A strong argument for supplementing machine learning with symbolic manipulation.
Machine learning has achieved amazing things, but we all know its limitations. For example, the very latest natural language processing tool still exhibits no common sense. Why? Well, here’s one reason: because machine learning is designed to answer the question, what, but not the question, why. It thrives on correlation and ignores causation. Gary Marcus rightly observes that this gap in AI research needs filling. His proposal is to create a hybrid AI architecture that combines machine learning with models of the world and relational understanding. Some voices in machine learning understand the problem. But there remains no consensus on how fix the problem. And fixing the problem is essential if we wish to forestall another AI winter.
Link to paper: https://arxiv.org/ftp/arxiv/papers/2002/2002.06177.pdf
You may also like to browse other AI papers: https://www.thesentientrobot.com/category/ai/ai-academic-papers/