You will know, if you watch this program regularly, that we've spent a lot of time in recent months talking about artificial intelligence, the dangers that it presents, but also the opportunities it will offer to improve our lives. One of the most valuable might be its capacity to improve our weather forecast. Crucially important, of course, to farmers, to emergency planners, to energy companies, in fact, so many sectors of the economy that would depend on timely, accurate information. The teams at Google Deep Mind are developing a forecasting system using AI that will not only process more quickly all the information and data that we accumulate, but will also look at historic patterns and previous events to forecast with greater precision. I've been speaking about it to our climate editor Justin Rowland.

They start with up to 800 million separate observations of what's happening in the atmosphere. Those could be from all sorts of places, from satellites, from weather stations, boys in the ocean. There are even sensors collecting weather data on the nose of commercial jet planes. They take that data, they decide which they think are the strongest but most important bits of data. Normally, whittle it down to something like 10 million data points. They then put them into a supercomputer. We are talking truly enormous computers that attempt to model the way that heat and moisture is transferred through the atmosphere. Look at what they use the data to say what the weather is like now and how that weather is likely to change into the future and the models which take huge amounts of processing power. We are talking thousands of trillions of calculations per second. I was clarifying it with them because it's such an enormous number in order to come up with the weather forecast that we routinely use, which tell you at the most prosaic, whether it's going to rain tomorrow and you should put a coat on, all the way up to predicting storms and other extreme weather, which gives people full warning, which means, of course, that they can protect themselves, evacuate the area, or battle down the hatches to protect themselves if heavy weather is on its way.

How would this new AI tool from Google DeepMind do it differently?

Well, in a way, what it does is a lot simpler. Instead of trying to actually model the way the atmosphere works, the complex physics of the atmosphere, what it does is it ingests, it takes in the historic data. It looks at what the weather has been and how it's changed over time. It's taken every hour of data for the last 40 or 50 years, takes that into the computer. What the computer is doing is looking for patterns. When you put in the weather conditions today, it says, Oh, I've seen that before, and normally that develops into this. It uses historic data to predict what's going to happen in the future. Now, the level of detail it produces is nothing like the detail that we get from the traditional models at the moment. That could change as they decide to make the resolution a bit tighter, but it is extremely accurate, more accurate on most measures than traditional weather forecasting systems. It is a breakthrough, but it uses the output of the traditional models, it is the data it learns on. It does need those traditional models. We're not going to shut the computers down and go over to AI.

They work side by side. Ai produces forecast far, far quicker. Instead of hours of supercomputer time crunching those numbers, it can come up with a forecast in less than a minute using basically the computer you might sit on a desktop, a big computer, but not a giant supercomputer.

It's going to be such a different future, isn't it, Ed? Justing rollout there, our climate editor. We'll take a short break. Stay with us.