It always makes me smile when someone talks about “artificial superintelligence behind the corner“, ready to surpass us in everything, then I read a recent study and discover that the most advanced AI systems on the planet can't even tell me what day it is, and what time the clock says. Isn't that ironic? These digital giants that generate photorealistic images, win at poker and complete our sentences before we finish them, are miserably lost in front of simple calendars.
It’s like watching a quantum physicist who can’t tie his shoelaces: brilliant in the lab, but embarrassing in the little things of everyday life. Calendars, objects that our ancestors mastered millennia ago, still represent an enigma to these supposedly “wannabe superior” intelligences. But why?
The strange temporal inability of machines
Have you ever wondered why a six-year-old child can read an analog clock (if we teach him, of course) while a computational system as powerful as a supercomputer gets confused? According to a study conducted by the University of Edinburgh, the most advanced AIs are able to correctly interpret the position of the hands less than 25% of the time. A frankly embarrassing result.
The issue becomes even more complicated when the hands are stylized or Roman numerals replace Arabic ones. And to think that we humans solved this problem millennia ago with sundials; today we even have wristwatches that measure the phases of the moon and the heart rate, yet artificial intelligence seems stuck in the Time Stone Age.
Even the presence or absence of the second hand doesn’t change the AI’s performance, suggesting that the problem is deeper than it seems. It’s not just a matter of visual recognition, but of spatial and contextual interpretation. And it doesn't stop there.
Calendars: A Puzzle for Digital Giants
What I find particularly fascinating is how calendars present an insurmountable obstacle for these artificial intelligences. Researchers have found that even the best models They miscalculate dates 20% of the time. Imagine relying on an AI to schedule important appointments with this margin of error: you risk missing them one in five times.
The tests conducted by the team led by Rohit Saxena of School of Informatics have highlighted how these machines fail miserably at tasks that a primary school child can easily perform. Identifying holidays or calculating future and past dates seem like abstruse concepts for systems that excel at generating complex text.
The Aryo Gema, another researcher involved in the study, pointed out the irony of the situation: systems designed for complex reasoning that get bogged down in everyday tasks. It's like having Nobel Prize winners (literally) who don't know how to install a light bulb.
The Implication for Our Digital Future
Imagine a world where home automation systems can't manage timetables properly; virtual assistants that wake you up at 3 am convinced that it's time to go to work; robots that arrive at appointments a day late. It sounds like the plot of a comedy, but it's the real state of the art.
These seemingly trivial problems have enormous implications for integrating AI into time-sensitive applications. Think of scheduling assistants, autonomous robots, or tools for the visually impaired—all of them depend on the correct interpretation of times and dates.
The study, which will be presented at the “Reasoning and Planning for Large Language Models” workshop in Singapore on April 28, 2025, highlights a significant gap in machine capabilities. And if we really want AI to become an integral part of our daily lives, it’s time for it to learn to read the clock.
Ultimately, however brilliant they may be in other fields, these systems demonstrate how long the road is still to an intelligence truly comparable to ours. Perhaps it is precisely in these small daily inabilities that the deepest difference between man and machine lies. And for once, I am almost pleased that the hands of time are still a mystery to them.
