AI faces mathematical accuracy challenges and societal risks

Research from the University of Chicago and the KTH Royal Institute of Technology indicates that current artificial intelligence models struggle significantly with basic arithmetic. Tests showed that some models failed to multiply four-digit numbers 99% of the time, often guessing the approximate shape of an answer rather than performing precise calculations. Anthropic reported that its Claude model achieved approximately 21% accuracy on raw tables, though this improved to 95% when grounded in a governed semantic layer.

This mathematical limitation has business implications, particularly regarding financial metrics like revenue and inventory where accuracy is essential. Companies like AtScale are positioning themselves to provide the necessary semantic layers to prevent AI from guessing business data.

Separately, Bill Gates has expressed concerns regarding the societal transition into the AI era. He noted that while AI has the potential to be a great equalizer, it also poses risks of increasing injustice and widening the gap between the rich and the poor if not managed correctly. He emphasized that the world is not yet adequately prepared for the turbulence this technological shift may cause.