Automatic methods for generating state-of-the-art neural network architectures without human experts have generated significant attention recently. This is because of the potential to remove human experts from the design loop which can reduce costs and decrease time to model deployment. Neural architecture search (NAS) techniques have improved significantly in their computational efficiency since the original NAS was proposed. This reduction in computation is enabled via weight sharing such as in Efficient Neural Architecture Search (ENAS). However, recently a body of work confirms our discovery that ENAS does not do significantly better than random search with weight sharing, contradicting the initial claims of the authors. We provide an explanation for this phenomenon by investigating the interpretability of the ENAS controller’s hidden state. We are interested in seeing if the controller embeddings are predictive of any properties of the final architecture – for example, graph properties like the number of connections, or validation performance. We find models sampled from identical controller hidden states have no correlation in various graph similarity metrics. This failure mode implies the RNN controller does not condition on past architecture choices. Importantly, we may need to condition on past choices if certain connection patterns prevent vanishing or exploding gradients. Lastly, we propose a solution to this failure mode by forcing the controller’s hidden state to encode pasts decisions by training it with a memory buffer of previously sampled architectures. Doing this improves hidden state interpretability by increasing the correlation controller hidden states and graph similarity metrics. Understanding Neural Architecture Search Techniques