By Hadj Batatia, Radu-Casian Mihailescu, Md Azher Uddin
The Association for the Advancement of Artificial Intelligence’s 2025 Spring Symposium Series was held in Dubai, UAE, May 20-May 22, 2025. There were four symposia in the spring program: AI-Driven Resilience: Building Robust, Adaptive Technologies for a Dynamic World, AI in Business: Intelligent Transformation and Management and Context-Awareness in Cyber-Physical Systems.
AI-Driven Resilience: Building Robust, Adaptive Technologies for a Dynamic World (S1)
The AI for Resilient Communities symposium explores the intersection of artificial intelligence, resilience, and adaptive technologies, highlighting AI’s transformative role in helping communities navigate environmental, economic, and social uncertainties. As societies face escalating challenges—from climate crises to shifting economic landscapes—the need for resilient, adaptive systems has never been more critical. This symposium is designed to foster innovation and dialogue around creating robust communities that can withstand and adapt to crises, evolving into stronger and more resilient entities over time.
This event uniquely promotes synergies between resilience research and AI, advancing the development of adaptive technologies that empower individuals and communities to respond to unforeseen challenges. Although the field of AI-driven resilience and adaptive technology is still emerging, it presents distinct technical, ethical, and operational challenges that require collaborative solutions. The symposium brings together AI researchers, social scientists, policymakers, and industry leaders to identify current gaps, share insights, and devise strategies to address these challenges effectively.
Attendees are invited to present innovative research, current projects, and theoretical perspectives that contribute to enhancing resilience through robust AI-driven technologies. Submissions that explore scalable AI applications, adaptable to diverse contexts, are especially encouraged. By uniting experts from various disciplines, this symposium aims to accelerate the development of adaptive technologies that build resilient communities capable of thriving in an uncertain world. No formal report was filed by the organizers for this symposium.
AI in Business: Intelligent Transformation and Management (S2)
The Symposium/Stream aims to explore the strategic deployment of Artificial Intelligence (AI) across various business domains to drive competitive advantage, enhance operations, and foster sustainable growth. The event will equip participants with actionable insights into the latest trends, challenges, and applications of AI in business transformation and management. The event aims to provide a platform for academics, practitioners, and policymakers to exchange actionable insights, explore the latest advancements, and address the challenges of integrating AI into business transformation and management. No formal report was filed by the organizers for this symposium.
Context-Awareness in Cyber-Physical Systems (S3)
The AAAI 2025 Summer Symposium on Context-Awareness in Cyber-Physical Systems convened researchers and practitioners to explore advancements in context-aware technologies within CPS. Context-aware CPS detect, interpret, and adapt to real-time environmental or user conditions, achieving situational awareness for intelligent autonomous behavior. Emphasizing the integration of AI, ML, and sensor fusion, the symposium highlighted applications across healthcare, industrial automation, and smart cities, underscoring the transformative potential of context-awareness in enhancing system adaptability and intelligence.
Dr. Radu-Casian Mihailescu and Dr. Md Azher Uddin served as co-chairs of this symposium. The symposium featured two keynote talks and four technical sessions (with 16 research papers). The program covered topics from machine learning and computer vision to trustworthy AI, privacy, human-centric design, and IoT autonomy
The organizing committee comprised:
- Radu-Casian Mihailescu, Heriot-Watt University
- Md Azher Uddin, Heriot-Watt University
- Hanan Salam, New York University Abu Dhabi
- Aziz Nasridinov, Chungbuk National University
- Sascha Ossowski, King Juan Carlos University
- Paul Davidsson, Malmö University
- Young-Koo Lee, Kyung Hee University
Two compelling keynote addresses showcased the breadth of context-aware cyber-physical systems across human-centeredness and sustainability, respectively. Dr. Hanan Salam from NYU Abu Dhabi, explored how socially and emotionally intelligent CPS can adapt to human cognitive and affective states. Drawing on her lab’s work with neurodivergent populations and AI-driven coaching systems for ADHD, she emphasized the role of real-time personalization using LLMs and affective computing to foster ethical and effective human-AI collaboration. From another perspective, Dr. Huma Zia, from Abu Dhabi University, delivered a talk sharing experimental deployments of smart agricultural CPS. Her work demonstrated how AI-enabled IoT systems, ranging from water-optimizing irrigation tools to autonomous weed removal robots, can enhance productivity, reduce environmental impact, and support food security. Together, the keynotes highlighted the diverse ways in which context-awareness can drive innovation and address critical societal challenges.
Context Modeling and Representation
The first session showcased new approaches for modeling context in CPS, moving beyond fixed rule-based methods by leveraging machine learning. For example, one study proposed using LLMs as an adaptive cognitive layer to flexibly interpret contextual information, while another introduced a knowledge graph framework for interpretable video-based activity recognition. The session also presented an explainable, retrieval-augmented method for detecting misinformation using context cues, illustrating how AI techniques can enrich context modeling across applications.
Security and Privacy in CPS
The session on security and privacy focused on robust AI-driven strategies for securing context-aware CPS. One presentation described a context-aware IoT security framework for elderly care facilities showing that incorporating contextual cues can improve threat detection in healthcare
IoT. Another applied hierarchical multi-agent reinforcement learning to harden CPS against cyberattacks, demonstrating the potential of adaptive multi-agent AI for system defense. Researchers further proposed an oversampling technique (E-SMOTE) leveraging group context to boost fraud detection, and evaluated online learning algorithms for Android malware detection.
Machine Learning for Context Recognition
The third session explored advanced machine learning techniques for context recognition and interpretation. For example, one team used an LLM-based model to guide autonomous drones with high-level contextual instructions, while another developed a deep spatiotemporal network to recognize words in Pakistani Sign Language. In the healthcare domain, researchers applied unsupervised learning to cerebrospinal fluid proteomics data to track Parkinson’s disease progression, and introduced a retrieval-augmented model to improve appendicitis diagnosis from sparse clinical notes.
Industrial Applications and IoT
The final session showcased real-world case studies where several projects demonstrated how AI-driven context awareness can improve efficiency and safety. For instance, the “I-TREES” framework adapts to environmental context to optimize energy use in forest sensor networks for tree health monitoring. In manufacturing, a digital twin platform for SOFC electrode engineering integrates AI and virtual simulations to fine-tune production conditions, enhancing consistency and performance. For workplace safety, a multimodal glove compliance system combines image segmentation with pose estimation to more reliably detect missing protective gear, reducing false positives and boosting detection accuracy.
In summary, the AAAI S25 symposium showcased how cutting-edge AI and machine learning methods are driving progress in context-aware CPS across all these themes. Whether through learning rich context models, enhancing security, interpreting complex data, or enabling autonomous behavior, AI is the key enabler for CPS that truly understand and react to their context. The diverse contributions demonstrated the growing impact of context-aware CPS across multiple domains, pointing toward more adaptive and intelligent cyber-physical systems ahead.
Human-AI Collaboration: Exploring diversity of human cognitive abilities and varied AI models for hybrid intelligent systems (S4)
Human-AI collaboration has become a cornerstone of next-generation intelligent systems, prompting a rethinking of design paradigms so that artificial agents function not merely as tools but as genuine partners in complex tasks. At its core lies a user-centered philosophy: AI systems are expected to understand, anticipate and adapt to human needs, preferences and cognitive styles, all while maintaining transparency and accountability. Achieving such synergy requires the integration of rigorous theoretical frameworks drawn from cognitive science, human-factors engineering, ethics and social psychology, each grounding interaction design in measurable, human-centered principles.
Four primary focus areas define current research and development efforts:
- Explainability and Interpretability: Methods that expose internal model reasoning in accessible formats, enabling non-expert users to follow decision pathways, build trust and exercise informed oversight.
- Human-in-the-Loop Methodologies: Collaborative workflows in which humans and machines alternate decision steps, leveraging human intuition for ambiguous or high-stakes judgements, and AI’s capacity for large-scale data processing.
- Immersive Environments: Applications of virtual and augmented reality that create rich, multisensory collaboration spaces, allowing AI to scaffold human creativity, situational awareness and skill acquisition.
- Diversity-Aware Systems: Adaptive architectures that recognize variability in human cognitive abilities and cultural backgrounds, tuning AI behavior to individual strengths and fostering inclusivity and equity.
The anticipated impact spans multiple sectors. In healthcare, human-AI teams promise earlier and more accurate diagnoses, personalized treatment planning and enhanced patient monitoring, while ensuring clinicians retain ultimate oversight and ethical responsibility. In education, intelligent tutoring systems and collaborative learning platforms support diverse learners, tailoring instructional strategies to individual needs and promoting deeper engagement. In creative industries, co-creative tools enable novel artistic expression and accelerate design workflows, preserving the human spark at the heart of innovation. By uniting robust theory, cutting-edge methodology and real-world applications, human-AI collaboration seeks to augment rather than replace human expertise, empowering people to tackle increasingly complex challenges with ethical integrity.
The AAAI Summer Symposium 2025 on Human-AI Collaboration comprised two keynote talks, one tutorial, 34 papers and a brainstorming session. Professor Merouane Debbah’s keynote on “Pioneering Intelligent Connectivity for the Future of Generative AI” explored TelecomGPT, redefining intelligent connectivity by enhancing customer service and optimizing network performance. It focused on the broader implications of Generative AI in telecommunications, transforming operational efficiency, enabling a 6G era of distributed, low-latency AI and unveiling new end-user services. The second keynote, by Professor Abbas-Turki Abdeljalil, addressed human-AI collaboration in automated vehicles within industrial environments, emphasizing seamless and safe co-habitation between humans and smart vehicles in sensitive scenarios without disrupting workflow efficiency. An immersive demonstration illustrated various optimization strategies to achieve this objective.
The tutorial by Professor Mohamed Quafafou, “Computing with Perception in Hybrid Human-AI Systems,” tackled the challenge of human perceptual diversity when intelligent agents collaborate with people. Perceptual diversity originates from biological, cognitive, cultural, emotional and contextual differences, introducing complex alignment, interaction and decision-making challenges. Drawing on perception theory, the tutorial introduced the mathematical concept of observer-dependent sets, which allows modelling variability in human perception. This methodological tool enables relative inference on data from diverse sources and interaction with humans while respecting their diversity.
The 34 papers covered methodological innovations and practical experiments in four domains: healthcare; creativity and design; computer programming and software; and ethics and other applications. Notable contributions in healthcare included scenarios that integrate expert knowledge in medical diagnosis with large language models, collaborations between clinicians and AI agents, methods to analyze and support mental health disorders (including virtual reality environments), and policy proposals to promote AI in healthcare across Africa. In creativity and design, papers addressed thought embedding in large language models, human-AI co-creation applied to furniture, crochet, fashion design and music generation. Many contributions focused on software synthesis and testing via human-AI collaboration, combining formal approaches (such as executable graphs and multi-agent models) with LLMs. Additional papers examined ethical issues in large language models, assessed worker anxiety induced by AI agents, and explored augmented cybersecurity through human-AI collaboration.
A brainstorming session explored the creation of an international network for Diversity AI, named HUMANIA, aimed at promoting multidisciplinary research into human cognitive, physiological, social and cultural diversity in hybrid AI systems. Ongoing efforts seek to establish the network as an AAAI chapter.
The symposium was well attended, and organizers received very positive feedback from participants. Kayvan Karim (Heriot-Watt University) represented the symposium at the plenary session. The symposium co-chairs were Hadj Batatia and Mohamed Quafafou. This report was written by Hadj Batatia.
Author Bios
Hadj Batatia is an associate professor at the school of Mathematical and Computer Sciences, Heriot-Watt University.
Radu-Casian Mihailescu is an Associate Professor at the Department of Mathematical and Computer Sciences, Heriot-Watt University, Dubai.
Md Azher Uddin is an Assistant Professor at the Department of Mathematical and Computer Sciences, Heriot-Watt University, Dubai.