AI-in-the-Loop -- The impact of HMI in AI-based Application
The paper discusses the integration of human-machine interaction (HMI) in AI development, particularly for embedded applications, where HMI is typically absent during AI architecture design and training. While human-in-the-loop (HITL) is widely used in data selection, cleaning, and evaluation, its role in AI model design is often overlooked. The proposed approach leverages HMI to identify unproductive layers in AI architectures, leading to lightweight and efficient models for embedded systems. The AI-in-the-loop concept is introduced, where AI handles tasks autonomously but requests human input via HMI when uncertain. This enhances AI reliability, explainability, and usability, making AI more accessible for real-world applications.
Read the full article here:
AI-in-the-Loop—The impact of HMI in AI-based Application
AI-in-the-loop concept enables the use of safe, secure, and reliable AI—the AI is the working horse and primarily solves the task and, with the support of minimal HMI, users’ domain expertises are integrated if needed.
Illustration of users’ workload needed for performing a complex task over the availability of systems.
1. User-in-the-loop systems: AI assists users but requires continuous human supervision and validation. These systems are common in aviation, healthcare, and cybersecurity, ensuring human control but maintaining a high workload.
2. AI-based expert systems: AI automates tasks using predefined rules and data analysis, reducing human effort. However, users still make the final decisions, and these systems struggle to adapt to unexpected scenarios.
3. AI-in-the-loop systems: AI operates autonomously but requests human input when uncertain. It continuously learns from user feedback, reducing workload while ensuring reliability, making it suitable for self-driving cars, defense, and aerospace.
4. Fully automated AI-based systems: AI makes all decisions without human intervention, using deep learning and adaptive algorithms. These systems require high safety standards and strict regulatory approval, with future applications in aviation and healthcare.
From point 1. to point 4., the estimated timeframe is approximately 50 years.
