ADESSE: Advice Explanations in Complex Repeated Decision-Making Environments

TL;DR

ADESSE addresses the challenge of human trust and decision quality in complex, repeated decision-making by generating explanations about a two-component adviser agent (prediction and DRL). It combines local SHAP-based top-predictor features with domain-specific DRL indices and a policy- visualization using arrows to produce concise, readable explanations. Computational experiments show scalability and smaller explanations than LIME across taxi and wildfire environments; a game-based user study demonstrates higher explanation satisfaction, faster decision times, and a trend toward higher rewards under ADESSE. The work highlights the value of tailoring explanations to both prediction and DRL components to improve human–AI collaboration in dynamic decision tasks.

Abstract

In the evolving landscape of human-centered AI, fostering a synergistic relationship between humans and AI agents in decision-making processes stands as a paramount challenge. This work considers a problem setup where an intelligent agent comprising a neural network-based prediction component and a deep reinforcement learning component provides advice to a human decision-maker in complex repeated decision-making environments. Whether the human decision-maker would follow the agent's advice depends on their beliefs and trust in the agent and on their understanding of the advice itself. To this end, we developed an approach named ADESSE to generate explanations about the adviser agent to improve human trust and decision-making. Computational experiments on a range of environments with varying model sizes demonstrate the applicability and scalability of ADESSE. Furthermore, an interactive game-based user study shows that participants were significantly more satisfied, achieved a higher reward in the game, and took less time to select an action when presented with explanations generated by ADESSE. These findings illuminate the critical role of tailored, human-centered explanations in AI-assisted decision-making.

Figures (8)

• Figure 1: An agent consisting of two components provides advice to a human decision-maker.
• Figure 2: An explanation generated by Adesse consists of (1) a list of top-ranked features for the prediction, (2) domain-specific indices summarizing the DRL input features, and (3) arrows visualizing the trained DRL policy.
• Figure 3: An example of Adesse explanation for the taxi environment. A is the taxi's current location and B is the advised next location, which lead to C-F in the next few steps following the trained DRL policy visualized as arrows. A list of top-ranked features would be displayed separately when the human selects one of these labelled locations.
• Figure 4: Mean and standard deviation of participant ratings on the explanation satisfaction scale comparing explanations generated by Adesse (top/green) and the baseline (bottom/gold). ($^{**}$ for $0.001 < p \leq 0.01$ and $^{***}$ for $p \leq 0.001$.)
• Figure 5: Distribution of the number of taxi trips in the New York City Yellow Taxi dataset in 2015 and 2016 visualized on a logarithmic scale through a 500m square grid.