This commentary discusses the paper "Building machines that learn and think like people" by Lake et al., highlighting its insights into the development of artificial intelligence (AI) from the perspective of human intelligence. The author agrees with most of the paper's points but adds complementary perspectives. The commentary emphasizes the importance of learning from biological evolution, which has developed inductive biases that enable efficient learning and robustness in the human brain. It argues that while biological evolution has created human-like intelligence, this does not necessarily prove the feasibility of creating an arbitrarily powerful artificial intelligence (AUI), as it would require impossible physical constructs. The author suggests that AGI can be achieved by building systems that strongly exploit inductive biases, while remaining open to the possibility that some assumptions may be relaxed with advances in optimization algorithms. The commentary also discusses the limitations of "human-level performance" as a benchmark for AGI, noting that it is often based on narrow tasks and does not reflect true generalization abilities. The author also highlights the importance of neuroscience in understanding the brain's generative models, which differ from those used in AI. Examples include the role of lateral connections in the visual cortex and the factorization of contours and surfaces. These insights could help in developing more effective AGI systems. Finally, the commentary suggests that message-passing (MP) based algorithms could offer a practical alternative to MCMC for inference in probabilistic models, combining the speed of neural networks with the flexibility of MCMC. The author concludes that cognitive science and neuroscience can provide valuable insights for AI research, particularly in understanding the limits of human intelligence and how to overcome them.This commentary discusses the paper "Building machines that learn and think like people" by Lake et al., highlighting its insights into the development of artificial intelligence (AI) from the perspective of human intelligence. The author agrees with most of the paper's points but adds complementary perspectives. The commentary emphasizes the importance of learning from biological evolution, which has developed inductive biases that enable efficient learning and robustness in the human brain. It argues that while biological evolution has created human-like intelligence, this does not necessarily prove the feasibility of creating an arbitrarily powerful artificial intelligence (AUI), as it would require impossible physical constructs. The author suggests that AGI can be achieved by building systems that strongly exploit inductive biases, while remaining open to the possibility that some assumptions may be relaxed with advances in optimization algorithms. The commentary also discusses the limitations of "human-level performance" as a benchmark for AGI, noting that it is often based on narrow tasks and does not reflect true generalization abilities. The author also highlights the importance of neuroscience in understanding the brain's generative models, which differ from those used in AI. Examples include the role of lateral connections in the visual cortex and the factorization of contours and surfaces. These insights could help in developing more effective AGI systems. Finally, the commentary suggests that message-passing (MP) based algorithms could offer a practical alternative to MCMC for inference in probabilistic models, combining the speed of neural networks with the flexibility of MCMC. The author concludes that cognitive science and neuroscience can provide valuable insights for AI research, particularly in understanding the limits of human intelligence and how to overcome them.