The rapid evolution of artificial intelligence (AI) has been accompanied by a growing role for open-source models. The recent release of DeepSeek R1 represents a notable shift towards customizable, independent, and privacy-conscious AI solutions. In the department Human-AI Interaction (HAI) at Fraunhofer IOSB, we investigate the implications of these advancements in all kind of AI-driven applications, e.g. production environments, or vehicle interiors (cars, trucks, etc.). Open-source models enhance local AI applications, rendering them more adaptive, secure, and efficient.
Unlocking the Potential of Open-Source AI
The introduction of DeepSeek R1 and analogous models presents novel opportunities for the application of AI. In contrast to proprietary, cloud-based models like ChatGPT, open-source AI gives us complete control over model customization, ensuring that our deployed AI adapts precisely to the specific workflows and environments of our customers. This independence is crucial for edge and local AI solutions, as it allows for AI-driven applications to function without reliance on cloud services.
AI models can now be fine-tuned for specific tasks, such as data analysis, image recognition, or even user interfaces. Tailored solutions that meet the precise operational requirements of customers are now within reach. On top of that, sensitive data remains within local infrastructure, ensuring compliance with GDPR and industry regulations, while edge AI capabilities allow these models to run directly on embedded hardware, reducing latency and dependency on external servers.
At Fraunhofer IOSB we already run our own prototypes locally e.g. in our research vehicle KALLE, a Level 3 automated Mercedes EQS we use for data collection, user studies, technology evaluations and development of customized AI interior systems.
GenUIn: Generated User Interfaces Using Open Source Generative AI
A key to adaptive user interfaces is not only recognizing the user’s context and needs, but to adapt assistance functions for intuitive control and dialog. The latter is enhanced tremendously using using generative AI to build Generated User Interfaces (GenUIn) – intelligent, adaptive interfaces that dynamically adjust to user behavior and context. Open-source AI, particularly multimodal models like DeepSeek R1, elevates this adaptability to a new level. This technology enables the comprehension of multimodal inputs, including text, speech, and images altogether, thereby facilitating the development of intuitive interfaces that adjust dynamically based on user needs. In contrast to static menus, AI-generated interfaces adapt in real-time, adjusting menus, assistance prompts, and UI components to ensure a seamless experience. These interfaces also provide personalized assistance, adapting to individual preferences and creating a tailored in-vehicle or industrial user experience. One of our key projects, in which we develop and evaluate generated user interfaces, is SALSA – a research project funded by the Federal Ministry of Economic Affairs and Climate Action in which we cooperate with Audi, MAN, Valeo, bast, CanControls, Elektrobit, Invensity, WIVW, and many more. Watch out for our demos of this technology in the upcoming IAA and other conferences and exhibitions.
Adaptive AI and User Expectations
A primary challenge in the realm of AI-driven interfaces pertains to achieving a balance between adaptability and predictability. Users have expressed a desire for interfaces that are intuitive yet dynamic, providing helpful responses without inducing feelings of unpredictability. Open-source AI plays a pivotal role in refining this balance by facilitating transparent customization and iterative improvements.
Take our advances inside the car, where we deploy our advanced occupant monitoring system to give the car eyes and understand the given situation between the passengers at any given time. AI-generated interfaces have the potential to personalize in-cabin experiences by adapting to driving conditions, passenger preferences, and real-time sensor data, ensuring that interactions are both seamless and effective. In addition, context-aware UI adaptations can mitigate motion sickness by adjusting screen brightness, content display, and interaction methods. Concurrently, AI ensures that interactions remain both predictable and adaptable, evolving naturally based on user feedback without introducing unnecessary complexity. Interested in finding out more? In our publication Activities that Correlate with Motion Sickness in Driving Cars – An International Online Survey we describe our exemplary implementation on motion sickness and activity recognition with local visual language models (VLM).
Security and Privacy: The Big Advantages of Local AI Models
In the context of automotive and industrial applications, ensuring security and reliability is crucial. DeepSeek R1 and analogous open-source models offer a distinct advantage over API-based AI solutions by ensuring that data remains within the system’s local confines. The local and controlled execution of AI computations eliminates all security risks associated with cloud dependencies, thereby ensuring consistent system availabilities without the need for internet connectivity or trustworthy third-party servers. Staying in our example of incabin monitoring and user interface adaptation: Running things locally meets all stringent data protection requirements of automotive safety and cybersecurity standards – something cloud-based alternatives can hardly keep up with. Part of our research is how to deploy AI in secure and safe ways. Our research project Anymos, which is funded by the Federal Ministry of Education and Research, particularly focuses on data privacy in deployed AI systems within the car’s interior.
Future Prospects: Open-Source AI in Next-Generation Interfaces
As AI technology advances, open-source models will become essential in designing next-generation multimodal interfaces. Our department for Human-AI Interaction is committed to driving this transformation by expanding multimodal AI capabilities, integrating language, vision, and sensor-based inputs to create holistic AI-driven interactions. To ensure the accessibility of these interfaces across diverse platforms, researchers are optimizing edge AI for low-power devices, thereby enabling the efficient operation of AI models on automotive and industrial-grade hardware. Through the continuous redefined conception of AI-assisted interaction, we are shaping the future of AI-driven UI, ensuring that systems evolve in accordance with user expectations.
Conclusion: We embrace Open-Source AI for Human-Centered Innovation
The advent of DeepSeek R1 and analogous open-source AI models signifies a pivotal moment in the evolution of AI-driven interaction. By capitalizing on these models, our research is at the forefront of pioneering secure, efficient, and adaptable AI applications. This encompasses a spectrum of innovations, from the generation of user interfaces to multimodal AI-driven automation. Consequently, open-source AI is laying the foundation for a future where human-centered AI solutions are not merely reactive, but genuinely intelligent and context-aware.
Get in touch with us if you want to find out more.