MIT’s Computational Model Outperforms in Predicting Emotions

MIT neuroscientists have developed a computational model that demonstrates remarkable success in accurately predicting human emotions within social situations. Using the prisoner’s dilemma game as a foundation, this model incorporates factors such as desires, expectations, and the influence of observers to deduce motivations, compare outcomes against expectations, and forecast emotions. By replicating human social intelligence, this model has surpassed other emotion prediction models, prompting researchers to explore its adaptability for wider applications.

The model designed by the MIT team approximates a fundamental aspect of human social intelligence by utilizing insights into how individuals perceive and understand the emotions of others. During interactions, we often engage in the cognitive task of anticipating how others will respond to our words and actions. This ability, known as theory of mind, assists in inferring beliefs, desires, intentions, and emotions of other individuals.

The newly developed computational model by MIT neuroscientists can predict a range of emotions experienced by others, including joy, gratitude, confusion, regret, and embarrassment. It effectively emulates the social intelligence of human observers. The model’s foundation lies in predicting the emotional responses of individuals involved in situations inspired by the prisoner’s dilemma, a classic game theory scenario where two participants must choose between cooperation and betrayal.

While previous research focused extensively on training computer models to infer emotional states based on facial expressions, MIT Professor Rebecca Saxe emphasizes that the key aspect of human emotional intelligence lies in the ability to predict emotional responses before events unfold. She states that anticipating others’ emotions prior to their occurrence is the most significant factor in understanding human emotions, as reactive emotional intelligence would be inadequate.

To model how human observers make these predictions, the researchers employed scenarios from the British game show “Golden Balls.” In this show, contestants are paired and presented with a $100,000 prize, requiring negotiation to decide whether to split the sum or attempt to steal it. The outcome of this decision elicits a range of emotions, such as joy and relief from a split, surprise and fury from a theft, or guilt mixed with excitement from a successful steal.

The researchers devised a computational model comprising three modules to predict these emotions accurately. The first module utilizes inverse planning to infer a person’s preferences and beliefs based on their actions. By predicting contestants’ motivations, this module provides insights into their expectations. Specific player knowledge, such as occupation, is integrated into the model to enhance its ability to infer likely motivations.

The second module compares the game’s outcome with the contestants’ desired and expected outcomes. Finally, the third module predicts the emotions experienced by the contestants based on the known expectations and outcome. The researchers trained this module using predictions from human observers, aiming to mimic how observers causally reason about each other’s emotions rather than representing actual emotions.

Once the three modules were operational, the researchers assessed their emotion predictions against those made by human observers using a new dataset from the game show. This computational model significantly outperformed previous models in predicting emotions.

The model’s success stems from its incorporation of key factors utilized by the human brain when predicting others’ reactions. These factors include evaluations and emotional responses based on desires, expectations, and concerns about how others perceive them. Rebecca Saxe highlights that the core intuitions of the model encompass these factors, illustrating that underlying emotions are influenced by desires, expectations, outcomes, and observers’ presence. She further suggests that people’s desires extend beyond material gains, encompassing concepts like fairness and avoiding exploitation.

Nick Chater, a professor of behavioral science at the University of Warwick, lauds the researchers for their deeper understanding of emotions’ impact on human behavior. He commends their work, which not only explains how people’s actions can reveal their underlying emotions but also illuminates emotions’ crucial and nuanced role in human social behavior.

In future studies, the researchers plan to refine the model to make more generalized predictions encompassing diverse scenarios beyond the game-show context. Additionally, they aim to develop models that can predict game outcomes solely based on contestants’ facial expressions after the results are announced.

The research received funding from the McGovern Institute, the Paul E. and Lilah Newton Brain Science Award, the Center for Brains, Minds, and Machines, the MIT-IBM Watson AI Lab, and the Multidisciplinary University Research Initiative.

Frequently Asked Questions (FAQs) about Emotion prediction

More about Emotion prediction

• MIT News: Decoding Emotional Intelligence: MIT’s Computational Model Excels in Predicting Emotions
• Philosophical Transactions A: Emotion prediction as computation over a generative theory of mind