AI can write essays, create art, and even hold conversations, but when it comes to telling time or reading a calendar, it still has a lot to learn.
A new study from the University of Edinburgh found that some of the most advanced AI systems struggle to read clocks and understand dates, tasks that most humans can do easily.
Researchers tested AI models that process both text and images—known as multimodal large language models (MLLMs)—to see if they could answer time-related questions by looking at pictures of clocks and calendars.
The results were surprising: AI got the correct time from a clock less than 25% of the time, and it also made frequent mistakes when answering questions about dates.
The team tested different clock designs, including those with Roman numerals, different colored dials, and clocks with and without second hands.
The AI performed especially poorly when the clocks had stylized hands or Roman numerals.
Even removing the second hand did not help, suggesting that AI struggles with the basic task of detecting and interpreting clock hands.
When asked calendar-related questions, AI models also made errors about one-fifth of the time. These included mistakes in recognizing holidays, calculating past or future dates, and working out simple scheduling problems.
The problem isn’t just about recognizing numbers or shapes—telling time and using calendars requires spatial awareness, context, and basic math.
AI systems are good at some types of reasoning, but they still lack the ability to fully understand how time works in a practical way.
The researchers say fixing these weaknesses is important for time-sensitive AI applications, such as scheduling assistants, automation in smart homes, robots that need to follow time-based instructions, and tools for visually impaired people.
The findings will be presented at a major AI conference (ICLR) in Singapore on April 28, 2025.
Rohit Saxena, the lead researcher, explained, “Most people learn to tell the time and use calendars at a young age. Our study shows that AI still struggles with these basic skills. If AI is going to be useful in scheduling, automation, or assistive technology, these issues need to be fixed.”
Aryo Gema, another researcher on the team, added, “AI research focuses a lot on complex reasoning, but ironically, many AI systems still can’t handle simple everyday tasks. If we don’t address these gaps, AI might remain unreliable for real-world use—stuck at the eleventh hour.”
This study highlights a surprising flaw in today’s AI and reminds us that sometimes, the simplest tasks are the hardest to master.
