{"slug": "are-we-still-friends-the-importance-of-memory-updating", "title": "Are We Still Friends? The Importance of Memory Updating", "summary": "Human memory constantly updates relationship statuses and recent events, a process called memory updating that is crucial for social interactions and daily decisions. AI systems struggle with this updating when they have large existing knowledge structures, highlighting a limitation compared to human cognition.", "body_md": "######\n[Memory](/us/basics/memory)\n\n# Are We Still Friends? The Importance of Memory Updating\n\n## Memory updating helps us track changes, but may be a limitation for AI systems.\n\nPosted June 26, 2026\n[\nReviewed by Monica Vilhauer Ph.D.\n](/us/docs/editorial-process)\n\n### Key points\n\n- Keeping track of the status of friendships requires constantly updating our memory.\n- Updating based on autobiographical memories lets us know where we parked our car and choose where to eat.\n- Scientists update theories based on more recent and better evidence.\n- AI models have problems updating based on recent evidence when they have a large existing knowledge structure.\n\nAre we still friends? Knowing the status of a relationship depends on tracking changes. What have we done together recently? Was it fun or were we fighting? Admittedly, keeping track of the changes to a relationship may have been overwhelming when you were in middle school. But we always work to update our relationship status.\n\n## Memory Updating and Relationships\n\nOne of the most amazing features of human [memory](https://www.psychologytoday.com/us/basics/memory) is our ability to track recent experiences while also knowing the general context. I’ve argued that this matters for maintaining and tracking relationships of all kinds (Hyman, 1999). We sometimes need to modify and change our memories to adapt to our social groups. We also decide what stories to share with one another based on how long it has been since we’ve seen each other (Drivdahl et al., 2014). When talking with family at dinner, we talk about what happened each day. When we meet friends we haven’t seen in months, we share the important events from that time period.\n\nKnowing what counts as recent depends on the context. People are amazingly good at this. How good we are at tracking recent events and conversations becomes particularly clear when someone loses the ability. [ Conversation loops occur in Alzheimer’s disease](https://www.psychologytoday.com/us/blog/mental-mishaps/201712/stuck-repeat-in-alzheimer-s-disease) because someone can no longer track what they’ve told you recently — sometimes as recent as five minutes ago — and sometimes a little longer. But for the rest of us, we generally know when we are repeating a previous conversation. We have updated our relationship status.\n\n## Where Should We Go for Dinner\n\nAutobiographical memory allows us to stay updated in areas well beyond relationship status. You want to know where your car is parked now, not where it was last week or where it is usually parked (da Costa Pinto & Baddeley, 1991; Hyman et al., 2013). Our ability to remember recent events is displayed across a great variety of experiences (Baddeley & Hitch, 1977). Everything from the last movie we saw to where we went on our last vacation comes to mind easily when we think about that type of event. Older memories tend to blur together into the general type of event – going to the movies or going on vacation.\n\nThe ability to remember the most recent events is called a recency effect. Recency effects allow us to solve one of life’s greatest challenges: Where should we go for dinner? We often think about places we’ve been recently. Maybe we want to go back, but most likely we decide to do something different. Of course, we do sometimes go back to the same places. When I meet a good friend for a beer, we generally go to our favorite local brewery.\n\n## Knowledge Updating in a Changing World\n\nWe not only update our personal experiences, but we also update our knowledge about the state of the world. We keep track of who the current president is. We have some ideas about the current news. We know about the release of new music from our favorite artists. If something new and important happens we track that information.\n\nKnowledge updating is particularly important for science. Sometimes we discard old ideas when new findings are reported. Of course, dropping old theories and beliefs is hard. Many of these ideas hang around in textbooks, the popular press, and in public beliefs. A [ zombie idea ](https://www.psychologytoday.com/us/blog/mental-mishaps/202110/how-to-kill-a-zombie-idea)is something that is known to be incorrect, but that keeps coming back around no matter how many times new data tries to kill it.\n\nWhen I teach, I often teach the old idea and then describe new evidence that disproves that theory. We then discuss our current theories. And I always note that the next set of research may require that we update our understanding of the world yet again. That’s the heart of the scientific method.\n\n## AI and Knowledge Updating\n\nIn thinking about this, I became curious about how some of the current [AI](https://www.psychologytoday.com/us/basics/artificial-intelligence) Large Language Models would handle knowledge updating. I was interested in cases when there is an older idea or theory that newer evidence disproves. AI systems are built with a lot of information. They use models that create text based on the entire set of knowledge. What happens when you ask about some of these topics? I tried a few.\n\nFirst, I tried: Does culture influence perception of the Muller-Lyer illusion. This is a classic illusion with lines the same length appearing different based on arrows at the end of the lines (see the example). The old idea was that culture was critical. People who didn’t grow up in context with lots of right angles didn’t perceive the illusion – this was called the carpentered-world hypothesis. When I asked Google, the AI answer presented this as the standard explanation. But more recent evidence shows the illusion works for all people, even with curves instead of angles. It shows up in some non-human animals. Google AI even cited the recent paper by Amir and Firestone (2025) summarizing this new understanding but didn’t actually use the current understanding.\n\nI have tried a few others with a variety of AI responses. It didn’t do too well with the zombie idea that listening to music makes kids smarter, the discredited Mozart effect (Mehr et al., 2013; Steele, et al., 1999). Google AI knew it didn’t increase [intelligence](https://www.psychologytoday.com/us/basics/intelligence), but thought it increased other cognitive abilities. It did better with the zombie idea that releasing pent up anger decreases [aggression](https://www.psychologytoday.com/us/basics/anger). But there is substantial and long running evidence against that false claim.\n\nI suspect that AI systems will have challenges updating knowledge. When a large amount of information is based on an old, but discredited, idea, AI may stick with the more well-known, but discredited response. Admittedly people struggle with this too. But in these cases, AI will continue to promote zombie ideas. AI may make it harder for us to update knowledge in the world based on newer and better evidence.\n\nKnowledge and memory updating is a critical human capability. Updating allows us to update our friendships and track the current state of the world. From knowing where our car is to knowing who our friends are, we have to continually update our knowledge.\n\nReferences\n\nAmir, D., & Firestone, C. (2025). Is visual perception WEIRD? The Müller-Lyer illusion and the cultural byproduct hypothesis. *Psychological Review*.\n\nBaddeley, A. D., & Hitch, G. J. (1977). Recency Reexamined. In *Attention and performance VI* (pp. 647-667). Routledge.\n\nda Costa Pinto, A. A. N., & Baddeley, A. D. (1991). Where did you park your car? Analysis of a naturalistic long-term recency effect. *European Journal of Cognitive Psychology*, *3*(3), 297-313.\n\nDrivdahl, S., Hiatt, R., Arnold, M., & Hyman, I. E., Jr. (2014, November). Playing to the crowd: Conversation partner determines temporal distance of past and future autobiographical events. Poster presented at the meeting of Psychonomic Society. Long Beach, CA.\n\nHyman, I. E., Jr. (1999). Creating false autobiographical memories: Why people believe their memory errors. In E. Winograd, R. Fivush, & W. Hirst (Eds.), *Ecological approaches to cognition: Essays in honor of Ulric Neisser* (pp. 229-252). Hillsdale, NJ: Erlbaum.\n\nHyman, I. E., Jr., Roach, A., & Drivdahl, S. (2013, June). The Updating Function of Autobiographical Memory: Keeping the Self Located in Time, Place, and Relationships. Paper presented at the meeting of SARMAC (Society for Applied Research in Memory and Cognition). Rotterdam.\n\nMehr, S. A., Schachner, A., Katz, R. C., & Spelke, E. S. (2013). Two randomized trials provide no consistent evidence for nonmusical cognitive benefits of brief preschool music enrichment. *PloS one*, *8*(12), e82007.\n\nSteele, K. M., Bass, K. E., & Crook, M. D. (1999). The mystery of the Mozart effect: Failure to replicate. *Psychological Science*, *10*(4), 366-369.", "url": "https://wpnews.pro/news/are-we-still-friends-the-importance-of-memory-updating", "canonical_source": "https://www.psychologytoday.com/us/blog/mental-mishaps/202606/are-we-still-friends-the-importance-of-memory-updating", "published_at": "2026-06-27 01:06:35+00:00", "updated_at": "2026-06-27 01:07:36.169033+00:00", "lang": "en", "topics": ["artificial-intelligence", "machine-learning", "large-language-models", "ai-research", "ai-ethics"], "entities": ["Alzheimer's disease", "Monica Vilhauer"], "alternates": {"html": "https://wpnews.pro/news/are-we-still-friends-the-importance-of-memory-updating", "markdown": "https://wpnews.pro/news/are-we-still-friends-the-importance-of-memory-updating.md", "text": "https://wpnews.pro/news/are-we-still-friends-the-importance-of-memory-updating.txt", "jsonld": "https://wpnews.pro/news/are-we-still-friends-the-importance-of-memory-updating.jsonld"}}