Gary Marcus, a researcher at New York University, presents a critical appraisal of deep learning in his article. While deep learning has seen significant progress since its resurgence in 2012, Marcus argues that it has limitations and is not a universal solution for artificial general intelligence. He outlines ten key concerns, including the fact that deep learning is data-hungry, lacks the ability to handle hierarchical structures, struggles with open-ended inference, and is not transparent. Additionally, he points out that deep learning systems are vulnerable to being fooled and are difficult to engineer with. Marcus suggests that deep learning should be supplemented with other techniques to achieve artificial general intelligence. He also warns against excessive hype surrounding AI, which could lead to an AI winter similar to the one in the 1970s. Instead of viewing deep learning as a universal solution, Marcus proposes that it should be seen as one tool among many, and that other approaches, such as unsupervised learning and symbolic AI, may be necessary for future progress in AI.Gary Marcus, a researcher at New York University, presents a critical appraisal of deep learning in his article. While deep learning has seen significant progress since its resurgence in 2012, Marcus argues that it has limitations and is not a universal solution for artificial general intelligence. He outlines ten key concerns, including the fact that deep learning is data-hungry, lacks the ability to handle hierarchical structures, struggles with open-ended inference, and is not transparent. Additionally, he points out that deep learning systems are vulnerable to being fooled and are difficult to engineer with. Marcus suggests that deep learning should be supplemented with other techniques to achieve artificial general intelligence. He also warns against excessive hype surrounding AI, which could lead to an AI winter similar to the one in the 1970s. Instead of viewing deep learning as a universal solution, Marcus proposes that it should be seen as one tool among many, and that other approaches, such as unsupervised learning and symbolic AI, may be necessary for future progress in AI.