Artificial intelligence can do many things that once seemed impossible. It can generate hyper-realistic war zones, deepfake politicians, and even manipulate historical narratives. But ask AI to generate an image of a person writing with their left hand, and many models fail—often spectacularly.
This strange limitation exposes something deeper about AI systems and the way they learn. While AI has become disturbingly effective at generating propaganda, misinformation, and biased outputs, it still struggles with minor but meaningful details. The inability to accurately depict left-handed people is not just a glitch—it is a symptom of how AI is trained, the data it relies on, and the biases it reinforces.
Why AI Struggles with Left-Handed People
Most AI models learn from vast datasets compiled from existing images and text. Since only 10-15% of the global population is left-handed, training data is inherently skewed toward right-handed imagery. When an AI model generates an image of a person writing, it defaults to the majority pattern—a right-handed grip—because that is what it has seen the most.
This issue was highlighted at the AI Action Summit in Paris, where Indian Prime Minister Narendra Modi made an intriguing observation:
“AI can interpret complex medical reports, but it struggles to show a person writing with their left hand.”
Following his remarks, fact-checkers tested AI tools like MetaAI and ChatGPT to generate accurate left-handed images. The results were clear—many still failed to depict left-handed individuals correctly, producing images where the pen or hand position was awkward, unnatural, or outright incorrect.
The Bigger Question: How Can AI Be So Good at Manipulation But Bad at Small Details?
If AI can fabricate realistic deepfakes, generate fake political conspiracies, and create hate-filled content, why does it struggle with something as simple as handedness? The answer lies in how AI models prioritize data processing.
AI does not “see” the world the way humans do. It processes patterns, probabilities, and dominant trends in data. This means that the more common something is, the better AI gets at generating it. Right-handed people dominate image datasets, so AI models naturally become better at depicting them.
But this same mechanism also means AI can scale bias and misinformation at an unprecedented level.
AI Doesn’t Just Reflect Bias—It Reinforces It
Bias in AI is not limited to image generation. It affects hiring, policing, finance, and news. A 2023 study found that AI-generated job applications favored male candidates over female ones—even when the credentials were identical. A biased dataset led to biased outputs, and because AI works at scale, it doesn’t just reflect human biases—it amplifies them.
The same pattern is seen in predictive policing, where AI-driven crime prediction tools disproportionately target certain communities, reinforcing racial and socioeconomic biases. Financial algorithms have denied loans to minorities at higher rates, even when applicants had similar financial profiles to approved candidates. AI models learn from historical data, and if that data contains discrimination, the models codify it into decision-making systems.
The Real Risk: AI’s Bias Becomes Reality
AI is no longer just a tool—it is shaping who gets hired, who gets financial access, what news people consume, and even how history is remembered. When bias exists in the underlying data, AI systems can systematically perpetuate inequality in ways that are difficult to detect and even harder to correct.
If left unregulated, these systems will not only reflect society’s flaws but embed them into the future, creating an algorithmic loop of inequality. The inability to correctly depict a left-handed person may seem trivial, but it points to a much deeper issue: AI models are trained on what is common, not what is correct.
