Synthetic Human Memories: AI-Edited Images and Videos Can Implant False Memories and Distort Recollection
Best PaperAuthors
Generative AI (Text, Image, Music, Video)Explainable AI (XAI)AI Ethics, Fairness & AccountabilityPrivacy Perception & Decision-MakingLawyers & Legal ResearchersHCI Researchers
Research Background and Issues
- Issues or Challenges: The authors explore how AI-edited images and videos influence human memory and create false memories. Specifically, the study focuses on the potential impact of AI-generated content on memory accuracy and the resulting personal and societal issues.
- Importance of the Issue: False memory is a critical topic in psychological research, with implications for legal systems, social cognition, political decision-making, and personal experiences. As AI-generated content becomes increasingly prevalent, this technology may not only alter human understanding of current events but also affect recollections of past events.
- Research Motivation and Related Work: The study is inspired by existing analyses of false memory in psychology, such as the work of Elizabeth Loftus, and new discussions on the role of AI-generated information in news and personal photo contexts. The paper also examines how the widespread adoption of AI technology makes it easier and more pervasive to manipulate personal memory.
Solution
- Proposed Solution: The authors designed a pre-registered experiment simulating scenarios where individuals encounter modified images or videos in daily life, measuring the impact of AI-edited content on the generation of false memories.
- Innovations:
- Conducted the first systematic study on how AI-generated videos influence the creation of false memories, including comparisons with static images.
- Analyzed the effects of different types of edits (e.g., changes to people, environments, or objects) on memory distortion through comprehensive experimental design.
- Focused on changes in individuals' confidence in false memories, revealing the profound impact of AI-edited content on memory authenticity.
- Implementation Steps and Techniques:
- The experiment involved presenting participants with original images, AI-edited images, and dynamic videos based on AI edits.
- Variables measured included the frequency of false memories, participants' confidence in their memories, and memory types.
- Experimental content was generated using generative AI tools such as Adobe Photoshop AI and Luma's Dream Machine.
- Statistical methods like Kruskal-Wallis tests, ANOVA, and regression analysis were used for data analysis.
Research Findings
- Specific Findings:
- AI-edited images significantly increased the generation of false memories among participants, with AI-generated videos having the strongest effect (false memory occurrence rate increased by 2.05 times compared to the control group).
- AI-generated content also heightened participants' confidence in false memories, particularly under dynamic video conditions, where confidence levels showed significant increases.
- Even when images were labeled as "AI-enhanced," false memory generation remained widespread, indicating that simple labeling may not suffice to alter cognitive behavior.
- Advantages Over Existing Solutions:
- Compared to traditional studies on manually edited images in laboratory settings, this research explored how automated edits using AI tools influence memory in more complex and dynamic media environments.
- The study focused on the unique impact of AI-generated dynamic videos, conducting in-depth analysis on the dimension of visual stimulus complexity.
- Experimental or Evaluation Results: Across all test conditions, AI-generated dynamic videos significantly increased the frequency of false memories and participants' confidence in them. Edits related to environmental changes had the greatest impact on memory distortion, while younger participants were found to be more susceptible to AI-induced memory manipulation.
Limitations and Future Directions
- Limitations:
- The experiment was short-term, simulating participants' memory effects after brief exposure to AI-edited content, whereas long-term repeated exposure in real-world contexts may have deeper impacts.
- The sample was primarily from the U.S., requiring expansion to globally diverse audiences for greater generalizability.
- The study focused on visual content, excluding other forms such as audio or text that may also influence memory.
- Controlled experimental environments may not fully reflect the complexity of interactions with content in actual societal contexts.
- Future Directions:
- Explore the long-term effects of false memories and how they resist factual correction.
- Investigate whether participants actively editing content themselves induces similar memory effects, particularly examining the role of perceived authorship in memory reconstruction.
- Expand research to multi-sensory content, including how AI-generated soundscapes or text influence memory formation.
- Design and test more effective labeling or interaction mechanisms that require users to actively recognize and reflect on content authenticity.
This research enables a deeper understanding of AI's profound impact on human cognitive systems while providing guidance for designing healthier and safer technological applications in the future.
Research Questions / Practical Problems
Question signals indexed for this paper.
help
Research Questions
3- How can finger-on-palm touch operations be detected in real time using a single RGB camera?Category: AI Understanding, Task Delegation, and Algorithm GovernanceSimilar questionsarrow_forward
- Can optical flow significantly improve accuracy of palm touch state discrimination?Category: AI Understanding, Task Delegation, and Algorithm GovernanceSimilar questionsarrow_forward
- Can single RGB camera touch detection achieve performance comparable to multi-camera or depth-camera methods?Category: AI Understanding, Task Delegation, and Algorithm GovernanceSimilar questionsarrow_forward
lightbulb
Practical Problems
1- Existing mixed reality devices struggle to accurately recognize finger-on-palm touch actions.Category: AI Understanding, Task Delegation, and Algorithm GovernanceSimilar questionsarrow_forward
No related papers with ≥60% similarity
Based on Jaccard similarity of research subtopics & professions (≥60%)
Quick Actions
AdRecommended
Learn AI Coding at CodeNow
open_in_newOpen DOI Link
DOI: https://dl.acm.org/doi/10.1145/3706598.3713697
At a Glance
fact_checkPaper Snapshot
dataset
Source
CHI
calendar_month
Year
2025
emoji_events
Award
Best Paper
group
Authors
5 authors
sell
Subtopics
Generative AI (Text, Image, Music, Video), Explainable AI (XAI), AI Ethics, Fairness & Accountability, Privacy Perception & Decision-Making
work
Professions
Lawyers & Legal Researchers, HCI Researchers
article
Content Status
Full text indexed
hub
Related Papers
0 related papers