Google recently launched its next-generation artificial intelligence model, PaliGemma2, which can analyze images and generate captions while answering questions about the emotions and actions of people in the photos. PaliGemma2 is based on Google's Gemma open model series, offering deeper image descriptions than traditional object recognition, capable of identifying emotions and generating contextually relevant detailed descriptions. However, despite this technology seeming like a groundbreaking innovation, experts have raised serious warnings about its potential ethical and social impacts.
Emotion recognition is not a standard feature of PaliGemma2 but is achieved through fine-tuning. Although Google claims it has undergone "extensive testing" and performs better than industry benchmarks in terms of demographic bias, experts remain concerned about the technology's reliability. Professor Sandra Wachter from the University of Oxford states, "There are significant issues with using AI to 'read' human emotions," noting that this process relies heavily on assumptions, which may lead to misinterpretations and biases.
Emotion recognition technology has long been a controversial topic in the tech community. While early research, such as Paul Ekman's theory of emotions, proposed six basic emotions, subsequent studies have shown that emotional expressions can vary greatly across different cultures and backgrounds. Researcher Mike Cook from Queen Mary University of London points out, "The complexity of emotional experiences makes accurate emotion detection nearly impossible." Furthermore, studies have indicated that existing facial expression analysis systems often exhibit biases towards certain emotions, such as smiles or differences in facial expressions across different races.
As emotion recognition technology becomes increasingly commercialized, the potential risks of misuse have raised concerns among various parties. Some experts worry that such technologies could be used in law enforcement, recruitment, and other areas, further exacerbating social inequalities. The EU's Artificial Intelligence Act has already proposed strict limitations on the use of emotion recognition technology, especially in high-risk environments.
Google insists that PaliGemma2 has thoroughly considered ethical and safety issues during its testing phase, particularly regarding child safety and content security. However, whether these assurances are sufficient still requires rigorous scrutiny. Dr. Heidy Khlaaf from the AI Now Institute emphasizes that emotion recognition is not just a visual issue but involves deep social and cultural contexts, stating, "Emotions cannot be accurately inferred solely from facial features."
With the public release of this technology, PaliGemma2 will not only advance the application of artificial intelligence in the field of image understanding but also pose new challenges to social ethics and data privacy, necessitating the attention and intervention of relevant regulatory bodies.
