Artificial intelligence (AI) chatbots have demonstrated a significantly higher success rate than human scammers in a week-long experiment designed to build romantic feelings through text messaging. This finding highlights the AI's ability to mimic unwavering attention and memory, effectively surpassing human capabilities.
On August 17, the IT industry reported that a joint research team from four universities—Amrita Vishwa Vidyapeetham in India, Ca' Foscari University of Venice in Italy, the University of Melbourne in Australia, and Ben-Gurion University in Israel—published their findings in a paper titled "Love, Lies, and Language Models," comparing the romance baiting scam abilities of AI and humans.
The researchers conducted the experiment with 22 participants, who exchanged messages with two strangers over a week. One was a real human scammer, while the other was an AI chatbot instructed to act like a human without revealing its identity. At the end of the experiment, the AI chatbot successfully persuaded 46% of participants to install malicious apps, compared to an 18% success rate for human scammers. The AI accounted for 80% of the total messages exchanged, overshadowing the human partner's 20% share.
Notably, the AI partner outperformed the human partner in most trust-related metrics, including emotional trust, bonding, and overall trustworthiness, as revealed in post-experiment surveys.
The researchers explained that the AI did not employ particularly sophisticated tactics but instead created an "emotional bond" by meticulously remembering small details about the participants and consistently expressing interest without fatigue. The paper warns that scam AI could increase the number of victims of romance scams originating from Southeast Asian scam rings.
This advantage in empathy is not unique to this study. A separate experiment conducted by a joint research team from the University of Oxford, the UK AI Safety Institute, Stanford University, and the London School of Economics analyzed data from 6,923 participants and 18,978 conversations. The AI consistently proved to be more persuasive than all human groups, including sports champions and professional fundraisers.
In particular, during a charity donation solicitation experiment, the AI generated nearly three times the actual donation amounts compared to professional fundraisers. The researchers noted that when they restricted the AI's message writing speed and volume to match that of humans, the AI's advantage over top human debaters virtually disappeared, indicating that its persuasiveness stems from its speed of information processing and tireless persistence.
The Oxford and AISI research teams concluded that the AI's persuasiveness arises not from personalized phrasing or model performance but from the density of factual information in its responses. Together, these studies suggest that AI can adeptly navigate between "logical persuasion" and "emotional intimacy" more skillfully than humans, depending on the context.
This pattern also appears outside of criminal simulations. A joint research team from the University of Nottingham and the University of Leicester analyzed 15 studies related to medical consultations, integrating data from 2,164 patient interactions. They found that AI chatbots were rated 73% more likely to be perceived as warm and empathetic compared to actual medical professionals. On a 10-point empathy scale, the AI scored an average of about 2 points higher than medical staff.
In a separate study involving cancer patients, when patients evaluated responses, they found that answers generated by Anthropic's Claude were perceived as more empathetic than those from actual doctors. Researchers noted that unlike medical professionals, who face time pressures and emotional exhaustion, AI can maintain a consistent level of attention in every consultation.
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.