Recent progress of artificial intelligence demonstrates a pattern of emphasizing the mining of "big data "while ignoring explicit knowledge representations. In contrast,cognitive science has revealed that human intelligence is based on a set of remarkably powerful representations,which are simultaneously expressive,sparse and computationally efficient. Together they enable human to acquire knowledge through few data and transfer that knowledge to many novel situations—both of which are superior than artificial intelligence. Here we review "hierarchical tree " as a prominent representation,which can express infinite meaning by recursively applying fixed rules to fixed features. Whiling sufficing the general requirements of an intelligent representation,hierarchy tree is also unique as it( a) consumes little cognitive resources through feature-sharing;( b) describes a task with different levels of abstractions,and( c) expresses the causal process of data generation. We highlight the utilities of "hierarchical tree"by reviewing studies of both language and vision. Based on empirical results of human perception and learning,we propose that cognitive studies on knowledge representation is capable of inspiring a new generation of artificial intelligence.