Receive free updates from Deepmind Technologies Ltd
We will send you a myFT Daily Digest e-mail covering the latest news from Deepmind Technologies Ltd. every morning
There is a saying about journalism: ‘If someone tells you it’s raining and another person tells you it’s dry, it’s not your job to quote both. Your task is to look through the window and find out what is true. ”
This is not literally what journalism is: readers can look out the window themselves so we can all be unemployed. In addition, the public wants to know if it will rain later.
This brings us to weather forecasts. They used to be only on radio and TV, and headbands inevitably wind their arms in front of giant cards. Now they are everywhere. They are on our phones. They’re on my laptop’s toolbar. Guys, wait, the chance of rain is 37%!
Such predictions are always wrong, as are opinion polls and strategic decisions by the governing bodies of football. Correction: such predictions are generally correct, but if they go wrong, people are taken hostage and blamed on the out-of-touch meteorological elite. Sorry, BBC Weather, but if I wanted an app to give me dangerous misinformation, I would have checked Facebook.
Due to the geography, Britain has unpredictable weather. Despite this, forecasts have improved tremendously. Dwight Eisenhower presumably said that the reason D-Day succeeded was that the Allies ‘had better meteorologists than the Germans’ (the Germans did not spy on the short period of calm in which the landings were possible).
Now says the Met Office: “A four-day forecast today is more accurate than a one-day forecast in 1980”. It adds that 92 percent of its temperature forecasts for the next day are accurate within 2C. It’s probably like the train company claiming incredible accuracy by recording all the trains in the middle of the day that no one uses, but I’m taking the point.
If this is still not reliable enough for you, there may be good news. This week, the Met Office and DeepMind, Alphabet’s artificial intelligence arm, unveils a new method between 5 and 90 minutes in advance the forecast of short-term rainfall. DeepMind’s previous work includes identifying breast cancer and severe kidney damage. Over time, it became something useful, like whether we should take an umbrella to the mailbox.
Existing supercomputers are accurate in predicting ‘low rainfall’, using data such as wind speeds, an academic paper by the researchers said. The new method, where algorithms are left to themselves to decipher the patterns between radar readings and rainfall, is better at “medium to heavy rain events” of the type that made climate change more likely. We will be able to predict how unpredictable we have made the planet.
DeepMind’s predictions are only 90 minutes ahead, and it has not been said how much more accurate they are than current methods. Yet, once AI achieves something, it is natural to ask if it will achieve it all. Will we one day have completely accurate weather forecasts? Would we want a world where we know what the weather is like every day of our lives? Is this what Australia looks like?
This is not going to happen because weather systems are so complex. But the death of uncertainty would be strange: yes, we have availability for weddings in June 2024, but do not book because it is going to rain. We will be less surprised, for better or for worse. There would be fewer cancellations because everything would be perfectly scheduled so that our lives would get busier and busier.
Would we talk more or less about the weather if there was no uncertainty? I reckon even more: people who have read the prophecy will want to spread the word. The likelihood of rain is still the apolitical icebreaker it is today.
Of course, it’s weird that we like to talk about the weather, but we’re afraid to talk about the climate. Would I prefer to know what the weather is like in two hours, or to change the climate in two decades? A prediction is only as good as the action it causes.