The jet that ran out of fuel at 41,000 feet

On 23 July 1983, Air Canada Flight 143 was flying a Boeing 767 from Montreal to Edmonton, stopping in Ottawa. There were 61 passengers, two pilots, Captain Robert Pearson and First Officer Maurice Quintal, and six cabin crew.

The 767 was Air Canada's first type with fuel gauges reading in kilograms. The rest of the fleet used pounds. On this plane a faulty part had left the fuel gauges blank, and under the airline's rules it should not have been sent out like that. Through a chain of misunderstandings, it was.

With no gauges, the crew and ground staff measured the fuel with a dipstick and converted the reading by hand. They used the factor 1.77, which turns litres into pounds, when they needed about 0.8 to turn litres into kilograms. The plane left Montreal, and then Ottawa after the same check was repeated with the same mistake, with about half the fuel everyone thought it had.

Over Red Lake, Ontario, at 41,000 feet, both engines stopped and the electronic cockpit displays went blank, leaving only a few basic standby instruments. Captain Pearson, an experienced glider pilot, steered for a disused military airfield at Gimli, Manitoba, which by then was partly used as a drag racing strip. He landed it with nobody killed and no serious injuries. A 1985 board of inquiry said calling it a simple metric mix-up was too simple, and blamed poor communication within the airline as well as individual errors.

Every step of the calculation looked reasonable. The numbers were real, the method was standard, and two separate checks agreed with each other, because both used the same wrong factor. Nobody asked the simple question: does this total match what this aircraft usually carries? A number with the wrong unit looks exactly like a right number.

AI makes this kind of mistake easy to repeat. Ask it to combine sheets in different currencies, units or date formats, and it will often produce a clean, confident total without flagging that the inputs did not match. Asking the same AI to check its own sum usually repeats the same assumption. A quick sanity check against a figure you already know catches what a second identical calculation cannot.

The idea behind this caseLesson: Spreadsheets and data →

1. 1Put the unit or currency in every column heading, and tell the AI the units of your data in the prompt.
2. 2Ask the AI to list the units and formats it assumed before you trust a total.
3. 3Compare every important result with a rough figure you already know, like last month's total.
4. 4Make the second check different from the first: another person, another method or another data source.
5. 5When the normal tool or gauge is broken, slow down. Manual fallbacks are where unit mistakes hide.