Lecture / Seminar Family-Friendly Wheelchair Accessible
[email protected]
(902) 679 8604
Affiliation in human-AI interactions based on shared psychological traits by Dan Lametti
Abstract:
People connect or affiliate with others who share their likes, interests, and psychological traits. Does the same phenomenon occur when humans interact with artificial intelligence (AI) instructed to mimic their psychology? This study addressed this question using the large language model (LLM) GPT-4. In Experiment 1, GPT-4 was prompted to use language that mimicked either an anxious or non-anxious (calm) psychological state. One hundred participants had text-based conversations with both GPT-4 personas and then had their own anxiety measured. Participants with higher anxiety reported a stronger connection to GPT-4 when it mimicked anxiety, a distinction reflected in the sentiment of the messages they exchanged with the AI. Experiment 2 extended these findings to the personality trait of extraversion-introversion. One hundred participants had text-based conversations with two versions of GPT-4, one instructed to use extraverted language and the other instructed to use introverted language. Extraverted participants preferred interacting with an extraverted AI. In a final study, 100 participants had conversations with a version of GPT-4 instructed to mimic their complete personality profile—effectively, an AI doppelgänger—and a version instructed to present their opposite personality. Participants reported significantly more affiliation with their AI doppelgänger, a distinction also reflected in the sentiment of the messages they sent. Taken together, the results support affiliation in human-AI interactions based on the linguistic presentation of a shared psychology. The findings have important implications for the development of AI companions and the use of AI in healthcare.
[email protected]
(902) 679 8604
Affiliation in human-AI interactions based on shared psychological traits by Dan Lametti
Abstract:
People connect or affiliate with others who share their likes, interests, and psychological traits. Does the same phenomenon occur when humans interact with artificial intelligence (AI) instructed to mimic their psychology? This study addressed this question using the large language model (LLM) GPT-4. In Experiment 1, GPT-4 was prompted to use language that mimicked either an anxious or non-anxious (calm) psychological state. One hundred participants had text-based conversations with both GPT-4 personas and then had their own anxiety measured. Participants with higher anxiety reported a stronger connection to GPT-4 when it mimicked anxiety, a distinction reflected in the sentiment of the messages they exchanged with the AI. Experiment 2 extended these findings to the personality trait of extraversion-introversion. One hundred participants had text-based conversations with two versions of GPT-4, one instructed to use extraverted language and the other instructed to use introverted language. Extraverted participants preferred interacting with an extraverted AI. In a final study, 100 participants had conversations with a version of GPT-4 instructed to mimic their complete personality profile—effectively, an AI doppelgänger—and a version instructed to present their opposite personality. Participants reported significantly more affiliation with their AI doppelgänger, a distinction also reflected in the sentiment of the messages they sent. Taken together, the results support affiliation in human-AI interactions based on the linguistic presentation of a shared psychology. The findings have important implications for the development of AI companions and the use of AI in healthcare.
Pricing & Tickets
Pricing: Free
32 University Avenue
Wolfville, Nova Scotia
[email protected]
902-585-5242
View Full Venue Info
| Other Events at K.C. Irving Environmental Science Centre |
|---|
| Fri Dec 5, 2pm ALLways Learning |
| Sun Dec 7, 2pm Garden Room Concert |
| Jan 16, 2026, 2pm ALLways Learning |
| Jan 23, 2026, 2pm ALLways Learning |