# AI-in-the-Loop -- The impact of HMI in AI-based Application

**Authors:** Julius Schöning, Clemens Westerkamp  
**arXiv:** [2303.11508](https://arxiv.org/abs/2303.11508)  
**Date:** 2023-03-21  
**Summary page:** https://www.summarizepaper.com/en/arxiv-id/2303.11508v1/

## Key points

- The paper discusses the importance of human-machine interaction (HMI) in AI-based applications.
- HMI is usually missing during the AI architecture design and training of an AI model.
- The human-in-the-loop concept is prevalent in all other steps of developing AI from data analysis via data selection and cleaning to performance evaluation.
- By using HMI during AI architecture design, unproductive layers of the architecture can be highlighted so that lightweight network architectures for embedded applications can be created easily.
- This approach reduces the resources needed for AI development by avoiding training and evaluating AI architectures with unproductive layers leading to lightweight architectures.
- Enabling HMI during an AI uses inference introduces the concept of "AI-in-the-Loop," which combines both humans' strengths with those of machines.
- The authors propose that developers should consider incorporating HMI into their designs to create more efficient and effective systems while reducing development costs.

## Summary

The paper titled "AI-in-the-Loop -- The impact of HMI in AI-based Application" by Julius Schöning and Clemens Westerkamp discusses the importance of human-machine interaction (HMI) in artificial intelligence (AI)-based applications. While AI and HMI are two important keywords, they usually do not fit embedded applications. However, during the steps needed before applying AI to solve a specific task, HMI is usually missing during the AI architecture design and training of an AI model. The authors highlight that the human-in-the-loop concept is prevalent in all other steps of developing AI from data analysis via data selection and cleaning to performance evaluation. The paper proposes that during AI architecture design, HMI can immediately highlight unproductive layers of the architecture so that lightweight network architectures for embedded applications can be created easily. By using this approach, users can instantly distinguish which AI architecture should be trained and evaluated first since high accuracy on the task could be expected. This approach reduces the resources needed for AI development by avoiding training and evaluating AI architectures with unproductive layers leading to lightweight architectures. The resulting lightweight architectures will enable HMI while running the AI on an edge device. By enabling HMI during an AI uses inference, it introduces the concept of "AI-in-the-Loop," which combines both humans' strengths with those of machines. In this approach, the machine remains the working horse and primarily solves tasks; however, if it is unsure whether its inference solves a task correctly, it asks the user to use appropriate HMI. The authors conclude that this approach will make AI more reliable and explainable since HMI will become available in many applications soon. Therefore, they propose that developers should consider incorporating HMI into their designs to create more efficient and effective systems while reducing development costs. Overall, this paper provides valuable insights into how combining human expertise with machine learning algorithms can lead to better outcomes in various fields such as healthcare, finance, and transportation.