Probabilistic machine learning and artificial intelligence, as discussed by Zoubin Ghahramani, emphasize the role of uncertainty in learning and decision-making. The probabilistic framework provides a way to model and manipulate uncertainty about models and predictions, playing a central role in scientific data analysis, machine learning, robotics, and cognitive science. This article introduces the probabilistic approach to machine learning and Bayesian inference, and reviews state-of-the-art advances such as probabilistic programming, Bayesian optimization, data compression, and automatic model discovery. The key idea is that learning involves inferring plausible models to explain observed data. Uncertainty is fundamental, as observed data can be consistent with many models, and predictions are uncertain. Probability theory provides a framework for modeling uncertainty. The article highlights five areas of current research: probabilistic programming, Bayesian optimization, probabilistic data compression, automatic model discovery, and hierarchical modeling. Probabilistic modeling uses probability theory to express all forms of uncertainty, including model uncertainty, parameter uncertainty, and structural uncertainty. Probability distributions represent uncertain quantities and their relationships to data. Learning involves transforming prior distributions into posterior distributions through Bayesian learning. Probabilistic models are compositional, allowing for the construction of complex models from simpler ones. Nonparametric models offer flexibility by allowing predictions to grow in complexity with the amount of data. Examples include Gaussian processes and Dirichlet processes. Probabilistic programming allows the representation of probabilistic models as computer programs, enabling the automatic inference of unobserved variables. This approach has the potential to revolutionize machine intelligence and scientific modeling by automating inference and enabling rapid prototyping of models. Bayesian optimization is a method for globally optimizing unknown functions, often used in machine learning and scientific applications. It involves sequential decision-making to find the maximum of an expensive-to-evaluate function. Bayesian optimization is closely related to reinforcement learning and has applications in areas such as drug design and robotics. Data compression is another area where probabilistic modeling is applied. Bayesian methods improve compression rates by learning better probabilistic models of data. Compression algorithms like Sequence Memoizer and PPM-DP are equivalent to Bayesian nonparametric models of sequences. Automatic model discovery involves learning and explaining statistical models from data. The Automatic Statistician is a system that automatically discovers plausible models and explains them in plain English. This approach is useful for fields reliant on data extraction and has applications in scientific discovery and data science. Probabilistic approaches to machine learning and intelligence are a growing area of research with wide-ranging impacts beyond traditional pattern recognition. These approaches handle uncertainty in a principled manner, leading to more robust and interpretable models. Challenges remain in computational efficiency, but recent advances in inference methods have made probabilistic methods computationally competitive with traditional approaches. Ultimately, probabilistic modeling will continue to play a central role in the development of more powerful machine learning and artificial intelligence systems.Probabilistic machine learning and artificial intelligence, as discussed by Zoubin Ghahramani, emphasize the role of uncertainty in learning and decision-making. The probabilistic framework provides a way to model and manipulate uncertainty about models and predictions, playing a central role in scientific data analysis, machine learning, robotics, and cognitive science. This article introduces the probabilistic approach to machine learning and Bayesian inference, and reviews state-of-the-art advances such as probabilistic programming, Bayesian optimization, data compression, and automatic model discovery. The key idea is that learning involves inferring plausible models to explain observed data. Uncertainty is fundamental, as observed data can be consistent with many models, and predictions are uncertain. Probability theory provides a framework for modeling uncertainty. The article highlights five areas of current research: probabilistic programming, Bayesian optimization, probabilistic data compression, automatic model discovery, and hierarchical modeling. Probabilistic modeling uses probability theory to express all forms of uncertainty, including model uncertainty, parameter uncertainty, and structural uncertainty. Probability distributions represent uncertain quantities and their relationships to data. Learning involves transforming prior distributions into posterior distributions through Bayesian learning. Probabilistic models are compositional, allowing for the construction of complex models from simpler ones. Nonparametric models offer flexibility by allowing predictions to grow in complexity with the amount of data. Examples include Gaussian processes and Dirichlet processes. Probabilistic programming allows the representation of probabilistic models as computer programs, enabling the automatic inference of unobserved variables. This approach has the potential to revolutionize machine intelligence and scientific modeling by automating inference and enabling rapid prototyping of models. Bayesian optimization is a method for globally optimizing unknown functions, often used in machine learning and scientific applications. It involves sequential decision-making to find the maximum of an expensive-to-evaluate function. Bayesian optimization is closely related to reinforcement learning and has applications in areas such as drug design and robotics. Data compression is another area where probabilistic modeling is applied. Bayesian methods improve compression rates by learning better probabilistic models of data. Compression algorithms like Sequence Memoizer and PPM-DP are equivalent to Bayesian nonparametric models of sequences. Automatic model discovery involves learning and explaining statistical models from data. The Automatic Statistician is a system that automatically discovers plausible models and explains them in plain English. This approach is useful for fields reliant on data extraction and has applications in scientific discovery and data science. Probabilistic approaches to machine learning and intelligence are a growing area of research with wide-ranging impacts beyond traditional pattern recognition. These approaches handle uncertainty in a principled manner, leading to more robust and interpretable models. Challenges remain in computational efficiency, but recent advances in inference methods have made probabilistic methods computationally competitive with traditional approaches. Ultimately, probabilistic modeling will continue to play a central role in the development of more powerful machine learning and artificial intelligence systems.