Can AI Understand Emotions?
Today, venting at length to a chatbot on your phone or feeling relieved because a virtual assistant ‘listens’ to your complaints about a heavy day is no longer strange. From customer service features that can detect frustration in a customer’s voice to cars that warn us when a face appears drowsy, artificial intelligence is now present in the most personal spaces of our lives: our feelings.
However, the question remains hanging: does the machine truly understand our emotions, or is it merely skilled at pretending?
When Machines Learn to Read Our Faces and Voices
The idea that computers could ‘recognising’ human feelings has long been proposed. Rosalind Picard, the scientist who pioneered the field of affective computing, explained that this effort is essentially an attempt to give computers emotional capabilities, ranging from recognising to responding to human expressions via sensors, cameras, and machine learning models.
This is the concept we experience daily: mental health applications that analyse speech patterns, or AI-based customer services that can ‘catch’ a caller’s irritation from their intonation and then redirect the call to a human.
Interestingly, despite being a pioneer in this field, Picard herself warns that even as artificial intelligence develops rapidly through deep learning and large language models, machines fundamentally do not possess feelings, do not think, and do not truly learn in a human sense. The issue lies in our language, which often errs in describing what is actually happening inside the machine. At this point, we begin to realise: the ability to detect is not the same as the ability to understand.
Between Algorithms and Complex Reality
Our daily reality is far more complex than what a camera or microphone can capture. Imagine someone smiling during a job interview because they are nervous, rather than happy, or a child appearing ‘gloomy’ on a screen during online learning when they are simply tired.
Researcher Kate Crawford, who has criticised the AI emotion recognition industry, highlights that implementing such technology in high-risk contexts—such as recruitment, law enforcement, and education—has the potential to perpetuate bias and produce erroneous conclusions. She asserts that the idea that a person’s face can accurately reveal what they are truly feeling is a fundamentally flawed assumption, as emotional expressions are heavily influenced by culture, context, and individual habits.
In real life, this means AI systems that assess employee ‘mood’ via work cameras, or applications that score a candidate’s honesty based on facial expressions, could misjudge individuals simply because their way of expressing themselves differs from the data used to train the algorithm.
Feeling Understood, But By Whom?
Ironically, even though AI does not technically ‘feel’, many people feel more understood by chatbots than by the humans around them. The phenomenon of confiding in AI, chatting with virtual assistants at night, or using digital companion apps as daily ‘friends’ is becoming increasingly common.
Yoshua Bengio, a major figure behind the progress of deep learning, explains that what makes people feel close to AI is not whether the machine is truly conscious or aware, but because the interaction feels like speaking to an intelligent entity with its own personality and purpose. This is why more people are becoming emotionally attached to the AI they use.
He also warns that this subjective perception of machine consciousness could lead people to make incorrect decisions, as we tend to project human understanding and empathy onto something that actually operates solely through statistical patterns. Here lies the deeper issue: it is not just about whether AI understands us, but how we, as humans, so easily feel understood by something that is actually just arranging word after word based on probability.
Re-evaluating What It Means to ‘Understand’
From these perspectives, a common thread emerges: AI can indeed detect patterns, mimic empathetic responses, and even make us feel heard, but understanding emotion in its full sense—involving life experience, pain, happiness, and complex cultural contexts—appears to remain beyond the reach of machines, at least for now.
In daily life, this is not a reason to avoid this technology, as its benefits are real: assisting in monitoring mental health, facilitating public services, or simply providing companionship during loneliness. However, it is wise to remain aware of its limits; behind those seemingly warm responses, there is no one who truly feels what we feel. Perhaps the more important question is not ‘can AI understand our emotions’, but ‘how prepared are we to maintain a healthy distance while still utilising its presence wisely’.