When individuals engage in social interactions with others, they encounter a range of emotions. Additionally, they make conscious efforts to either evade or predict these emotional responses based on the words spoken or actions taken. Referred to as the theory of mind, this ability empowers people to deduce the thoughts, wishes, objectives and feelings of those around them.

A computational model which enables forecasting of a range of emotions in individuals was developed by MIT neuroscientists, including joy, gratitude, confusion, regret, and embarrassment. This model closely mimics the social intelligence exhibited by human observers.

It was specifically designed to anticipate the emotional responses of individuals involved in a scenario based on the prisoner’s dilemma. It is a classic game theory scenario in which two people must decide whether to help and cooperate with their partner or betray them.

The construction of the model involved integrating various factors that are believed to impact an individual’s emotional responses. These factors encompassed the person’s desires, expectations in each situation, and whether their actions were being observed. By considering these elements, the researchers aimed to create a comprehensive framework that could capture the complexities of human emotional reactions.

By incorporating these factors, the computational model developed by the researchers aimed to approximate how individuals might express emotions in different contexts. This computational modelling advancement brings humanity closer to unravelling the mysteries of human emotions and enhances the understanding of how individuals perceive and respond to various situations.

Rebecca Saxe, the John W. Jarve Professor of Brain and Cognitive Sciences, a member of MIT’s McGovern Institute for Brain Research, and the study’s Senior Author stated that although comprehensive research has focused on training computer models to infer an individual’s emotional state through facial expressions, it is not the most crucial element of human emotional intelligence. The most critical factor is the capability to anticipate and predict someone’s emotional reaction to events before they occur. This ability holds greater significance in human emotional intelligence.

To simulate the prediction-making process of human observers, the researchers utilised scenarios taken from a British game show named “Golden Balls.” Depending on the game’s outcome, contestants may experience various emotional states, such as joy and relief when both contestants choose to share the winnings, surprise and anger if one contestant steals the pot, or a mix of guilt and excitement when successfully stealing the winnings.

The researchers devised three distinct modules to develop a computational model capable of predicting these emotions. The first module was trained to infer a person’s preferences and beliefs by analysing their actions, employing a technique known as inverse planning.

The second module assesses the game’s outcome with each player’s desired and anticipated outcomes. Subsequently, the third module utilises this information along with the contestants’ expectations to forecast the emotions they might be experiencing.

After implementing and activating the three modules, the researchers employed them on a new dataset obtained from the game show to evaluate the accuracy of the models’ emotion predictions compared to those made by human observers. The results demonstrated a significant improvement in the model’s performance compared to any previous model designed for emotion prediction.

In the future, the researchers are ready to enhance the model’s capabilities by further extending its predictive performance to various scenarios.

Machine-learning models are utilised in the real world to assist radiologists in identifying potential diseases in X-rays; however, these models are intricate and their prediction process remains elusive even to their creators. To address this, researchers employ saliency methods, techniques that seek to offer insights into the model’s behaviour and elucidate its decision-making procedure.

Researchers from the Massachusetts Institute of Technology (MIT) and a multinational technology company have collaboratively developed a tool with a new method to assist users in selecting the most suitable saliency method for their specific requirements. Therefore, they introduced saliency cards, providing standardised documentation summarising how a particular process of saliency operates, including its strengths, weaknesses, and explanations to aid users in correctly interpreting the method’s outputs.

The Co-lead Author, Angie Boggust, a graduate student in electrical engineering and computer science at MIT and a member of the Visualization Group of the MIT Computer Science and Artificial Intelligence Laboratory (CSAIL), expresses the team’s aspiration that users equipped with this knowledge will be able to consciously select a suitable saliency method based on the specific machine-learning model being employed and the task it aims to accomplish.

Boggust explains that saliency cards are purposefully crafted to provide a concise and easily understandable overview of a saliency method while highlighting the essential attributes most relevant to human users. These cards are intended to be accessible to a wide range of individuals, including machine-learning researchers and even those unfamiliar with the field and seeking guidance in selecting a saliency method for the first time.

Choosing the “wrong” saliency method can have serious consequences. For instance, one saliency method known as integrated gradients compares the importance of features in an image to a meaningless reference point. Features with the highest priority compared to this reference point are considered the most meaningful for the model’s prediction. If an unsuitable saliency method is chosen, it can lead to incorrect or misleading interpretations of the model’s behaviour and predictions. Therefore, selecting a saliency method appropriate for the specific task requirements is crucial to avoid these consequences.

Saliency cards can assist users in avoiding choosing “the wrong method” by reducing the operational details of a saliency method into ten user-centric attributes. The attributes encompass the methodology for calculating saliency, the connection between the saliency method and the model, and how users interpret the outputs generated by the method.

The saliency cards can also serve as a valuable resource for scientists by revealing areas where further research is needed. For instance, the researchers from MIT encountered a challenge in finding a saliency method that was both computationally efficient and applicable to any machine-learning model. This highlights a gap in the research space that warrants further exploration and development.

In the future, the researchers aim to delve into the less-explored attributes of saliency methods and potentially create task-specific saliency techniques. They also seek to enhance their understanding of how individuals perceive saliency method outputs, with the potential for developing improved visualisations. Furthermore, they have made their work accessible through a public repository, inviting feedback from others that will contribute to future advancements.

Boggust is optimistic, envisioning these saliency cards as dynamic documents that will evolve as new saliency methods and evaluations emerge. Ultimately, this marks just the beginning of a broader discussion regarding the attributes of saliency methods and their relevance to different tasks. Boggust believes that in the future, there will be other researchers who will further develop this discovery.