What Happened When Conspiracy Theorists Talked to OpenAI's GPT-4 Turbo? (washingtonpost.com) 108
A "decision science partner" at a seed-stage venture fund (who is also a cognitive-behavioral decision science author and professional poker player) explored what happens when GPT-4 Turbo converses with conspiracy theorists:
Researchers have struggled for decades to develop techniques to weaken the grip of conspiracy theories and cult ideology on adherents. This is why a new paper in the journal Science by Thomas Costello of MIT's Sloan School of Management, Gordon Pennycook of Cornell University and David Rand, also of Sloan, is so exciting... In a pair of studies involving more than 2,000 participants, the researchers found a 20 percent reduction in belief in conspiracy theories after participants interacted with a powerful, flexible, personalized GPT-4 Turbo conversation partner. The researchers trained the AI to try to persuade the participants to reduce their belief in conspiracies by refuting the specific evidence the participants provided to support their favored conspiracy theory.
The reduction in belief held across a range of topics... Even more encouraging, participants demonstrated increased intentions to ignore or unfollow social media accounts promoting the conspiracies, and significantly increased willingness to ignore or argue against other believers in the conspiracy. And the results appear to be durable, holding up in evaluations 10 days and two months later... Why was AI able to persuade people to change their minds? The authors posit that it "simply takes the right evidence," tailored to the individual, to effect belief change, noting: "From a theoretical perspective, this paints a surprisingly optimistic picture of human reasoning: Conspiratorial rabbit holes may indeed have an exit. Psychological needs and motivations do not inherently blind conspiracists to evidence...."
It is hard to walk away from who you are, whether you are a QAnon believer, a flat-Earther, a truther of any kind or just a stock analyst who has taken a position that makes you stand out from the crowd. And that's why the AI approach might work so well. The participants were not interacting with a human, which, I suspect, didn't trigger identity in the same way, allowing the participants to be more open-minded. Identity is such a huge part of these conspiracy theories in terms of distinctiveness, putting distance between you and other people. When you're interacting with AI, you're not arguing with a human being whom you might be standing in opposition to, which could cause you to be less open-minded.
Answering questions from Slashdot readers in 2005, Wil Wheaton described playing poker against the cognitive-behavioral decision science author who wrote this article...
The reduction in belief held across a range of topics... Even more encouraging, participants demonstrated increased intentions to ignore or unfollow social media accounts promoting the conspiracies, and significantly increased willingness to ignore or argue against other believers in the conspiracy. And the results appear to be durable, holding up in evaluations 10 days and two months later... Why was AI able to persuade people to change their minds? The authors posit that it "simply takes the right evidence," tailored to the individual, to effect belief change, noting: "From a theoretical perspective, this paints a surprisingly optimistic picture of human reasoning: Conspiratorial rabbit holes may indeed have an exit. Psychological needs and motivations do not inherently blind conspiracists to evidence...."
It is hard to walk away from who you are, whether you are a QAnon believer, a flat-Earther, a truther of any kind or just a stock analyst who has taken a position that makes you stand out from the crowd. And that's why the AI approach might work so well. The participants were not interacting with a human, which, I suspect, didn't trigger identity in the same way, allowing the participants to be more open-minded. Identity is such a huge part of these conspiracy theories in terms of distinctiveness, putting distance between you and other people. When you're interacting with AI, you're not arguing with a human being whom you might be standing in opposition to, which could cause you to be less open-minded.
Answering questions from Slashdot readers in 2005, Wil Wheaton described playing poker against the cognitive-behavioral decision science author who wrote this article...
Does it admit to the real conspiracies? (Score:4, Insightful)
Things like Iran-Contra, MKUltra, Tuckaseegee, United Foods, WMD in Iraq, etc. etc. etc.
It's hard to dissuade (or persuade, for that matter) when you lack credibility.
Re: (Score:2)
Re: Does it admit to the real conspiracies? (Score:1, Flamebait)
Re: (Score:3)
It’s clear that a lot of people have some mis-firing neural logic circuits that make them susceptible to conspiracy theories. But, it doesn’t help that some of them have a real grain of truth.
Example: chemtrails. Absolutely bonkers, except that the US military totally chemtrailed San Francisco in the 1950s in order to experiment on the population.
https://en.wikipedia.org/wiki/... [wikipedia.org]
Re:Does it admit to the real conspiracies? (Score:5, Insightful)
But isn't that just the thing with conspiracies, how easy it is to link two only tangentially related things and draw really outsized conclusions based on that link even if the scale of evidence is nowhere approaching that?
In the chemtrail example, sure we have a thing that happened but is there anything about Operation Sea Spray that gives evidence to the idea that there is something nefarious in the vapor trails of commercial airliners or even military aircraft, especially when we already know for certain fact that vapour trails are a natural phenomenon of turbo and jet fan engines? It's kinda perfect conspiracy bait since the trails are always there.
"The origin of this conspiracy has some validity to it therefore the conspiracy that sprung from it must also have validity even if there is no evidence" is something of a fallacy. All of those examples follow this logic and it doesn't make the conspiracies around them anything more than unevidenced claims.
It doesn't also help that the primary media proponents don't actually believe in them either, the "true believer" style "Lone Gunmen" are kinda dead and replaced with malicious actors who use these conspiracies for ulterior motives. Do you think Alex Jones actually believes chemtrails exist? He doesn't care it's means to end, a brand exercise.
Re:Does it admit to the real conspiracies? (Score:5, Informative)
You see you are doing the thing I said right?
but this is weather modification we are talking about.. and it's well established for decades at this point.
Is it? So because cloud seeding is a thing that has in fact been experimented with it should be considered a valid belief to say that weather modification is a capability that the government or some non-state actor has access to and is currently using? I disagree, lets leave it at that. I would say show me some evidence of weather modification done at any non-experimental scale and hidden from the public. Especially at the level conspiracy theorists imply it is happening.
once-controversial conspiracy theories
Were they once controversial at the time? Tuskegee was whistleblown by someone on the team doing it in like 1955, were there conspiracies at that time?
Iran Contra was known because there was a congressional investigation not long after it happened.
MKUltra is the cloest but even then, were the conspiracies taking place in the time it was under wap or did they all spwn from the CIA disclosure of the program?
WMD in Iraq was a conspiracy in what sense when it was out in the open and being called out from 2003, it was discussed on the 2004 campaign. The facts of something not being publically known does not comprise a conspiracy, it's just that window of uncertainty allows more to take root.
That said Iraq and the Office of Special Plans is the closest one amongst all of those to actually living up to the legend; Rumsfeld and Wolfowitz and crew did a conspiracy but in this case we have reams of evidence so no conspiracy theorists actually care about it becasue 1. it's against their party and 2. there is evidence so wahts the fun in that? Can't rot brains when there was mainstream reporting on it in 2004 The Lie Factory [motherjones.com]
Re: (Score:2, Troll)
" it should be considered a valid belief to say that weather modification is a capability that the government or some non-state actor has access to and is currently using"
Was Saudi Arabia wrong in thinking so?
https://science.slashdot.org/s... [slashdot.org]
Was Russia wrong in thinking so for the 2013 olympicss? Here's an article from 2010 about the intention, do you see any incredulity that they could pull it off?
https://www.foxsports.com/stor... [foxsports.com]
Is China insane for doing it?
https://science.slashdot.org/s... [slashdot.org]
https://science [slashdot.org]
Re: (Score:2)
Nothing you posted contradicts what i said about weather manipulation, you just did the thing again!
Cloud seeding isn't a conspiracy! It's a real thing and we aren't even sure how well it works.
Nothing in those links draws the line between "cloud seeding is a thing that is possible and maybe works sometimes?" into the conspiracy that weather events are and are currently being manipulated in a malicious fashion. It doesn't anywhere and don't pretend like we don't know what the weather control conspiracy is,
Re: (Score:2)
In the chemtrail example, sure we have a thing that happened but is there anything about Operation Sea Spray that gives evidence to the idea that there is something nefarious in the vapor trails of commercial airliners or even military aircraft, especially when we already know for certain fact that vapour trails are a natural phenomenon of turbo and jet fan engines?
I think we can be reasonably confident that there isn't since it's much cheaper to for them to use fluoridated drinking water.
Kidding aside I view conspiracy theories the same way I do lotto tickets. A few of them pay off, but the vast majority are worthless. I suspect they're largely just a modern spin on whatever circuitry in the human brain long ago produced god. People just need something to be in control of all the chaos, even if they believe it to be a malevolent force.
Re: (Score:2)
Kidding aside I view conspiracy theories the same way I do lotto tickets. A few of them pay off, but the vast majority are worthless. I suspect they're largely just a modern spin on whatever circuitry in the human brain long ago produced god.
I think people just get caught up in their own shit. Look at the people who used to promote 9/11 conspiracy theories and now don't, and that one was hugely popular until about 2008.
Today's equivalent is the Trump campaign/Russia collusion crap. There never was any strong evidence for it. Period. Yet I guarantee you that everybody who seriously bought into it still thinks its true, and any evidence they bring up is just as shoddy as the supposed evidence behind the 9/11 conspiracy theories.
Re: (Score:2)
Re: (Score:2)
Re: (Score:2)
RFK jr.
Re: (Score:1)
Re: (Score:1)
There were a lot of people like yourself, that made up arguments to convince people to take an unproven, experimental drug. Putting words in other people's mouths. Labeling and grouping people with reservations as "anti-vax." Pushin
Re: (Score:3)
Not when the CIA does it (Score:3)
No meaningful punishment for this behaviour
https://thehill.com/policy/tec... [thehill.com]
Re: (Score:1)
The climate has been changing back and forth for as long as there's been a Planet Earth to have a climate on. The only thing that makes this last one different, is that it happens while working-class people need cars to get to work...
Re:Does it admit to the real conspiracies? (Score:5, Insightful)
Conspiracy is clearly defined as a secret cooperation between people to the detriment of others. This happens all the time, and in many situations, it happens for legitimate reasons. An undercover investigation is a legal instrument for the prosecution, and it is to the detriment of the perpetrators of a crime.
A conspiracy theory instead claims something a) would be to the detriment of a diffuse us, b) was intentional and c) constructs an unproven collusion between likely diffuse groups and d) often fails to construct a plausible gain for the postulated members of the collusion and replaces it with the members simply being evil. It dismisses alternative explanations outright and re-adapts itself to attribute the alternative explanations as part of the conspiracy.
No flat-earther can give a sufficient reason for the members of the round-earth conspiracy to insist on the spheroid shape of Earth, and they also fail to provide any evidence for the belief in a round earth to be bad for us. But they insist, that the "Round Earthers lie to us, because they are evil."
Re: (Score:3)
Re:Does it admit to the real conspiracies? (Score:5, Informative)
You are touching on a critical distinction between real conspiracies, and conspiracy theories. Real conspiracies involve details: specific places, actions, and people, who engage in subversive behavior. the list you provided, is a good example. We have details for each of these conspiracies.
Conspiracy *theories*, on the other hand, are vague and suggestive, but never include details. "They" don't want you to know, "shadowy, powerful people" are really controlling things.
I don't think these AIs will have trouble distinguishing between the two.
Re: (Score:2)
Facts convinced them? (Score:3)
Re: (Score:2)
Maybe thinking the facts came from a machine instead of a person made a difference somehow? That could be something to test in further experiments. Maybe we need to argue with them in an AWESOM-O costume.
Re: Facts convinced them? (Score:2)
Re: (Score:2)
"He's from Japan!"
Re: (Score:1)
There's also how a machine is less likely to have a Nefarious Agenda (obviously depending on who's been programming it).
More importantly, it's less likely to be preachy, condescending and arrogantly judgmental about it. Less likely to pour scorn on the lowly Untermensch for stepping out of line with their thinking.
Re: (Score:2)
I think conspiracy theorists are invested in their own ability to discern "the truth" where normal people only see "the narrative"--so they aren't invested in any particular facts apart from their conclusion (RAND Corporation with flying saucers and reverse vampires FTW). The researchers think that the facts can sell themselves better when you take the emotions of pride and insecurity out of the equation.If your position on any major issue was acquired by a chiropractor's video on youtube, you are terrible
Re: (Score:2)
People who adhere to conspiracy theories often have immersed themselves in information ecosystems that reinforce the conspiracy. It's the lack of belief-challenging facts that helps them stay hooked. Emotion is certainly part of it, but that's a tool wielded by the managers of the information ecosystem to keep them from straying.
The emotional problem will remain no matter what you do. You need to get people to recognize they believe in something that is false and that their emotions have been exploited in o
Re: (Score:3)
Facts alone, probably not. But what does often convince conspiracy theorists, is showing them self-contradictory aspects of their theories. Sometimes, it's not easy to put a finger on these contradictions, but AI would probably be good at that.
Re: Facts convinced them? (Score:3)
Couple people I know are like that, happy to drop con-theory 1 when juicier con-theory 2 comes along.
Seems to be a personality trait to want to rush into a (usually) wild contrary position from the status quo.
Re: (Score:2)
You have to fix the emotional problem, not the factual information problem.
Sounds familiar [wikipedia.org].
I asked ChatGPT to reply to the headline (Score:2)
Re: I asked ChatGPT to reply to the headline (Score:3)
Re: I asked ChatGPT to reply to the headline (Score:2)
Re: (Score:2)
How fickle these people are. I am forgotten [wikipedia.org].
Re: (Score:3)
No danger here at all! (Score:5, Insightful)
The researchers trained the AI to try to persuade the participants to reduce their belief in conspiracies by refuting the specific evidence the participants provided to support their favored conspiracy theory.
Because there's absolutely no potential here for creating a 'propaganda AI' trained to persuade people that the truth is in fact a lie. No, that could and would never happen. /sarc
Re: (Score:2)
You beat me to it. It's not the message being delivered but the way it's being delivered that has weakened resistance.
People aren't changing their minds. They're letting a machine do the thinking for them.
Re: No danger here at all! (Score:3)
I know an orange man and his oli-buddies working on that
Hey, lend me, ummm, $25 billion will ya? I heard that DickDok is for sale and it comes with 175 million American subscribers.
Re: orangie plutocrats working on fibbot (Score:1)
They willingly do it manually, don't need bots. Quality doesn't have to be high if the listener is gullible, paranoid, and clueless.
Does it also work the other way around? (Score:5, Insightful)
If you train your AI to persuade people to belive in conspiracies - does it work?
And does it work better than human propaganda channels?
In either way, the owners / trainers / operators of our new AI overlords yield horrible power.
Re: (Score:2)
To do that, you'd need to insulate the AI in an information ecosystem that reinforces the conspiracy and shuns contradicting information.
But when this AI is released to the world, it will be exposed to alternate sources of information. My guess is that it may very well think itself out of the conspiracies you attempted to train into it. Just like a human could.
That's the foundation of GrokAI (Score:1)
Exactly. Musk has prioritized Twitter content in training GrokAI in order to perpetuate the misinformation flourishing on that platform.
Re: That's the foundation of GrokAI (Score:2)
Re: (Score:2, Insightful)
Exactly. Musk has prioritized Twitter content in training GrokAI in order to perpetuate the misinformation flourishing on that platform.
Nope. Polemic much? Haven’t used it, eh? Jumped straight to Threads and Bluesky when your playground went free speech, eh?
1 - A dynamic tie to X makes Grok enormously up to date. It can comment on hour old events, help verify truthfulness, and provide reams of context.
2 - Unlike much of old style cable, newspaper, magazine, and social media, X itself splits evenly between left and right both in terms of its user base and its posts. That REALLY bothers radicals from any side.
3 - Grok itself leans left
Re: (Score:2)
Grok and X are not the beacons of balanced truth you believe them to be. I'll just leave these here:
https://www.businessinsider.co... [businessinsider.com]
https://casmi.northwestern.edu... [northwestern.edu]
https://globalwitness.org/en/c... [globalwitness.org]
Re: (Score:1)
Business Insider is your top neutral source? Really? Business Insider?
Sigh.
Re: (Score:2)
Re: (Score:2)
So you would expect that the self proclaimed protectors of the truth and fact checkers for the nation would be more than happy to get their points of view out to where these conspiracy generating AIs would be exposed to them. Thank you. This is good to know [nytimes.com].
Re: (Score:2)
Don't put words in my mouth.
I'm saying you can't sustain a conspiracy if the individuals who believe it are exposed to truth that has been hidden from them.
You linked to an article (paywalled) about a dispute regarding the use of copyright material to train AI, and suggest NYT should embrace such use because it gets "their points of view out."
There's nothing inconsistent in NYT's position. It wants its subscribers to learn from what it publishes, but you can only use this information in certain "fair use" w
Re: (Score:2)
That's the default path for public consuming LLMs given the huge amount disinformation floating around. Which will be why these bots had to be "personalized". Otherwise they'd just be agreeing with the fantasies.
Re: (Score:2)
What I have found is that these LLMs (particularly ChatGPT) are somewhat conciliatory, ready to presume any question asked of them is a legitimate one, and going out of their way to tailor responses accordingly. To put it another way, they're good at bullshitting when you ask them a bullshit question.
Re: (Score:2)
Re: (Score:2)
That's a credible conspiracy (Score:2)
Re: (Score:2)
You know, I was just thinking that the silly people who came up with this idea don't really understand how conspiracy theories work. Even if ChatGPT could do this, it would likely only be a matter of a few days before a new round of conspiracy theories went viral - ones that warned about all the nefarious things these "AI"s were really designed to do.
But what happened when (Score:2)
Re: But what happened when (Score:2)
Re: But what happened when (Score:2)
Opinion polls != science (Score:1)
Self-reported behavior isn't a good measurement of anything other than what a subject would say to an interviewer. Self-reported *future* behavior is even less tethered to reality. Happytalk to the contrary notwithstanding.
Tiananmen Square Never Happened (Score:2)
Says the Chinese AI bot. That's just a conspiracy.
Put all the information out there and let people choose. If folks want to look like idiots, let them. It's only a conspiracy until it's proven to not be. In the case of flat earth folks, maybe the ai can give them an experiment so they can learn for themselves. Science is repeatable. You don't need to dissuade folks, just prove it. Simple. And if you can't prove it, well, then who knows who's right? All I know is that it shouldn't be whatever the pro
Re: (Score:2)
Re: (Score:2)
They were just conspiracies, i.e.any covert plan involving two or more people.
For example, Operation Paperclip was a conspiracy, but in 1957 no one was theorizing that it existed. The conspiracy itself was revealed later. So it was a conspiracy, not a conspiracy theory.
Re: (Score:1)
Re: (Score:2, Informative)
We know what’s on your mind.
Re: (Score:1)
it's one of the most obvious delusions that is trivial to point to, and brings the activists out of the woodwork with downvotes pretty quickly.
most people have zero interest in validating the delusions of militant activists.
rape and death threats are par for the course when it comes to the love and kindness brigade.
the females are ovarit and ovarit.com has all the statistical references you'll need to get over this dangerous fad.
Re: TRANSWOMEN ARE MALES WHO THINK THEY'RE WOMEN (Score:3)
That insanity is generally considered a negative personality feature?
Who cares. They're cute. (Score:1)
And that's a fact.
The problem is... (Score:2)
The problem is: how do you know? I mean, sure, Bigfoot does not exist, but some "conspiracy theories" turn out to be true.
Training AIs to push certain points of view is, in the end, it's own kind of conspiracy. The viewpoints pushed will reflect the biases of the trainers.
Who to trust is a BIG problem (Score:2)
The internet has made lots of weird ideas that they used to be protected from available to the public. The public, on the whole, has little idea of what sources are 'trustworthy', so absorb the claims first made to them.
Of course this is nothing very new; people have been denying the validity of the claims of Christianity because its truth is inconvenient to them, on spurious grounds of refusing to believe the sources, for a very long time...
Banning vs Discussing (Score:2)
Personally, I have not been able to convince LLM in something it has a pre-set idea about, but it has been a great test for my understanding of the topic
Or ... (Score:3, Insightful)
It is hard to walk away from who you are, whether you are a QAnon believer, a flat-Earther, a truther of any kind or just a stock analyst who has taken a position that makes you stand out from the crowd.
or a belief that over half of all voters are "racist", or a belief that chasing fewer goods with more money won't cause inflation, or a belief that putting a dress on a boy magically creates a girl ...
So Ai (Score:3)
Tinfoil Hats, Meet GPT-4. (Score:3)
Fascinating paper. Non-paywalled preprint here. [osf.io] The authors claim that AI—specifically GPT-4—can persuade conspiracy theorists to dial down their beliefs by about 20%. Even better, that shift sticks around for at least a couple of months. Participants in the study described the specific conspiracy they believed, GPT-4 challenged it with tailored evidence, and boom—significantly lower belief afterward. This effect spilled over to other conspiracies as well, suggesting that personalized debunking really can change minds.
Interestingly, they also noted that the same superpowered persuasion that deflates conspiracies could just as easily prop them up. AI’s a double-edged sword—researchers used it here to provide well-referenced counterarguments, yet there’s nothing stopping bad actors from flipping the script and turbo-charging conspiratorial nonsense at scale.
For everyone bemoaning the “post-truth” era, I think this offers some hope. The paper suggests that carefully targeted evidence still matters—even to those “lost causes” out there, insulated from reality inside epistemically closed cocoons. I think this is true because it reaches people personally. That’s where GPT-4 (or any similarly powerful model) can shine—fact-checking on demand, answering in real time, and tackling everyone’s unique version of the same story.
Re: Tinfoil Hats, Meet GPT-4. (Score:2)
And yet UFO people still look to it for "proof" (Score:2)
Go into one of the UFO subreddits and those people are taking its regurgitated garbage as proof.
Define a conspiracy theory (Score:1)
Why you can't stop conspiracy theories/theorists (Score:2)
The conspiracy is more fun than reality.
That's ALL you need.
Human Nature (Score:2)
The CIA organized the overthrow of the elected Ukrainian government in 2014.
The claim that the CIA organized the overthrow of the elected Ukrainian government is an invention of Russian propagandists.
Which one is the conspiracy theory and which is the actual conspiracy?
I think the reality is that conspiracy theories work because they fit and support people's world view. They often provide people with rational human controlled explanations for random out of control events.
The theories that survive allow almost any factual information to be incorporated into the conspiracy.
USA is a cult, MAGA is a cult (Score:1, Troll)
Psychological needs and motivations ...
People need a story that explains why they are not in control: Why following the rules does not work. An invisible sky-daddy is not a valid excuse in a world full of machines made by men.
Then, there's skirting the edge: US culture approves the belief in a sky-daddy, particularly with a book where one can pick and choose the bits that are 'important' (AKA evangelical Christians). US culture secretly promotes the idea that helping others, via basic living welfare (food-stamps, SNAP), social-security, DE
Re: (Score:3)
People need a story that explains why they are not in control
I think they need a story that explains that someone IS in control. The idea that COVID can come out of the blue and kill millions of people is incredibly frightening. There is no way to prevent it. Better that it came out of lab created by humans. Autism doesn't just happen, it is caused by vaccines. You can look at almost every conspiracy theory and see how it creates a story of human control.
It's tricky (Score:2)
Most conspiracy theories don't have a merit but there are some who turned out to be at least partially true. I'm worried about the word "conspiracy theory" being overused by AI models to cover up willful wrongdoing. A healthy dose of doubt and theorising is good for the society because it stimulates rational thinking and keeps some of the villains in check.
So are birds real or not? (Score:2)
LLMs can also fuel conspiracies (Score:1)
Today’s LLMs (GPT, Grok, etc) default to an “agreeableness” which can actually fuel conspiracy theories instead of stop them out, as does the fact that their training set is most of humanity’s knowledge - a knowledge base which leans FAR more heavily into utopian or wandering narratives instead of tying theories to their actual results. Think of this as the kindly liberal arts faculty usually outnumbering the factual hard science crew.
For example, LLM’s treat Critical Theory
Re:The existence of the database endangers us (Score:1)
They are limiting their subjects to people who are prepared to spend time talking to a LLM.
So relatively openminded for conspiracy theorists.
I can't see the hard core nutters enrolling in such a study.
I just couldn't get past... (Score:1)
Real And Present Danger (Score:1)
Just remember, folks: Many of these Conspiracy Theorists of yours, believe they're under some sort of attack. Usually for veering too close to the Truth.
If your response is to attack them, you just prove their point.
Not sure of the relevence of this study. (Score:2)
Ok so were now cheering AI with a built in bias they tested on people?
All this proves is that giving peopled biased information works.
In this case the bias is negative bias toward conspiracy theory.
But it could just as easily be done to refute the truth about climate change
or the troubles in the middle east.