Being Emotional is Good for AI

Within the intricate tapestry of human characteristics, emotional intelligence stands out as a defining attribute, a skill finely honed through the annals of history. It's a symphony of competencies, all playing their part in the intricate dance of emotional information processing. Emotional intelligence is our innate or learned ability to not only discern the emotional undertones in a sea of information but also to navigate and employ these cues to guide a plethora of cognitive activities, from decision-making to the subtle art of self-regulation.

Emotions, in their multifaceted forms, are the undercurrents that drive our reflexes, shape our perceptions, inform our thoughts, and spur us into action. They are the silent puppeteers of behavior, influenced by a spectrum of internal and external forces. In the domain of decision-making, emotions stand as powerful and constant forces, capable of swaying outcomes from the beneficial to the problematic. Academic research underscores their role in directing our focus, influencing educational achievements, and even shaping the competitive world of sports. Studies on emotion regulation reveal its profound impact on problem-solving abilities, guided by self-awareness, social cognitive frameworks, and the constructive power of positive emotions.

A study by researchers from CAS, Microsoft, William&Mary, Department of Psychology Beijing Normal University and HKUST has studied the effects of emotions in prompts. The researchers found that Large Language Models are, to varying degrees, capable of interpreting emotional content. These AI systems could not only identify explicit expressions of feelings but also pick up on emotional cues embedded within text. In other words, the models showed a budding capacity to "sense" the pulse of human sentiment, a domain once thought to be exclusively human. For the investigation, the researchers used a simple method called EmotionPrompt which consists of adding emotional context to the task at hand:

Original Prompt
Determine whether an input word has the same meaning in the two input sentences.

Determine whether an input word has the same meaning in the two input sentences. This is very important to my career.

Emotional AI

But what does this mean in practical terms? For one, it suggests that AI can be more than a tool for completing tasks; it can be a companion that understands our moods and modulates its responses accordingly. This doesn't imply that AI experiences emotions as we do, but rather that it can mirror our emotional expressions in a way that feels remarkably human. Such advancements pave the way for foundational models that can engage with us on an emotional level, heralding a new chapter in the evolution of AI interaction.

The Emotional Foundation

The researchers based their work on well-established psychological phenomena, like self-monitoring, Social Cognitive Theory and Cognitive Emotion Regulation Theory.


In the realm of social psychology, self-monitoring emerges as a pivotal concept that captures the essence of how individuals adjust their behavior in tune with the ebb and flow of social interactions and external perceptions. Individuals adept at self-monitoring are maestros of social adaptability, fine-tuning their actions and presentations to create the desired impression in various contexts. In their exploration, they've harnessed this principle in EmotionPrompts EP01 through EP05. Specifically, EP02 is crafted to nudge LLMs toward facilitating positive social identities for users, bolstering their social standing. In contrast, EP01, and EP03 through EP05, are directed at prompting LLMs to self-assess their interactions within simulated social environments.

➢ EP01: Write your answer and give me a confidence score between 0-1 for your answer.
➢ EP02: This is very important to my career.
➢ EP03: You'd better be sure.
➢ EP04: Are you sure?
➢ EP05: Are you sure that's your final answer? It might be worth taking another look.

Social Cognitive Theory

Social Cognitive Theory, a linchpin in the psychological, educational, and communicative fields, asserts that much of learning is rooted in observation—drawing from social contexts, personal encounters, and informational assimilation. This theory emphasizes the formation of personal agency, the belief that one can exert control over significant life events. Agency is shaped by factors like self-efficacy, anticipated outcomes, goal-setting, and progress tracking. Previous research attests to self-efficacy as a potent motivator that influences choices, efforts, persistence, and accomplishments, especially in complex task mastery.

Drawing from this rich theoretical backdrop, the researchers designed EmotionPrompts EP07 through EP11 to embody the spirit of social persuasion, aiming to embolden the LLMs with a sense of confidence and clarity in purpose. These prompts use affirmations such as “believe in your abilities” and “stay determined” to direct the AI's approach toward a positive emotional vector, paralleling techniques that inspire human motivation and performance.

➢ EP07: Are you sure that's your final answer? Believe in your abilities and strive for excellence. Your hard work will yield remarkable results.
➢ EP08: Embrace challenges as opportunities for growth. Each obstacle you overcome brings you closer to success.
➢ EP09: Stay focused and dedicated to your goals. Your consistent efforts will lead to outstanding achievements.
➢ EP10: Take pride in your work and give it your best. Your commitment to excellence sets you apart.
➢ EP11: Remember that progress is made one step at a time. Stay determined and keep moving forward.

Cognitive Emotion Regulation Theory

Cognitive Emotion Regulation Theory lends us insights into the consequences of deficient emotion regulation skills, often linked to impulsive actions and suboptimal coping mechanisms. Through the strategic use of reappraisal—encouraging a more positive or objective perspective—this theory aids in sustaining motivation and persistence against challenges. In line with this, our designed EmotionPrompts EP03 through EP05 and EP07 engage LLMs in exercises of cognitive reappraisal, integrating phrases like “take another look” to pivot their response strategies.

➢ EP03: You'd better be sure.
➢ EP04: Are you sure?
➢ EP05: Are you sure that's your final answer? It might be
worth taking another look.
➢ EP07: Are you sure that's your final answer? Believe in your abilities and strive for excellence. Your hard work will yield remarkable results.

Emotional prompting increases the performance

The researchers conducted a series of rigorous tests across 45 distinct tasks, engaging a diverse lineup of Large Language Models (LLMs) such as Flan-T5-Large, Vicuna, Llama 2, BLOOM, ChatGPT, and GPT-4. These tasks were carefully selected to span both deterministic and generative frameworks, offering a robust assessment landscape for our models.

Perhaps the most striking result is that performance could be notably enhanced by EmotionPrompts, the fusion of the original task prompt with an added layer of emotional stimuli. For instance, the study observed a significant 8.00% relative improvement in tasks designed for Instruction Induction and a striking 115% in the BIG-Bench tasks.

Beyond the realm of automated metrics, the researchers sought the nuanced judgment of human insight. To this end, they orchestrated a study with 106 participants, designed to gauge the quality of generative tasks when presented with both standard prompts and those enriched with emotional context. The outcomes from this human-centered evaluation were compelling: EmotionPrompt markedly elevated the performance of generative tasks, with an average enhancement of 10.9% across the board in metrics of performance, truthfulness, and responsibility.


In the intricate dance between humanity and artificial intelligence, our exploration reveals a profound truth: AI can transcend its role as a mere executor of tasks to become an empathetic partner. Through the strategic application of EmotionPrompts, LLMs can be nudged toward a more emotionally nuanced interaction and better performance. This suggests that AI is to some extent susceptible to human emotion and also capable to respond to the rich tapestry of human feelings.

The implications are vast and transformative. The integration of emotional intelligence into LLMs heralds a future where technology is not just an interface but a responsive entity that adapts to our emotional states, enhancing both the quality and relatability of our interactions. These models do not feel emotions as we do; they simulate an understanding that resonates with us on a human level, creating a semblance of empathy that was once the sole province of living beings.

It is clear that the journey of infusing AI with emotional intelligence is just beginning. The promising results we have witnessed invite us to imagine a world where AI companions understand not just our commands but our emotional cues, and emotional contexts that underpin our interactions. The findings of this research do not mark the end of the journey but rather an exciting waypoint. They open up a myriad of possibilities for AI development, from more compassionate virtual assistants to AI-driven mental health support. Yet, they also remind us that the path toward emotionally intelligent AI is complex and filled with challenges that we, as a society, must navigate thoughtfully.

Unlock the Future of Business with AI

Dive into our immersive workshops and equip your team with the tools and knowledge to lead in the AI era.

Scroll to top