Judea Pearl, a pioneering figure in artificial intelligence, argues that AI needs to understand cause and effect to be truly intelligent. Current AI systems lack this understanding, leading to limitations in their reasoning abilities. Pearl suggests that AI should be taught to reason about interventions and counterfactuals, which could lead to a new level of intelligence.
Pearl’s work has led to the development of a three-level causal hierarchy. The first level is association, which involves observing regularities in data. The second level is intervention, where one manipulates variables to see the effect. The third level is counterfactuals, where one imagines alternative realities.
Pearl’s view is not universally accepted. Some argue that deep learning, a current AI technique, can achieve true intelligence without understanding cause and effect. Pearl counters this by saying that deep learning is stuck at the level of associations and cannot reason about interventions or imagine alternative realities.
Pearl’s work is seen as a potential path to developing AI systems that can understand and reason about the world in a way that is closer to how humans do. However, implementing this in practice remains a significant challenge. Critics argue that it’s unclear how to teach machines to understand cause and effect, especially when dealing with complex real-world situations.
Go to source article: https://www.quantamagazine.org/to-build-truly-intelligent-machines-teach-them-cause-and-effect-20180515/