How Good Is ChatGPT at Recognising Human Faces?
For the past year, ChatGPT has expanded its capabilities to analyze not just text but also images with the introduction of its latest version - GPT-4V(ision). This new feature allows ChatGPT to process images and provide relevant information based on their content.
For example, by uploading a picture of your refrigerator, ChatGPT can identify the items in the photo and suggest meal ideas along with recipes. Similarly, a hand-drawn sketch of a website design can be converted into HTML code by ChatGPT. Additionally, providing a still frame from a movie enables ChatGPT to recognize the film and summarize the plot up to that point.
One intriguing aspect of ChatGPT's image analysis capability is its handling of human faces - specifically, matching different images of the same person. To evaluate ChatGPT's proficiency in recognizing faces, various tests commonly used in psychology to assess facial recognition abilities were employed.
Testing ChatGPT's Facial Recognition Abilities
Firstly, the "reading the mind in the eyes" test was conducted. In this test, participants view only the eye regions of individuals in photographs and select the appropriate descriptive word that corresponds to the person's thoughts or feelings. ChatGPT performed impressively by successfully answering 29 out of 36 questions on this test.
Moving beyond facial expressions, the "Glasgow face matching test" was administered. In this test, participants are presented with pairs of face images where they must determine if the two images portray the same person or different individuals. ChatGPT excelled on this test with a score of 92.5%, surpassing the average participant score of 81.3%.
Lastly, face recognition was explored using the "famous faces doppelgangers test". Here, participants must identify a celebrity's face from a pair of similar-looking faces. ChatGPT achieved a perfect score of 100% across all trials, outperforming typical participant scores of 81.5%.
Implications and Future Prospects
Based on these evaluations, ChatGPT demonstrates proficiency in tasks related to human face recognition and identification, showcasing its capabilities in this domain. Utilizing a large language model (LLM) AI framework, ChatGPT has been trained on extensive text and image datasets, allowing it to generate informed responses to a wide range of queries.
While ChatGPT does not store specific images, its training data includes a significant amount of facial images, enabling it to recognize patterns and associations related to human faces. Through continual refinement, ChatGPT's performance is expected to further enhance with each subsequent version release, solidifying its position as a versatile and adept AI tool in various applications.
