Artificial Intelligence Could Rewrite Human Memories, Making People Believe Events That Never Happened 

Artificial intelligence may be doing more than transforming how people work, communicate and access information. New scientific research suggests that AI-generated images, videos and conversations can influence human memory, sometimes convincing individuals that they witnessed events that never actually occurred.

The findings raise concerns about the growing relationship between artificial intelligence and human cognition, particularly as millions of people increasingly rely on AI systems to summarize information, reconstruct past events and assist with everyday decisions.

Researchers warn that the technology’s ability to generate convincing but inaccurate information could create new challenges for personal memory, legal investigations and public trust.

Researchers at the Massachusetts Institute of Technology, led by MIT Media Lab professor Pattie Maes, conducted experiments examining how artificial intelligence influences people’s recollections.

In one study, 200 participants viewed 24 ordinary photographs depicting scenes such as weddings, city streets and sporting events.

After a short break, some participants saw the original photographs again, while others were shown AI-modified versions containing altered facial expressions, additional people or environmental changes.

The results revealed that participants exposed to manipulated images were more likely to remember details that had never appeared in the original photographs.

Even more concerning, many expressed greater confidence in those inaccurate memories.

In related experiments, AI-generated videos based on altered photographs produced approximately twice as many false recollections as unmodified images.

The findings demonstrate how realistic synthetic media can influence not only what people remember but also how certain they feel about those memories.

Researchers also investigated whether conversations with AI chatbots could change how people remembered a crime.

Participants watched a simulated criminal incident before answering questions from different types of interviewers.

Those questioned by a generative AI chatbot developed more than three times as many immediate false memories as participants in the control group.

Some inaccurate recollections persisted when participants were tested again one week later.

These results have significant implications for criminal investigations, where eyewitness testimony can influence legal proceedings.

If AI-assisted questioning introduces misleading details, witnesses may later confuse those suggestions with events they actually observed.

The research suggests that using AI in sensitive investigative settings requires careful safeguards.

Other studies have demonstrated that AI-generated misinformation can influence memories involving entertainment, politics and historical events.

In one experiment, nearly half of participants reported remembering fictional movie remakes that had never been produced, including an imaginary version of The Matrix starring Will Smith.

Additional experiments found that misleading AI-generated material could encourage people to remember political scandals that never occurred.

However, researchers noted that manipulated videos were not consistently more effective than written misinformation.

These findings suggest that the problem involves more than sophisticated deepfake technology.

The persuasive presentation of inaccurate information, whether visual or written, can affect human recollection.

A separate experiment involving 331 participants examined how AI summaries influence memories of a traffic accident. Participants watched an animated video showing a vehicle striking a pedestrian.

Later, they received summaries describing the incident.

Among those who read accurate summaries, approximately 84% correctly remembered a traffic sign.

Among participants exposed to misleading summaries, that figure fell to approximately 45%.

Importantly, warning participants that the summaries were AI-generated did not eliminate the effect.

The results challenge the assumption that human oversight can always correct artificial intelligence errors.

Scientists are increasingly studying a phenomenon sometimes described as cognitive surrender, in which people rely heavily on technology rather than independently evaluating information.

AI systems can provide valuable assistance with learning, memory and decision-making, but researchers caution that excessive dependence may create new vulnerabilities.

The findings do not establish that ordinary AI use inevitably damages memory or causes permanent cognitive decline.

Instead, they reveal how misleading AI-generated information can alter recollections under experimental conditions.

As artificial intelligence becomes integrated into education, healthcare, journalism and legal systems, understanding these effects will become increasingly important.

The central concern is that AI may not simply change how people access information about the past—it could also influence what they believe the past contained.