Imagine a world where technology simply doesn't "see" you. For people with disabilities, this has been a frustrating reality with AI image generators – until recently. One woman's experience highlights both the progress and the persistent challenges in achieving true inclusivity in artificial intelligence.
Yasmin Rufo, reporting for BBC News, shares the story of Jessica Smith, a former Australian Paralympic swimmer. Smith embarked on a seemingly simple task: to generate a professional-looking headshot using an AI image generator. She uploaded a full-length photo of herself and specifically instructed the AI to depict her missing left arm from below the elbow. Surprisingly, the AI, in this case ChatGPT, struggled immensely. Despite multiple attempts and detailed prompts, the results were consistently inaccurate, showing either a woman with two arms or one with a robotic prosthetic.
When Smith inquired why the AI was struggling, it responded that it lacked sufficient data to accurately represent her condition. "That was an important realization for me that of course AI is a reflection of the world we live in today and the level of inequality and discrimination that exists," she says. This highlighted a critical issue: AI, in its current form, often mirrors the biases and underrepresentation present in the data it's trained on. But here's where it gets controversial... Is it the AI's "fault" that it reflects societal biases, or is it our responsibility to ensure the data used to train these systems is truly representative?
Recently, Smith decided to try again. To her amazement, ChatGPT successfully generated an accurate image of a woman with one arm. "Oh my goodness, it worked, it's amazing it's finally been updated," she told the BBC. "This is a great step forward." While it might seem like a small victory, this represents a significant milestone for millions of people with disabilities.
"Representation in technology means being seen not as an afterthought, but as part of the world that's being built," Smith emphasizes. "AI is evolving, and when it evolves with inclusion at its core, we all benefit. This is more than progress in tech it's progress in humanity." OpenAI, the company behind ChatGPT, acknowledged the issue and stated that they had recently "made meaningful improvements" to their image generation model. They also admitted that "challenges remain, particularly around fair representation, and we're actively working to improve this - including refining our post-training methods and adding more diverse examples to help reduce bias over time."
However, the journey toward inclusive AI is far from over. Naomi Bowman, who has sight in only one eye, experienced a similar problem. When she asked ChatGPT to blur the background of a picture, the AI not only failed to do so accurately but also "changed my face completely and evened out my eyes." And this is the part most people miss... even when Bowman explicitly explained her eye condition and requested that her face remain untouched, the AI struggled to comprehend. Initially, Bowman found it humorous, but she later expressed sadness, recognizing the "inherent bias within AI." She advocates for rigorous training and testing of AI models to reduce bias and ensure that datasets are broad enough to represent and treat everyone fairly.
It's important to note that the creation of images by AI also comes with an environmental cost. Professor Gina Neff of Queen Mary University London points out that ChatGPT "is burning through energy," and the data centers powering these AI systems consume more electricity annually than 117 countries combined. This raises important ethical questions about the trade-offs between technological advancement and environmental sustainability. Do you think the benefits of AI outweigh the environmental costs?
Bias in AI often mirrors the blind spots prevalent in society. Abran Maldonado, CEO of Create Labs, emphasizes that diversity in AI begins with the people involved in training and labeling the data. "It's about who's in the room when the data is being built," he explains. "You need cultural representation at the creation stage." If the people creating and training AI systems don't have diverse perspectives and lived experiences, AI will inevitably perpetuate existing inequalities. One stark example is a 2019 US government study that revealed facial recognition algorithms were significantly less accurate at identifying African-American and Asian faces compared to Caucasian faces.
Despite living with one arm, Jessica Smith doesn't identify as disabled. She believes the barriers she faces are primarily societal. "If I use a public toilet and the tap has to be held down, that impacts my ability, not because I can't do it, but because the designer hasn't thought about me." She warns that the same oversight could easily occur in the world of AI, leading to systems and spaces built without considering everyone. When Smith shared her initial experience on LinkedIn, someone messaged her claiming his AI app could create an image of a woman with one arm. However, when she tried it, she encountered the same problem. She informed the person, but he never responded, a reaction she describes as typical of conversations surrounding disability. "The conversation is too awkward and uncomfortable so people back away."
This highlights a crucial point: true inclusivity requires more than just technical solutions. It demands open and honest conversations, a willingness to confront uncomfortable truths, and a commitment to building AI systems that reflect the diversity of the human experience. What steps do you think are most important to ensure AI becomes truly inclusive and representative of all people? Share your thoughts in the comments below!
