More than 10 years ago, when climate scientist Tapeo Schneider at the California Institute of Technology first modeled how clouds form, he had to painstakingly adjust the equations describing how water droplets, airflow, and temperature interact. But in 2017, artificial intelligence (AI) technologies such as machine learning became his right-hand man. Schneider stated that machine learning models are faster and provide more satisfactory models, making climate modeling and climate science more interesting. The UK's Nature website recently reported that scientists are using various AI technologies to accelerate climate modeling and improve its performance, with the aim of improving model accuracy while reducing energy consumption. Of course, given the existence of a "black box" in AI, not everyone fully trusts models based on machine learning technology. Traditional models have shortcomings. Traditional climate models use mathematical equations to describe how the interactions between land, ocean, and air affect climate. These models run well and provide climate prediction information that can be used to guide global policy-making. But these models require powerful supercomputers to run for several weeks and consume extremely high energy. Traditional models simulate a century of climate and consume up to 10 megawatt hours of energy, which is approximately equal to the average annual electricity consumption of American households. In addition, these models are difficult to simulate small-scale processes such as how raindrops form, but these small processes play an important role in large-scale weather simulations. Machine learning refers to computer programs learning by discovering patterns in a dataset. Computer scientist Adia Grover from the University of California, Los Angeles has pointed out that a series of innovations in machine learning have the potential to "showcase their skills" in climate modeling. Simulators are fast and accurate. There are currently three main ways for researchers to model climate using AI. The first approach requires the development of a machine learning model called a simulator, which can provide the same results as traditional models without performing all mathematical calculations. In 2023, climate scientist Vasily Kizios and colleagues from the Commonwealth Scientific and Industrial Research Organization in Australia developed 15 machine learning models to simulate 15 physics based atmospheric models. They use physical models to train the QuickClim system. These physical models are designed to predict atmospheric temperature in 2100 for both low-carbon and high carbon emissions scenarios. In the scenario of moderate carbon emissions, the trained QuickClim predicts atmospheric temperature in 2100, which is very consistent with physics based models. Once trained on all three carbon emission scenarios of low, medium, and high, QuickClim can quickly predict the changes in global temperature this century, which is about one million times faster than traditional models. Coincidentally, in 2023, scientists from the Allen Institute of Artificial Intelligence also developed machine learning models for a physics based atmospheric model