Literature review was carried out to analyze existing approaches at the crossroads of xAI and CAs. The social sciences are indicating that xAI is conversational [1]. Moreover, the embedding of xAI within a conversational interface such as chatbots is an underexplored area.

Initially we started working with the original use case, which is the smart manufacturing use case, but we shifted to a simpler dataset [2], with the intention to revisit the original use case later.

We identified user questions could be organized into categories and mapped to specific xAI techniques [3,4].

Building on this understanding, we implement the question bank into an intents-based framework, mapping each category to an appropriate xAI technique were relevant, while recognizing that some cases do not require one [5]. Furthermore, we incorporated a LLM to generate responses for each question category.

Having learned key lessons from the hello world use case, we have returned to the smart manufacturing case, applying the acquired knowledge to enhance, refine and further advance this use case.

Now, we have two distinct scenarios, each with their own xAI Question Bank, and their own implementation:

1. Apples dataset: Developed for a broader audience that does not have prior knowledge in manufacturing, easier to understand by the general public
2. Smart manufacturing: Our primary use case, developed for individuals with a background in the manufacturing field.

Conclusions

To integrate xAI into a CA for smart manufacturing presents challenges, but is definitely feasible. This synergy holds potential to greatly improve decision-making processes in smart manufacturing and drive the advancement and adoption of Industry 4.0.

Technologies

References