At the AI Action Summit in Paris, Prime Minister Narendra Modi drew attention to an unexpected flaw in artificial intelligence. Beginning his address with a simple experiment, he said, “If you upload your medical report to an AI app, it can explain in simple language, free of any jargon, what it means for your health.” But then came the surprising part.  

Advertisment

PM Modi pointed out, “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 reason? AI is trained on datasets dominated by right-handed images.  

Also Read | 'Need to be careful about AI biases', PM Modi addresses AI Summit in Paris

His statement highlights a deeper issue— despite AI’s rapid progress, it still struggles with certain human features, especially hands.  

Advertisment

Also Read | WATCH | Did Canada's Trudeau smirk at PM Modi during his speech at AI Summit in Paris?

PM Modi’s observation underscores a critical aspect of AI development: bias in training data. If AI tools are learning from incomplete or skewed datasets, they will continue to produce flawed outputs.  

As he emphasised, “While the positive potential of AI is absolutely amazing, there are many biases that we need to think carefully about.” 

Advertisment

Why AI fails to get left-handed writing right

As per Abp news, AI-generated images have long faced issues with accuracy in human anatomy. Hands, in particular, remain a challenge. The reason lies in the quality and diversity of training data. Since right-handed writing is more common in datasets, AI models struggle to recognise and replicate left-handed writing correctly.  

Take a look:

 

 

 

Also Read | 'Need for rules, framework of trust' for AI development, says French PM Macron at AI Summit

Even advanced AI tools often produce images with distorted or incorrect hand positioning. The problem isn’t limited to hands—AI also struggles with details like ears and teeth, which frequently appear misshapen in generated images.  


(With inputs from agencies)