Is AI how we finally beat conspiracy theories? The answer might be yes, according to a recent study out of MIT & Cornell. What they found & my hunch on what’s behind the very encouraging results: 1500+ people who believed in a conspiracy theory of some kind sat down for a 5-minute conversation with ChatGPT about said theory…with a twist. 1/3rd knew they were talking to AI, but 1/3rd thought it was an expert, and 1/3rd thought it was another study participant who disagreed with them. Those 5-minute conversations worked. Out of the people who knew they were talking to AI, just under a third “switched from the false or unsupported belief to its more accurate counterpart.” Even more people—a full third—changed their minds when they thought they were talking to a human expert instead. Obviously, that leaves a majority of people who didn’t fully change their position. But those other participants reported their false beliefs had been weakened, even if not fully undone. And this was true even for people whose beliefs were very strongly held! What I find most impressive of all is that researchers checked in after two months and this persuasion stuck. If those people had gotten sucked right back in, this would be a much less hopeful paper. Instead, AI really helped them move on. The big question: why did these 5 minute chats work? The paper theorizes it’s thanks to AI’s ability to personalize its arguments. I could buy that, but my hunch is it’s something a little more emotional. With AI, there’s no sense of judgement. Unlike humans, AI can be infinitely patient, explain things over and over in different ways, and respond to questions without getting frustrated or heated. Zero eyerolls. In a perfect world, it’d be 100% (or at least close) of people who gave up their conspiracy theories. And these numbers could still go up: what if they talked for more than 5 minutes, for example? In the meantime, this is a huge, hopeful first step. Longer write-up here: https://lnkd.in/eST7FaEh
The data should be based on the amount of conspiracy theories that turned out true. Speaking with an AI that only knows the news isn't going to help anyone as even Copilot has told me things like Andrew Tate is a convicted human trafficking. Which was wrong, it was so biased and hating Andrew Tate I couldn't believe it. But all it has was the media for information. The same with conspiracy, what would it really know about Jeffrey Epstine or Puff Daddy when people are speaking to it before all he news breaks about them. The whole time the people are crazy conspiracy theorist, untill their not. 🙄 I don't understand how this will work.
*Could the same traits that make AI effective at dismantling conspiracy theories also make it a powerful tool for spreading them?* Fascinating study and very encouraging results! It’s impressive to see how AI’s ability to engage patiently and non-judgmentally can create meaningful change. But your post raises an important counter-question. If AI can personalize arguments and engage without frustration, it might also excel at reinforcing false beliefs in the hands of someone with malicious intent. For example, could AI systems be leveraged to tailor content that deepens a conspiracy theory’s appeal or even creates new ones? This dual potential makes me wonder how we can ensure AI is developed and used responsibly—not just for persuasion, but for safeguarding truth. What measures can we put in place to prevent the misuse of this incredibly effective technology?
By this stage, we know AI is still not intelligent and follows the data it is fed. Therefore, it can ( and based on where that data comes ) it would be biased ( I hope I'll be proven wrong and AI shows zero bias and prejudice which means high integrity of its human engineers ). Therefore, right now it cannot disprove theories that haven't been proven false ( because it cannot find the data provided by humans ). However, this paper tells me it can be used as a new medium to direct people's perspectives on the world by "those" who want to "direct" people's perspectives on the world. Newspapers, magazines,TV news, movies, social media have already been used for that goal and now a new tool would be added to the toolkit. ( The use of media for that has already been proven true. )
So basically what you are suggesting is using AI as a tool to homogenise people's beliefs and how the perceive the world and squash people's free thought in the process? There are 'conspiracy theories' that over time have proven to be absolute fact. There's also conspiracies such as collusion of big tech and government happening in real time. I'm sure AI will certainly silence those ones. I mean come on, what is this BS 'science'? More like a self reinforcing echo chamber typical of linkedin feeds
Ha - my goodness, this post did not go over well, Mustafa Suleyman. You have the non sequitur of a premise that every conspiracy theory is false, which is absurd. Many have a grain of truth. And there's also a fine line between systemic injustice and "conspiracies". Just yesterday I wrote this post about a prostitute in Nairobi who was basically enslaved for 20 years by AIDS researchers who raised $22 million for their study, but forced her to work as a prostitute for decades because they needed her to be continuously be exposed to the virus. How is this not a conspiracy? https://kinyarwanda.me/2024/12/19/data-colonization-in-medical-research/
Mustafa Suleyman There is an AI hype for money 💰🤑💸. Yes, AI can do many things, but NOT everything. Why don't we use AI? Because it can't solve our problem. Do you or any of your contacts need our expertise and intellectual property (IP), a copyrighted Chinese-English multilingual metadata, to do the data analysis that AI can't do? A five-year-old kid knows the fact that we need menu to order our food at a restaurant. Similarly, we need metadata to search/retrieve the right data for data analytics. Metadata is a copyrighted content, NOT technology. It is fundamental to enable data analytics. Without metadata, NO data can be found/retrieved, even by the most advanced technologies, like AI, quantum computer. https://lnkd.in/g-aJFnXR Let's try the go/no-go test here to see if AI works without metadata, and then, at what cost. Is there any data solution, AI or NOT, that can answer the following questions of business intelligence? "How many entities, in the Ontario province of Canada or in the "江蘇" province of China, have new US patents granted on the nearest Tuesday (Eastern Time), when the USPTO releases the newly granted US patents on a weekly basis?" With our IP, we can answer, even by an ordinary laptop.
It looked like it could be possible provided that the person continuously in touch with the AI assistant so that it can persuade. Persuading for a few days is easy but persuading to change a belief is difficult. This research shows results after 2 months which is a reasonable amount of time to believe that an AI assistant might work well here. But here comes the problems too. It might persuade people to do certain things, which should not be done.
It's encouraging to see that even brief interactions can have a lasting impact on beliefs, especially without the emotional baggage that often comes with human discussions on contentious topics.
"With AI, there’s no sense of judgement. Unlike humans, AI can be infinitely patient, explain things over and over in different ways, and respond to questions without getting frustrated or heated. Zero eyerolls."
Author/German Instructor/Translator ⚖️Law 🌍Business/Marketing 📚Books. Helping you communicate effectively in German and English. Ideal positioniert für englische und deutsche Übersetzungen. Seit 1998. +1 419 320 7745
1dSo you're talking about conspiracy theories that have been previously debunked. Those are the only theories AI could opine on and be correct because it uses factual data collected from humans who created them. Any unproven or non-debunked theories would be hard for AI to convince people on because AI does not know if they're true or not. It's again all about just sequencing words based on data AI was trained on. No thinking, no sentience, no reasoning.