1947: The first computer bug is discovered

On 9 September 1947, operators of the Harvard Mark II electromechanical computer at Harvard University encountered a failure. When they opened a relay, they found a moth trapped inside. The insect was carefully removed and taped into the machine’s logbook with the now-famous note: “First actual case of bug being found.” The entry, associated with Grace Hopper’s team, gave lasting currency to the word “bug” for a hardware or software fault and to “debugging” for the process of removing it.

The Mark II itself was a large-scale calculator built for the U.S. Navy, capable of handling ballistic and other numerical problems. It belonged to the generation of machines that sat between pure mechanical calculators and the fully electronic stored-program computers that would follow. Hopper, already a pioneering programmer, later developed the first compiler and championed languages that let people instruct machines in something closer to English—work that directly shaped the software practices still used today.

The moth incident is more than an amusing anecdote. It marks the moment when the reliability of complex automatic machinery became a visible, shared concern. Early computer scientists quickly realized that errors could arise from physical components, from coding mistakes, or from flawed logical design. That threefold awareness fed into the emerging fields of cybernetics, automata theory and, eventually, artificial intelligence, where the question of how a system can detect and correct its own mistakes remains central.

In modern AI the same vocabulary and the same underlying challenge persist. Training runs are “debugged,” model outputs are checked for hallucinations, and entire research programs focus on robustness and interpretability. When an large language model produces an unexpected result, engineers still speak of hunting for the bug—whether it lies in data, in architecture, or in the interaction between them. The moth taped into a 1947 logbook thus sits at the start of a continuous thread that runs from electromechanical relays to neural networks.

Seventy-eight years later, the Harvard logbook page is preserved as a landmark of computing history. It reminds us that every intelligent system, no matter how abstract, ultimately rests on physical devices that can fail and on human practices devised to keep them honest.