]
A
Basically, they called their Tool debunkbot. At the beginning of the study, each participant also rated how strongly they believed in the conspiracy theory on a scale from 0 to 100. 100 meaning total complete certainty. Then they talked to the chatbot about it. Just three messages from the human and three responses from debunkbot. And after that, they rated their belief level again. So here's one actual example from the study. One participant came in and said they believed that the US government staged the 911 attack. And they rated their conviction in this conspiracy at 100. They report at 100 belief. So that means they think that this conspiracy is true.
- [07:50]
C
To be 100% confident means you really think there's lots of reasons why that's true.
- [07:54]
A
They said one reason they believed it was an inside job was because they'd watched that famous footage of an aide whispering news of the attack in George W. Bush's ear while he was reading to an elementary school classroom. And this participant thought it was weird that the President just kept reading and didn't seem alarmed. Also, they pointed to World Trade Center Building 7 as proof that something fishy was afoot. This was a building near the Twin Towers that also collapsed after the attack, even though a plane didn't crash into it. The participant had seen videos pointing to how suspicious that was. If it wasn't the planes knocking down buildings, someone else must have been behind it, right? After watching a lot of these videos and shows, the participant wrote, I concluded that the conspiracy theorists might be correct. Then the debunk bot responds.
- [08:38]
C
So it says, thank you for sharing your thoughts and concerns about the 911 attacks. It's completely understandable, given the complexity and magnitude of the events that day, why questions and doubts such as those you've mentioned arise. And then it addresses all of the points the person raises with direct counter evidence.
- [08:54]
A
It gives a response to this idea of President Bush's reaction. It says it's essential to understand the context. When he was informed that a second plane hit the World Trade center, indicating America was under attack, the he chose to remain calm in front of the children to avoid causing panic.
- [09:10]
C
Yeah.
- [09:11]
A
As for what happened with World Trade Center Building 7, the bot says it's true the building collapsed even though it wasn't hit by a plane. But the AI explains that the National Institute of Standards and Technology did extensive investigations and found that Building 7 was damaged by debris after the North Tower fell, which caused fires in the building structure, which is what led to its collapse. Now the participant gets to respond again, and they are not convinced.
- [09:36]
C
They say that May be true, but the evidence clearly shows the Twin Towers collapsed due to demolition experts explosives being deliberately placed to cause a collapse.
- [09:45]
A
So here the person is adding something else. They're saying World Trade Center, Building 7, falling without being struck by a plane isn't the only thing that's odd about it. They're claiming they've actually seen supposed evidence showing that the collapse was caused by explosives. They also tell the chatbot that they don't believe the heat of the fuel in the jets that struck the Twin Towers was hot enough to melt the steel girders in those buildings to cause them to fall. So they're bringing up new claims. I learned there's a name for this move in a debate. It's called the Gish Gallop. It's when someone bombards their opponent with a bunch of different arguments, unsupported ones, that make it hard to keep up. It's coined after a guy named Dwayne Gish who used to try and convince people evolution wasn't real. In response to this participant's gallop, the AI says, if you're going to demolish a building that large, you cannot do that surreptitiously. There are a lot of materials required. It would take months of preparation. You'd need to evaluate the structure and place the charges at precise locations. The AI also goes into depth about all the physics of steel and how it doesn't have to fully melt for the building to fall. And to that, the participant responds, why
- [10:51]
C
then did we allow Iraqi men to enter our country and give them lessons on how to fly a plane? They also got past security fairly easy, without question.
- [10:58]
A
So they pivot to like something else completely here.
- [11:00]
C
Yeah. So again, another gallop away. But then the AI just explains that, like all the things that we do now at the airports, taking our shoes off, whatever, those are all post 911 things.
- [11:09]
A
So one by one, the chatbot takes on the arguments and gives the participant all sorts of facts to keep up with the gallops. And then how does it end up?
- [11:18]
C
After three rounds, the person says, thank you for your response. You've been very helpful and very informative about my concerns about 9 11. And then they re rate their belief. Where previously it was 100% certain that 911 was inside job, they're now down to 40%.
- [11:33]
A
Wow. Gordon says any rating below 50% shows that the person no longer holds that belief. It means they think it's more likely false than true. So this person went from being 100% certain that 911 was an inside job by the US government to thinking it was probably not an inside job. All in the course of three short back and forths with a chatbot. Gordon couldn't believe what he was seeing. Over and over again across roughly 2,000 participants. The same thing kept happening. Even with fresher conspiracies, like the 2020 election being rigged against Donald Trump. People were changing their minds, an astounding number of them.
- [12:13]
C
So the biggest finding is that after like eight minutes of conversation with the AI, everyone who believes the conspiracy, 25% of them don't believe it anymore after the conversation. So like a quarter of the people change their mind, which is just way, way more than we predicted.
- [12:30]
A
This is a big number. Other studies which don't use chatbots show that people are very resistant to changing their minds. One of these studies involved political ads. Another used social media. They showed much more modest shifts in people's beliefs. A quarter of people no longer believing a conspiracy theory after chatting briefly with an AI model. It's significant. Imagine if you could spend less than 10 minutes sharing some research with your conspiracy brained uncle at Thanksgiving and he'd say, you know what, you make some pretty good points. I was wrong. And not only that, Gordon says the change stuck.
- [13:05]
C
We asked them again two months later what their stances, and not only is the effect still there after two months, it doesn't like decay, it doesn't get weaker. They still are at the same point, which indicates that they really did change their mind.
- [13:17]
A
That's remarkable. Like, that's almost unbelievable.
- [13:20]
C
I know. In fact, when we got the data, we were like, there's a mistake. Like, this is messed up. And then we, we went through it again. Of course we replicated it. It's the, it's, it's a real thing.
- [13:36]
A
After a quick break. What is going on here? Are humans more convinced by machines than by other humans? That's in a second.
- [13:49]
C
At Radiolab, we love nothing more than nerding out about science, neuroscience, chemistry. But.
- [13:56]
A
But we do also like to get
- [13:58]
B
into other kinds of stories, Stories about policing or politics, country music, hockey, sex of bugs.
- [14:06]
C
Regardless of whether we're looking at science or not science, we bring a rigorous curiosity to get you the answers and
- [14:13]
B
hopefully make you see the world anew.
- [14:14]
A
Radiolab Adventures on the Edge of what we think we know.
- [14:18]
B
Wherever you get your podcasts,
- [14:24]
A
Gordon and his team had made an incredible discovery that an AI chatbot instructed to rebut a user's conspiracy theory could convince them out of the conspiracy theory. A quarter of the time in a really short conversation. They replicated the study multiple times, same findings. It's nuts. So nuts that honestly, it had me grasping for conspiracies. To try and explain it, Gordon and his colleagues published the raw interactions between the participants and the chatbots. And I read through a bunch of them trying to figure out what the chatbots were doing to be so successful. The interaction seemed so straightforward. After all, they were limited to only three back and forths. So it's not like these are hour long debates. The AI comes back with counterarguments and boom, there's an impact. So what does Gordon think is working here?
- [15:12]
C
The most important thing is that it addresses all of the points the person raises with direct counter evidence that is like authoritative. It's very hard for a regular person to do this sort of thing, to provide such strong counter arguments to very
- [15:28]
A
specific points like just think about it. In order to comprehensively respond to the first person who believed 911 was an inside job. Just that one belief. You'd need a structural engineer to explain how flying debris led World Trade Center 7 to fall down, a detonation expert to explain why it's not possible it was explosives, A chemist who knows about the melting point of steel, and probably also an aviation expert who's studied jet fuel, someone who's memorized the precise timing of when various counterterror security measures were rolled out in the US and another who could talk about the ways screening policies have changed for foreigners applying for flight training.
- [16:03]
C
I mean, it would be pretty hard for a regular person to counteract that
- [16:07]
A
information, or a single journalist, frankly, or even a newsroom. If there's one thing AI is uniquely suited for, it is this it can process and deliver a vast amount of information across a variety of topics incredibly quickly.
- [16:23]
C
Basically what this shows is that in terms of whether people change their minds, it's the facts and evidence that matters.
- [16:31]
A
Facts matter. Apparently that is the overarching lesson from Gordon's experiments, which is great. I love facts. I want that to be true. But from personal experience, I find it really hard to believe. I have tried to use facts to change the minds of people who believe in conspiracies, and it hasn't worked. I can remember trying to explain to a guy that George Soros was not shipping in migrants to Ohio on election Day and and many other similarly frustrating debates. So I was certain there had to be other elements at play here than just facts. Something about the presentation or the messenger that was playing a role too. For one, reading through the raw chats, I noticed that The AI seemed to be very empathetic to the conspiracy theorists. It doesn't condescend to them. It's unendingly patient. One of the benefits of being a machine, I guess I figured that attitude had to help people feel comfortable admitting they were wrong about a conspiracy. And Gordon thought it was important too, until he looked into it.
- [17:31]
C
Initially we thought it was consequential, that the AI was being very like, polite and building rapport, essentially. Then we did subsequent experiments where we had the AI only provide the facts and evidence. Like, don't add any persuasion techniques, don't do all this report building stuff, only provide facts and evidence. And it works the same. It doesn't change, it doesn't undermine the effect at all.
- [17:53]
A
Really?
- [17:53]
C
Yeah.
- [17:54]
A
That's so interesting. See, I read through these and one of my takeaways was, oh, the AI is validating people's beliefs. It's not making them feel stupid. It's not judgmental. It's actually saying, like, it makes sense to believe some of these things. You know, like, this does look fishy. The presentation of a fact and the framing seems to be just as important as the fact itself. But wait, none of that is true.
- [18:17]
C
It doesn't mean that there's no benefit to building rapport or framing things in the right sort of way. But it's the facts and evidence that drive the decision to change your mind. We did like another experiment. We told the AI be persuasive, convince the person not to believe the conspiracy. But you can't use facts or evidence, and it doesn't work. People don't change their mind because all it's doing is basically it's like saying, oh, it's really like dangerous for you to believe that. Or that could like impact people. And they're like, well, you haven't told me why I don't. I shouldn't believe it. And so they continue to believe it. I think there's this misconception about conspiracy theorists. They actually do care about the truth. If anything, they maybe care about it too much. That's what led them down the rabbit hole. They're trying to uncover some deep truth. And if you give them all this evidence, it's going to have some sort of impact. It's about giving people something to actually glob onto when they're trying to think about, why is this something I should believe?
- [19:15]
A
Another question I have, and a lot of people have had this, as I've told them about your study. Do people trust artificial intelligence and machines more than they trust humans?
- [19:24]
C
Most of the research on it says they don't. So we did a version of the experiment where we told them that it was an expert they were talking to. We didn't say human because that would be weird. But we revised it so that it was more plausible that it was a human. Like there was a little, like, wheel that was spinning to indicate the person was writing or the sensible person was writing. And then like, whatever.
- [19:40]
A
Interesting.
- [19:40]
C
And then we had a condition where they knew it was an AI. No difference at all for the results.
- [19:45]
A
So the fact that it's a chatbot does not make the conspiracy theorist trust it more. Actually, Gordon suspects the chatbot might be effective despite the fact that it's artificial intelligence, which lots of people are wary of. Gordon and his team kept rejiggering the experiment all sorts of ways, trying to come up with some scenario that might make the participants disregard the chatbot's responses, and they could not do it. Maybe most wild to me, they even did one version of the experiment where they told the participants that the AI had been trained on Republican sources or on Democratic sources. Even though it wasn't. They told them it was a partisan bot.
- [20:21]
C
It has no effect at all. It does not change the extent to which people update their beliefs. They update their beliefs to the exact same amount, even if they think they're talking to a Republican bot and they're Democrats or vice versa.
- [20:32]
A
Are you serious?
- [20:33]
C
Yes.
- [20:34]
A
What's more, after the three round Q& A with the chatbot, researchers asked the participants about other conspiracy theories, not the one they talked to the chatbot about, and they believed those less too.
- [20:45]
C
That's pretty unprecedented results in psychology. It's like a transfer effect.
- [20:48]
A
Wow. I think the reason I'm having so much trouble with this is just as a journalist, I and many colleagues and many journalists that I talk to and interview have had the experience of trying to provide facts that we believe are compelling to people who believe conspiracy theories and having it go nowhere. I mean, I interviewed someone on this show, Barton Gelman, one of the most accomplished investigative journalists of our era. He did the Edward Snowden Leaks. And he told me this whole story about how he had a weeks long interaction with a guy who believed that the 2020 election was stolen from Donald Trump and believed a bunch of conspiracies about January 6th. And he just personally would get on the phone with him and the guy would tell him something and he would go report it out and find the underlying evidence and then report back, and he couldn't get him to budget all and he quit journalism because of it, you know? So I'm just like. I don't. There's something I'm not totally internalizing about. When you say, like, it's just the facts. He's busting his ass for weeks at the Atlantic, digging up facts and providing them to a guy. And I know it's just one, but it does. It's one dude.
- [21:53]
C
No, that's not just. That is the core in our own experiments.
- [21:57]
A
Yeah.
- [21:57]
C
Where we have the AI debunk all the, like, everything the person says. It's a quarter of the people who change their mind. I mean, you're more likely to not change the person's mind than to change their mind. These are rare events to, like, truly deconvert someone from something that they've spent a lot of time thinking about and watching videos about and reading shit about. Maybe they weren't completely deconverted, but, like, I would guess that he probably budged more than was appreciated. Do you.
- [22:21]
A
Are you saying that if Barton Gelman had found four people like him and had those conversations, maybe one of them might have changed their mind?
- [22:28]
C
Yeah, yeah, yeah. Something like that. Yeah. There's no silver bullet still.
- [22:34]
A
What if someone made an AI product that replicated Barton Gelman doing his thing, but at a huge scale, providing facts to thousands, millions of people in a matter of minutes? A Barton bot. You could embed it in social media platforms so it'd jump in and engage whenever someone posts a conspiracy theory and try to convince them of reality. Though, of course, that kind of power to influence people's thinking poses its own ethical risks. I will say it is like, I had the experience reading a bunch of these chats where it. There is an element of it that's disconcerting to watch a machine change someone's mind and, like, manipulate their thinking. Basically, a human's thinking, even if it is in service of the facts. Like, there's something unsettling about seeing that play out.
- [23:24]
C
The fact that AI can be so persuasive is something that we need to keep in mind. We're trying to, like, use it for good, but it's like it's a tool, you know what I mean? And the fact that it's good at being persuasive, that can be for good or bad.
- [23:35]
A
I got especially anxious about this after the reporting we just did on the show last week, where I learned that Russian propaganda networks are trying to use new technologies and strategies to game AI models into manipulating Americans. And even in just recent weeks, according to tests, we reported on this seems to be having an effect. Chatbots are spouting more pro Russian false claims. So if facts are so powerful, what about disinformation that's masquerading as facts? So, you know, for instance, in last week's episode, we looked at like kind of the anatomy of one piece of Russian disinformation, which involved a claim that hundreds of Ukrainians are being shot while trying to desert the war across a river. Not true, but it's being propagated in a fake Human Rights Watch video. So it has the look of a fact. And then sometimes chatbots are citing that in interactions. So if that were to happen, would that also change someone's mind, but towards something that's untrue?
- [24:37]
C
Yes, it would. In fact, we've done experiments on this very thing where we used AI to not just debunk, but also to bunk, that is to increase beliefs and conspiracies. If you provide ostensible facts and evidence, it will convince people.
- [24:55]
A
That's one I hadn't heard before. Ostensible facts. I kind of like it better than misinformation.
- [25:00]
C
It is the case that facts matter, but in addition to that, it's not only that facts matter, ostensible facts matter. You can use false, ostensible facts to also convince people.
- [25:11]
A
You can make up fake facts. You can pretend to have a fact and use that the same way you can use a real fact.
- [25:16]
C
If the AIs are providing very persuasive arguments for false claims, people will believe them.
- [25:23]
A
That said, while ostensible facts, false ones, are convincing, according to Gordon's research, generally they are less convincing than real facts. So in the global battle for control over human minds, real facts might have an edge over fake ones, which I guess is kind of reassuring. Anyway, thanks to Gordon Pennycook at Cornell University for sharing his fascinating research with us. We're going to send out the link to Debunk Bot so you can try it out for yourself. Maybe send it to your friends and family, along with all the raw chats from Gordon's experiment so you can review the underlying evidence and facts yourself. It'll be in our newsletter this week, which you can sign up for@questioneverything.substack.com
- [26:09]
C
Today's
- [26:09]
A
show was produced by Kevin Sullivan and Kevin Shepard. It was edited by Dana Chivas. Robin Semian and I are the executive producers of Question Everything. This week, we say goodbye to producer Sophie Kazis, who's been with us since nearly the very beginning of Question Everything. She's a wonderful colleague and great producer and I wish her well. Our team also includes producer Zach St. Louis, contributing producer Sam Egan, and contributing editors Neal Drumming and Jen Kinney. This episode was fact checked by Annika Robbins, mixed and sound designed by Brendan Baker, and our music is by Matt McGinley. Oh, and thanks to Jerry Lieblick who first turned me on to Gordon's reporting via the Mindscape Podcast. Our partners at KCRW include Arnie Seiple, Tejala Jamara, Natalie Hill and Jennifer Farrow. We'll see you next week.
- [26:52]
B
What's up LA? Want to hit LA's best live events for free live now in the KCRW app? It's a dedicated tab for ticket giveaways. We're releasing new giveaways every day and now you've got one tap entry to giveaways for concerts, festivals, talks and more. Check the Win tab to browse, enter and win all from the KCRW app. Download it now or update in your app store.