Toward Accountable and Explainable Artificial Intelligence Part Two: The Framework Implementation

Abstract

This paper builds upon the theoretical foundations of the Accountable eXplainable Artificial Intelligence (AXAI) capability framework presented in part one of this paper. We demonstrate incorporation of the AXAI capability in the real time Affective State Assessment Module (ASAM) of a robotic system. We show that adhering to the eXtreme Programming (XP) practices would help in understanding user behavior and systematic incorporation of the AXAI capability in Machine Learning (ML) systems. We further show that a collaborative software design and development process (SDDP) would facilitate identification of ethical, technical, functional, and domain-specific system requirements. Meeting these requirements would increase user confidence in ML and AI systems. Our results show that the ASAM can synthesize discrete and continuous models of affective state expressions for classifying them in real-time. The ASAM continuously shares important inputs, processed data and the output information with users via a graphical user interface (GUI). Thus, the GUI presents reasons behind system decisions and disseminates information about local reasoning, data handling and decision-making. Through this demonstrated work, we expect to move toward enhancing AI systems' acceptability, utility and establishing a chain of responsibility if a system fails. We hope this work will initiate further investigations on developing the AXAI capability and use of a suitable SDDP for incorporating them in AI systems.

Toward Accountable and Explainable Artificial Intelligence Part One: Theory and Examples

“…Part two of this paper shows how developers and practitioners would engage in the process of incorporating and evaluating the efficacy of the proposed framework. Also, translating the AXAI capabilities into a set of system design requirements is demonstrated in part two of this paper [55]. Together, the two papers will be useful in developing the system requirements and producing a design process model as shown in Fig.…”

“…As explicit in this paper and part two of this paper [55], the AXAI framework also provides design guidelines and encourages provision of separable and quantifiable parameters of accuracy, comprehensibility and accountability. This makes the proposed AXAI capability framework different from existing XAI incorporation methods.…”

“…Delineating the three ML systems in a 3D AXAI capability space demonstrated that the proposed framework was helpful in system design and assessment of ML systems. Furthermore, the AXAI capability framework also provided an opportunity to systematically address ethical and professional issues, such as those highlighted in [40,55] while building ML systems. As evident in the above comparison, the nine measurable components of the AXAI capability framework ensured paying attention to system details, ethical responsibilities and moral duties during the conceptual design and functional analysis stages.…”

“…It is important to mention that the three components of the AXAI's comprehensibility vector rely on users' ability to inspect and understand information as we did not discuss or recommend any method of measuring the data inspectability in this part of the work. However, part two of our work [55] addresses the issue and recommends a collaborative system building approach that requires system developers and users to agree on the quality of inspectable information [64].…”

“…The ASAM was developed for real time multimodal analysis of facial expressions and speech for assessing affective states. The design and implementation of the ASAM is detailed in an accompanying paper entitled "Toward Accountable Explainable Artificial Intelligence Part two: The Framework Implementation" published in this journal [55]. The provisions of explainability and accountability in the ASAM were ensured by adding the AXAI features listed in Table 1.…”