Researchers from the Massachusetts Institute of Technology and the University of California, Irvine have found that artificial intelligence can imperceptibly alter human memories, even if people know they are seeing generated content.
Researchers from the Massachusetts Institute of Technology and the University of California, Irvine have found that artificial intelligence can imperceptibly alter human memories, even if people know they are seeing generated content.
According to Bloomberg, a team of psychologist Elizabeth Loftus and MIT scientist Pat Pataranutatorn conducted a series of experiments that prove that AI is capable of creating false memories in the mind no less effectively than people previously did during interrogations or surveys.
Elizabeth Loftus proved back in the 1970s that memory is not a «record of events» but a flexible construct that can be altered through suggestion. Modern AI tools can scale this influence, making it faster and almost imperceptible, posing new challenges for media literacy and cybersecurity.
In the first study, participants were shown a video of a robbery and then asked some of them leading questions, including about a «car thief» who didn’t actually exist. A week later, a third of these people remembered the fictional fact, and the group that received questions from a chatbot formed 70% more false memories than those who saw the text questions.
The second experiment demonstrated that manipulative summaries from or dialogues with an AI can change the plot of a read story in a person’s memory and reduce confidence in the true details.
In the third part of the experiment, 200 participants were shown images ranging from real to fully generated and converted into videos. Even a warning about the use of AI did not stop the effect: the group that watched the artificially generated videos had the highest level of distorted memories.
According to the authors, the danger lies not only in fake images or videos, but in AI’s ability to «rewrite» memory, amplifying the effect of mass manipulation on social networks.
We’ve previously written about how cybercriminals are actively implementing artificial intelligence to scale attacks and generate revenue. Malware generation, automated phishing, and self-learning AI worms are just some of the tools used in modern cybercrime.
