• How does AI help us recognise emotions?

AI can significantly enhance our ability to recognise emotions. How? By leveraging advanced machine learning algorithms and affective computing techniques that create sophisticated mappings between observable affect measurements and corresponding emotion labels.

There are measurements such as facial expressions, vocal tones, body language, and physiological signals like heart rate or skin conductance, that are captured and processed to create affect datasets.

Machine learning models are then trained on those datasets, deriving complex relationships between the inputs (like a smile or raised voice) and the corresponding emotional states (such as happiness or anger).

This process of mapping affect measurements to emotion labels via machine learning is central to emotion recognition, enabling AI to detect and interpret subtle emotional cues that humans might overlook or misunderstand, thereby offering more profound insights into human emotions.

 

• There are challenges hindering the advancement of AI in this regard. What are they?

The complexity of the relationship between AI and human emotions presents one of these challenges. The context-dependent nature tied to specific settings and environments of current affect models are a significant hurdle.

Traditional models often rely on large datasets that assume statistical correlations can sufficiently capture emotional states, but these models fail to account for the causal and anti-causal relationships that influence emotions.

AI struggles to answer deeper questions such as "Why do these emotions appear?" or "What elements of the context trigger specific emotions?"

 

• So how is Malta leading the way in resolving this?

Through the ERICA Project.

The ERICA Project, led by Dr Konstantinos Makantasis, a lecturer at the Department of AI within the Faculty of ICT at the University of Malta, is attempting to revolutionise the field of affective computing by addressing the limitations of current emotion recognition models, which are context-dependent and unable to identify the causal mechanisms underlying human emotional responses.

ERICA proposes shifting from traditional statistical learning approaches to a causation-aware paradigm, allowing AI to answer not just "what" emotions are present but "why" these emotions arise.

The project focuses on two key objectives:

1. first, to discover and identify high-level predictors that are causally or anti-causally related to emotions, thereby enabling the development of context-independent models; and second,
2. to create modular models of affect that can generalise across different environments by reusing robust cause-effect relationships.

By integrating causal inference tools and emphasising modularity, ERICA proposes an innovative approach towards building more adaptable and accurate emotion-aware AI systems with far-reaching applications in psychology, mental health, education, marketing and robotics. 

Ultimately, ERICA's vision is to foster a deeper connection between humans and technology, where AI understands and responds to humans’ emotions in ways that genuinely improve our lives.