The emergence of emotionally intelligent agents—designed for both stable temperament and dynamic interaction—has commenced, as evidenced by two unrelated research papers released last week.
The timing is critical. Nearly every day, news reports have been highlighting cases where chatbots have inadvertently encouraged emotionally vulnerable users to harm themselves or others. However, when viewed collectively, the studies indicate that AI is advancing into a domain where personality and emotion can significantly influence how agents reason, communicate, and negotiate.
One research team illustrated how to prepare large language models with enduring psychological archetypes, while another showcased that agents can develop emotional strategies throughout multi-turn negotiations.
Personality and emotion are no longer mere superficial enhancements for AI—they are evolving into essential features. Static temperaments render agents more predictable and reliable, while adaptive strategies enhance performance in negotiations and create interactions that feel strikingly human.
However, this same credibility prompts difficult questions: If an AI can flatter, persuade, or debate with emotional sophistication, who is accountable when these methods verge on manipulation, and how can one even assess “emotional alignment” in systems designed to influence feelings as well as logic?
Also see: South Africa Denies Nigeria Landing Authorization for Delegation
In the study titled Psychologically Enhanced AI Agents, Maciej Besta from the Swiss Federal Institute of Technology in Zurich and his colleagues introduced a framework known as MBTI-in-Thoughts. Instead of retraining models, they utilize prompt engineering to secure personality traits along the dimensions of cognition and affect.
”Utilizing the Myers-Briggs Type Indicator (MBTI), our approach equips agents with specific personality archetypes through prompt engineering,” the authors stated. This enables “control over behavior along two fundamental dimensions of human psychology, cognition and affect,” they further explained.
The researchers validated this by assigning language models characteristics such as “emotionally expressive” or “analytically minded.”
