A police officer hears an eyewitness account of a car accident, a doctor diagnoses an illness based on symptoms listed by a patient, or a friend recommends a restaurant. Other people’s memories are a source of information for us that influences how we act. Cognitive and neuroscience researchers from Ben-Gurion University of the Negev have investigated how people judge the reliability of these memories and the difference between human and machine abilities.

The findings were is published recently in the journal Proceedings of the National Academy of Sciences.

Years of research suggest that our memories are subject to forgetting and distortion. Human memories are not accurate descriptions of the past, but are prone to error, even after short periods of time. The information from these memories is significant to us because much of our knowledge of the world is based on information from the memories of others. So how can we base information on memory that is not always reliable?

Dr. Talya Sadeh at the Department of Cognitive and Brain Sciences at Ben-Gurion University of the Negev decided to investigate this question and understand how people can recognize and verify the accuracy of others’ memories.

“Many people’s knowledge comes from sharing episodic memories with each other, knowledge that we use to make decisions, form opinions, and so on,” she noted. “My research explored how we manage to base knowledge, sometimes really important knowledge, on memories that are not always reliable, and can natural language processing models, like the ones we all know (eg ChatGPT and others), help us identify the truth of the memory?”

To this end, she conducted a study that simulated real-life situations where a person had to judge whether memories told by another person were true or not. For example: “I remember a woman was at a party because I remembered she was late and wearing a really nice dress” or “I remember the car didn’t stop at a red light because I noticed the speed before it reached the intersection , while the traffic light changed from green to yellow.”

Participants were asked to directly judge whether they believed the memories to be true or not based on such descriptions. They were then asked to rate the quality of the other person’s memory by giving a quantitative rating on questions such as how vivid and detailed the memory was and how confident the speaker sounded about it.

The comparison with a machine learner was made based on the words from the memory descriptions that were most indicative of correct (or incorrect) memory, among humans and the machine learning model. Of the 20 words that best indicated the accuracy of memory, 14 were shared by human and machine. Therefore, the results suggest that people have the ability to directly evaluate other people’s memories and determine whether they are true or false, and they do so based on much of the same information as a language model (and as well as the model).

However, the reliability of others’ memories can be predicted even better (with a 10% gap) if – instead of relying on a direct assessment of memory reliability – we ask evaluators to rate the characteristics of the memory: how much they believe it is associated with a rich, vivid and detailed sensory experience, and how safe the memory sharer sounds.

While a language model is based on the extraction of statistical rules, people’s decisions depend on their sensitivity to information that indicates the qualities of memory. This allows people to evaluate other people’s memories very well and perhaps even better than a language model trained to extract statistical rules from texts describing people’s memories.

Given that we use language to share mental statesthoughts and beliefs, this lifelong language learning serves as a tool for us to validate the memories and experiences of others. This study is an important step in deciphering how people think memory sharing and its importance in creating social knowledge.

“Humans have the ability to take advantage of being social creatures to quickly learn from others. Much of the knowledge about humans comes from the fact that we share real experiences with each other, and we have managed to show that the machine cannot yet take our place when it comes to personal memory“, concludes Dr. Sadeh.

The research team included Avi Gamran from the Department of Psychology and Lilach Lieberman from the Department of Cognitive and Brain Sciences at Ben-Gurion University of the Negev, prof. Ian Dobbins from Washington University in Missouri and prof. Michael Gilead of Tel Aviv University.