My friend told me about the phenomenal weather prediction model that’s invented by Google’s DeepMind. So, I decided to take a closer look at it, and it completely blew my mind.
Google has developed an AI called GraphCast that can predict the weather more accurately than current supercomputers that monitor real-time weather patterns. Their secret weapon? A massive amount of data, not just from the globe’s many data points, but also dating back to 1970. By analyzing this vast trove of information, GraphCast identifies patterns in weather shifts, allowing for more accurate predictions.
GraphCast acts as an organizer that recognizes patterns and predicts what the weather will be like today by comparing the data in the past. The GraphCast uses statistics and advanced calculations to recognize patterns like after a very hot weather, rain is most likely to follow. Just like you can guess what will happen nesxt in a movie based on the genres you’ve seen before, Graphcast can do that with the weather.
What I think is uniquely amazing about this model is that it teaches us the value of our history.
Instead of trying to predict the future based on current events, it is better to use patterns from the past to estimate when something might occur. Recognizing the pattern and being in the vicinity of the next likely occurrence is better because it is 1. Effective 2. Less time consuming 3. Energy efficient.
This is very intuitive if you think about in retrospect. For example, imagine you’re playing catch. If you wait until the ball is almost near you to try to catch it, you’ll have to use a lot of energy and might even miss it. But if you watch how the ball has been thrown before, you can figure out where it’s likely to land and stand there to catch it easily,
The future implication of this model is immense.
There is an old saying that history repeats itself. This adage is even more close to my heart by seeing how AI models being developed to make our world more effective and easier to make the best decisions.
Chong



Leave a comment