Google continues to push the boundaries of artificial intelligence with PaliGemma 2, a new family of AI models that can analyse images and describe their content with unprecedented detail. The innovation that has ignited debate, however, is the ability of these models to identify the emotions of the people portrayed in the photos. PaliGemma 2 does more than just recognise objects or faces. Using sophisticated machine learning algorithms, the AI can generate accurate and contextualised descriptions of images, including information about actions, relationships between elements and, indeed, emotions. Google says this capability can be useful in several areas, such as moderating online content, providing accessibility for the blind, and creating more engaging interactive experiences.
But is it really possible to “read” emotions?
PaliGemma 2’s ability to identify emotions has raised a hornet’s nest of criticism and concern from ethicists and researchers. Central to the debate is the complexity of human emotions, which can hardly be reduced to simple facial expressions or postures. According to Mike Cook, a Queen Mary University researcher specialising in AI, emotions are subjective and multifactorial experiences influenced by context, culture and personality.
“We think we can understand what others are feeling by looking at them, but it’s an illusion,” Cook says. Heidy Khlaaf, a scientist at the AI Now Institute, warns against the risk of bias and discrimination. “The interpretation of emotions is influenced by cultural and personal factors,” Khlaaf explains. “AI trained on limited data could perpetuate stereotypes and biases, with serious consequences for people.” Sandra Wachter, professor of data ethics at the Oxford Internet Institute, paves the way toward a “dystopian future” in which emotion analysis could be used to discriminate against people in employment, finance or education.
An uncertain future
The arrival of PaliGemma 2 and its ability to “read” emotions opens a new and complex chapter in the history of artificial intelligence. On the one hand, the potential of this technology is enormous, but on the other hand, the ethical and social risks are equally significant.
It is crucial that Google and other industry players engage in an open and transparent dialogue with civil society to ensure that AI is developed and used responsibly, respecting human rights and dignity. Only in this way will we be able to avoid the dystopian scenarios feared by experts and take full advantage of the positive potential of artificial intelligence. In the face of this criticism, Google entrenched itself behind a “no comment,” generically reiterating its commitment to ethical and responsible artificial intelligence. An answer that smacks of something already heard and does not reassure much.
Mind you, the potential of PaliGemma 2 and emotion recognition models is undeniable and fascinating. But it is equally true that we are dealing with a disruptive and, in some ways, controversial technology that raises nontrivial ethical and social questions. The scientific community and regulators will be called upon to strike the right balance between innovation and rights protection because it is one thing to use AI to automatically tag our vacation photos and quite another to entrust it with the task of interpreting our most intimate and private emotional sphere. Not to mention the risks of mass surveillance and profiling.
