Indian PM Modi Exposes AI Bias: The Left-Handed Writing Challenge

Indian Prime Minister Says AI Can't Generate Images of Left-Handed People Writing

Indian Prime Minister Narendra Modi pointed out that AI image generators are unable to make pictures of left-handed people while speaking at the Artificial Intelligence Action Summit in front of world leaders at the Grand Palais in Paris, France. Modi used the example to highlight AI's shortcomings, noting that text-to-image models cannot generate pictures of a person writing with their left hand, which sparked a wave of people testing it out for themselves. "If you upload a medical report to an AI app it can explain in simple language, free of any jargon, what it means for your health," Modi explained during the summit's opening remarks yesterday. "But if you ask the same app to draw an image of someone writing with their left hand, the app will most likely draw someone writing with their right hand." The Prime Minister adds that AI's inability to create a picture of a leftie is due to what's in the training data. He uses the anecdote to warn of the "many vices we need to think carefully about" when it comes to AI. PetaPixel tested Modi's hypothesis and found that he is right: AI image generators will not make pictures of left-handed people. Despite being explicitly asked to "make a picture of someone writing with their left hand," ChatGPT's AI image generator, DALL-E, produced an almost identical image to the one in the X video above. Stable Diffusion 3.5 received the same prompt and generated a bizarre, top-down image showing a right hand and a multi-dimensional pen drawing a sketch. While X's Grok created four pictures of people all writing with their right hands. Modi is correct when he says it is down to the training data. According to Healthline, roughly 10 percent of the population is left-handed and that means most photos of people writing show the pen in their right hand. In a Google Image search for "people writing", not a single leftie appears and it is the same when searching Shutterstock, one of the world's largest stock photo websites. Why is this relevant? Because AI image generators are known to have been built via huge scrapes of the open web and so if there are none or very few images of people writing with their left hand then the AI will only know how to recreate an image of a right-handed person. It is a similar problem to AI image generators struggling to generate accurate pictures of human hands. This is just one of the many pitfalls of the technology which is extremely controversial since, in most cases, the copyright holders weren't asked before their photos were used for machine learning.

AI Action Summit 2025: PM Modi highlights AI bias with a brilliant example

Artificial Intelligence (AI) is shaping the world in ways we never imagined. From healthcare to creative arts, AI models are transforming industries with their capabilities. However, a fundamental problem with AI systems remains - bias in training data. This issue was brought to global attention when Indian Prime Minister Narendra Modi addressed the AI Action Summit 2025 in France. PM Modi's speech highlighted an important but often overlooked concern: AI's inherent bias due to skewed training data. The issue of AI bias isn't just about images; it extends to medical diagnoses, hiring processes, and more. In this article, we explore what PM Modi said, how AI models responded to the same experiment, and why fixing AI bias is crucial. At the AI Action Summit in France, PM Modi took the stage to talk about AI's potential and its pitfalls. His primary focus was on AI bias - a problem that can have real-world consequences if left unaddressed. To illustrate this, he proposed a simple experiment - asking an AI model to generate an image of a person writing with their left hand. According to him, the AI would mostly generate images of right-handed individuals, demonstrating how biases in training data lead to incorrect or skewed results. His speech has since received praise for its clear articulation of AI-related concerns and its emphasis on the need for unbiased AI systems that cater to all users equally. We conducted the same experiment across multiple AI models, including Grok, Midjourney, and DALL·E. The results were consistent with what he had stated: nearly all images depicted individuals writing with their right hand, even when explicitly asked to generate left-handed individuals. Also Read: AI Action Summit 2025 in Paris: 5 things you need to know We took it a step further by specifying well-known left-handed personalities, such as Jimmy Hendrix and Sachin Tendulkar. Despite clear photographic evidence of both Sachin and Hendirx being left-handed, AI models continued to generate images of then using their right hand. This highlighted a strong bias in AI training data, favoring right-handed images over left-handed ones. Here are the results for you to have a look -- AI bias is more than just an inconvenience - it can have far-reaching consequences, impacting various sectors and amplifying societal inequalities. Recent research highlights how bias manifests in different AI applications and why it must be addressed urgently. A research paper titled "Unmasking Bias in Artificial Intelligence: A Systematic Review of Bias Detection and Mitigation Strategies in Electronic Health Record-Based Models" highlights how AI models trained primarily on Western medical data can misdiagnose or underdiagnose patients from other regions. This issue extends beyond incorrect diagnoses; it leads to a lack of personalized treatment recommendations, potentially putting lives at risk. Without addressing this bias, AI-driven healthcare solutions may continue to fail patients from underrepresented demographics. Also Read: Safer Internet Day 2025: India's AI evolution and cybersecurity landscape Beyond healthcare, AI-generated content exhibits biases in visual representation. A study, "Bias in Generative AI," found that AI-generated images systematically favor specific racial and gender depictions, often reflecting outdated stereotypes. When AI models repeatedly generate images that reinforce biases - such as associating leadership roles with men or scientists with a particular race - they influence perceptions in subtle but powerful ways. These biases in AI can shape societal norms, perpetuating discrimination rather than challenging it. One of the most concerning aspects of AI bias is its self-reinforcing nature. The paper "AI AI Bias: Large Language Models Favor Their Own Generated Content" underscores how AI models trained on biased data will continue to generate biased content, which in turn influences future training datasets. This loop creates a compounding effect where AI systems increasingly favor biased outputs over time. If left unchecked, AI will not just reflect existing biases but actively amplify them, leading to greater inequalities in decision-making and representation. Also Read: AIoT revolution: The future of smart homes with next-gen motion recognition The best way to mitigate AI bias is through diverse, representative training datasets. A study titled "Policy Advice and Best Practices on Bias and Fairness in AI" emphasizes the importance of ensuring AI is trained on inclusive data that represents different genders, ethnicities, and socioeconomic backgrounds. This is particularly relevant for India, where demographic diversity is vast. Relying solely on AI models trained on Western data will continue to yield inaccurate or exclusionary results. Instead, India must invest in developing its own AI models built on indigenous data that better reflect its unique cultural and social diversity. PM Modi's emphasis on developing locally relevant AI aligns with these research findings. By training AI on diverse, high-quality datasets, we can create more accurate, fair, and useful AI applications that serve the needs of all individuals, not just a select few. Our independent testing across multiple AI models confirmed his claim, proving that AI models overwhelmingly generate right-handed individuals, even when asked to depict left-handed ones. Research findings further support the idea that AI bias extends far beyond image generation, affecting healthcare, content generation, and decision-making systems.