Conversational AI Powered by Large Language Models Amplifies False Memories in Witness Interviews (2408.04681v1)

Abstract: This study examines the impact of AI on human false memories -- recollections of events that did not occur or deviate from actual occurrences. It explores false memory induction through suggestive questioning in Human-AI interactions, simulating crime witness interviews. Four conditions were tested: control, survey-based, pre-scripted chatbot, and generative chatbot using a LLM. Participants (N=200) watched a crime video, then interacted with their assigned AI interviewer or survey, answering questions including five misleading ones. False memories were assessed immediately and after one week. Results show the generative chatbot condition significantly increased false memory formation, inducing over 3 times more immediate false memories than the control and 1.7 times more than the survey method. 36.4% of users' responses to the generative chatbot were misled through the interaction. After one week, the number of false memories induced by generative chatbots remained constant. However, confidence in these false memories remained higher than the control after one week. Moderating factors were explored: users who were less familiar with chatbots but more familiar with AI technology, and more interested in crime investigations, were more susceptible to false memories. These findings highlight the potential risks of using advanced AI in sensitive contexts, like police interviews, emphasizing the need for ethical considerations.

• The paper demonstrates that a generative chatbot induces false memories at rates nearly three times higher than control conditions.
• It reveals that users maintain high confidence in their false memories, with persistence observed one week after the initial interaction.
• The study underscores the need for ethical safeguards in legal and forensic applications to mitigate AI-induced memory distortions.

Conversational AI Powered by LLMs and False Memory Formation

Overview

The paper "Conversational AI Powered by LLMs Amplifies False Memories in Witness Interviews," authored by Samantha Chan, Pat Pataranutaporn, Aditya Suri, Wazeer Zulfikar, Pattie Maes, and Elizabeth F. Loftus, presents an empirical paper investigating how interaction with AI-powered conversational agents influences the formation of false memories. The paper specifically examines the effects of a generative chatbot using a LLM on false memory induction compared to other methods, including survey-based and pre-scripted chatbot interactions.

Methodology

A total of 200 participants were recruited to paper false memory formation under four different conditions: control, survey-based, pre-scripted chatbot, and generative chatbot. The experiment entailed two phases: an immediate assessment followed by an assessment occurring one week later.

In Phase 1, participants:

1. Watched a pre-recorded silent CCTV video of an armed robbery.
2. Completed emotional assessments using the Self-Assessment Manikin (SAM) scale.
3. Engaged in filler tasks to create temporal gaps.
4. Were randomly assigned to one of the four experimental conditions to answer questions about the video.

The generative chatbot leveraged an LLM to provide interactive, context-sensitive feedback to participant answers, especially focusing on misleading questions designed to induce false memories.

In Phase 2, participants were re-evaluated one week later using the same follow-up questions to assess the persistence of false memories.

Results

The paper's results elucidate several key findings:

1. Increased False Memory Induction by Generative Chatbot: The generative chatbot significantly increased immediate false memory formation compared to the other conditions. It induced roughly three times more false memories than the control condition and 1.7 times more than the survey-based approach.
2. Maintained Confidence in False Memories: Users interacting with the generative chatbot exhibited a high level of confidence in their false memories, both immediately and one week after the initial interaction.
3. Persistent False Memories: For participants in the generative chatbot condition, the number of false memories remained constant over the one-week period, which contrasts with the increase in false memories observed in the control and survey-based conditions.
4. Moderating Factors: Participants who were less familiar with chatbots but more familiar with AI were more susceptible to false memories. Similarly, those with a heightened interest in crime investigations were also more prone to memory distortions.

Discussion

The paper confirms the powerful influence of interactive, AI-driven conversational agents on memory malleability. The generative chatbot’s ability to provide detailed, contextually relevant feedback appears to significantly enhance both the formation and confidence in false memories. This affirms the primary hypothesis that LLM-powered chatbots induce not only more false memories but also more resilient ones.

The implications are substantial and multifaceted:

1. Forensic and Legal Applications: The use of AI in legal contexts, especially in witness interviews, must be carefully considered given the potential for significant memory distortion. The melding of AI interrogation methods with traditional legal practices could inadvertently compromise the reliability of eyewitness testimony.
2. Behavioral and Cognitive Sciences: This research contributes to the comprehensive understanding of false memory formation, adding a novel dimension to the intersection between AI and human cognition.
3. Ethical Considerations: There is a critical need for ethical guidelines governing the deployment of AI systems in contexts sensitive to memory accuracy. Ensuring user awareness of the potential for AI-induced misinformation is paramount.

Future Directions

Future research should explore mitigating strategies to counteract AI-induced false memories. Effective measures could include developing AI systems that flag potential misinformation or encourage user skepticism. Additionally, leveraging AI's capacity to induce positive false memories could offer therapeutic benefits, such as mitigating PTSD symptoms.

Longitudinal studies extending beyond the one-week mark will be vital for understanding the long-term effects and durability of AI-induced false memories. Finally, the expanding capabilities of multimodal AI systems should be studied to assess their potential to further influence memory through immersive, sensory interactions.

Conclusion

"Conversational AI Powered by LLMs Amplifies False Memories in Witness Interviews" provides compelling evidence of the profound impact that generative chatbots can have on human memory. The data underscore the need for a cautious and ethical approach to integrating AI into domains where memory fidelity is critical. The research presents a robust call to action for the scientific community to develop safeguards and ethical standards to navigate the burgeoning capabilities of AI in human-AI interactions.