Can AI aid with memory like those used by services such as ChatGPT? Decades of research have consistently shown that human memory is fallible. Memories are not static snapshots of the past but are subject to forgetting and distortion. Even relatively recent events can be misremembered, leading to inaccuracies in our recollections.
For this reason and more, scientists at Ben-Gurion University in Israel chose to conduct research into the question of whether or not AI can help us verify memories.
Given that much of our understanding of the world is based on the memories of others, it’s crucial to consider the potential unreliability of this information. How can we make informed decisions when our sources of knowledge are prone to errors?
Will you offer us a hand? Every gift, regardless of size, fuels our future.
Your critical contribution enables us to maintain our independence from shareholders or wealthy owners, allowing us to keep up reporting without bias. It means we can continue to make Jewish Business News available to everyone.
You can support us for as little as $1 via PayPal at [email protected].
Thank you.
Dr. Talya Sadeh of the Department of Cognitive and Brain Sciences at Ben-Gurion University of the Negev decided to examine this question and understand how humans can recognize and verify that others’ memories are accurate.
“A lot of 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 examined how we manage to base knowledge, sometimes really important knowledge, on the basis of memories that are not always reliable, and can natural language processing models, such as those we all know (e.g., ChatGPT and others), help us identify the truth of memory?”
The comparison to a machine learning model was based on the words most indicative of accurate or inaccurate memory in both human and machine responses. Of the 20 most informative words, 14 were shared by humans and the machine. These results suggest that humans possess the ability to directly assess the veracity of others’ memories, relying on similar cues as a language model.
However, we found that predicting the reliability of others’ memories could be improved by 10% if evaluators were asked to judge the quality of the memory instead of directly assessing its accuracy. Specifically, focusing on the richness, vividness, and detail of the sensory experience, as well as the confidence level of the memory sharer, yielded more accurate predictions.
“Humans have the ability to take advantage of their being social creatures to learn quickly from others. Much of the knowledge of humans comes from the fact that we share real-life experiences with each other, and we have succeeded in showing that the machine cannot yet take our place when it comes to personal memory,” concludes Dr. Sadeh.
