We’ve all heard about the evils of artificial intelligence. Indeed, here at the Misinformation Desk, we have written recently about inaccuracies produced in response to queries by ChatGPT and several years ago (before the true heyday of AI) with regard to deepfakes. Obviously, AI is here to stay, so what if we could use AI to diminish misinformation rather than amplify it? Psychological science has some bad news for us, but then some good news (and some better news).
Conspiracy Theories Abound
Let’s start with the bad news: Conspiracy theories are shockingly widespread and a majority of people believe in at least one. For example, a study of about 2,000 people in Australia and Aotearoa New Zealand asked participants whether they knew about various conspiracy theories and whether they believed them (Marques et al., 2021). Participants read a varied list of international and local conspiracy theories, including “Scientists are creating panic about climate change because it is in their interests to do so” and “The Christchurch [New Zealand] mosque shootings were orchestrated with the intent of restricting gun laws in New Zealand.” Almost everyone in the study reported that they knew about at least one of the listed conspiracy theories. Even worse: A majority believed at least one.
Unfortunately, none of us is immune to this type of thinking. As noted by these authors, there have been similar findings from other countries. Even worse, research suggests that although some people naturally recover from their conspiracy-theory beliefs, it’s rare (Williams et al., 2024). After following about 500 research participants over six months, researchers reported that “beliefs in conspiracy theories were highly stable,” a general susceptibility that is worrisome for the future.
AI to the Rescue
Can conversations with AI diminish beliefs in conspiracy theories such as alien invasions? Source: danielarealpeg/pixabay The good news comes from a research article published in the prestigious journal Science (Costello et al., 2024). Participants were randomly assigned to an experimental group in which AI would try to break down their belief in a conspiracy theory or a control group in which they would chat with AI about a non-conspiracy-related topic. The researchers asked each participant to describe a conspiracy theory that they believed, including any evidence that participants thought supported their chosen conspiracy theory. Participants’ proffered conspiracy theories included “classic” tropes (e.g., related to the assassination of John F. Kennedy or the existence of aliens) and more current ones (e.g., related to COVID-19 or the 2020 U.S. presidential election).
Next, the researchers asked participants in the experimental group to engage with AI in “real-time, personalized interactions.” The AI platform used a large language model (LLM) “that has access to vast amounts of information and the ability to generate bespoke arguments.” Over three rounds, the AI countered participants’ supposed evidence for their beliefs. The researchers reported that the AI intervention led to an average decrease in participants’ level of belief in their selected conspiracy theory of 20%, across all types of conspiracies). Moreover, the decrease lasted for the two months of the study (and, hopefully, longer).
Even better news: The researchers asked a fact-checker to verify a sample of 128 of the statements that AI made to counter the conspiracy theories. Over 99% of AI-made statements were true; none were false; only a tiny fraction were misleading. This means that the AI was doing a fantastic job at staying true to real-world data! Even better than that: Researchers documented a decrease in conspiratorial beliefs generally among participants in the experimental condition (as compared to the control condition), not just with respect to the conspiracy theory debunked during the study. Participants in the experimental group also reported that they were now more likely than those in the control group to push back against others who shared conspiratorial beliefs.
An editor’s summary of this article encapsulates this good news: “This intervention illustrates how deploying AI may mitigate conflicts and serve society.” Here at the Misinformation Desk, we have long argued that media literacy must be integrated into education at every level. This research supports adding another tool to our kit. It also showcases how AI may be a boon to truth and not just a threat.