Rationalization: A Neural Machine Translation Approach to Generating Natural Language Explanations

Abstract

We introduce AI rationalization, an approach for generating explanations of autonomous system behavior as if a human had performed the behavior. We describe a rationalization technique that uses neural machine translation to translate internal state-action representations of an autonomous agent into natural language. We evaluate our technique in the Frogger game environment, training an autonomous game playing agent to rationalize its action choices using natural language. A natural language training corpus is collected from human players thinking out loud as they play the game. We motivate the use of rationalization as an approach to explanation generation and show the results of two experiments evaluating the effectiveness of rationalization. Results of these evaluations show that neural machine translation is able to accurately generate rationalizations that describe agent behavior, and that rationalizations are more satisfying to humans than other alternative methods of explanation.

Friend, Collaborator, Student, Manager

“…Codella et al [9] demonstrated how explanations could improve model accuracy on three tasks, but required that every sample be hand-labeled with an explanation and treated explanations from different authors as equivalent. Ehsan et al [17] made use of explainable AI for automated game playing. Their approach relies on a secondary machine learning interpretation of the original behavior, rather than visualizing or explaining the original model as our approach does.…”

Extractive Adversarial Networks: High-Recall Explanations for Identifying Personal Attacks in Social Media Posts

“…While feature-based explanations are the most common approach, other forms have been proposed, including: similarity to learned "prototypes" which represent clusters of items from the training data (Li et al, 2017); high-precision feature interaction rules (Ribeiro et al, 2018); reference to predefined human-friendly concepts ; and generated natural language (Ehsan et al, 2017). Likewise, many evaluation criteria have been proposed.…”

Artificial Intelligence Bias in Health Care: Web-Based Survey

“…XAI aims to establish transparency by explaining what decisions led to the creation of the algorithm in addition to its inputs and outcomes, which provide the basis for trusting the algorithm [ 21 ]. Methods through which XAI can be established include layer-wise relevance propagation [ 22 ] and rationalization [ 23 ]. Furthermore, Friedrich et al [ 24 ] discussed the role and benefits of statistics, which they see as a natural partner in AI developments, for example, in calculating the sample size but also for bias control.…”