Artificial intelligence has made significant strides in inclusivity, as evidenced by Jess Smith’s experience where ChatGPT now accurately generates images of her with one arm, highlighting progress in reducing bias in AI systems.
Jess Smith, a former Australian Paralympic swimmer who is missing her left arm below the elbow, initially struggled to get AI image generators like ChatGPT to create a realistic depiction of her. When she uploaded a photo this summer and specified her disability, the tool repeatedly produced images with two arms or inaccurate prosthetics, failing to represent her accurately.
Upon questioning the AI, it responded that it lacked sufficient data, leading Smith to realize that AI mirrors societal inequalities. This insight underscored the importance of diverse data sets in technology development to avoid perpetuating discrimination.
Recently, Smith tried again and was astonished to find that ChatGPT could now generate a correct image of a woman with one arm. She described the moment as groundbreaking, emphasizing that such advancements are crucial for representation and inclusion in the digital age.
A spokesperson for OpenAI confirmed that recent improvements to their image generation model have addressed some biases, though challenges remain. They are actively working on refining methods and incorporating more diverse examples to enhance fairness over time.
However, not all cases are resolved; Naomi Bowman, who has sight in only one eye, shared that ChatGPT altered her face to “even out” her eyes despite her instructions, revealing persistent biases. She advocates for rigorous testing and broader data sets to ensure equitable representation.
Experts like Abran Maldonado of Create Labs stress that diversity in AI starts with inclusive teams during data training and labeling. Without input from people with lived experiences, AI systems risk overlooking marginalized groups, as seen in past issues like facial recognition inaccuracies.
Smith believes that societal barriers, not her disability, limit her, and similar oversights could occur in AI if not designed inclusively. She hopes that continued improvements will lead to more empathetic and accurate technologies that benefit everyone.
This progress in AI image generation represents a step forward in tech ethics, signaling a move towards systems that recognize and respect human diversity. As AI evolves, embedding inclusion at its core is essential for creating a fairer digital world.