Human-centric AI

AI technology is reshaping a wide range of human activities and disciplines. The Human-Centric AI Research Group (HCAIR) at Heriot-Watt University Dubai develops AI-enabled technologies that place the human at the centre of system design, aligning technology with human needs and values, and complementing rather than replacing human skills and abilities. Our research sits at the intersection of machine learning, multimodal AI, intelligent agents, computer vision, natural language processing and distributed intelligence.

How can we build AI systems that work effectively with people while remaining adaptive, trustworthy and aligned with human goals?

Human-centricity, for us, is a design constraint across the AI lifecycle: the data used for learning, the decision procedures and their explanations and the conditions under which a system yields measurable human benefit. HCAIR connects to the University's Global Research Institute (GRI) in Health and Care Technologies, spanning Heriot-Watt's Scotland, Dubai and Malaysia campuses to turn research into practical health and care solutions.

Our mission

Our mission is to advance AI from isolated prediction towards systems that perceive, reason, collaborate and adapt under real-world constraints. We pursue three closely connected goals:

• Advance the science of adaptive and collaborative intelligence: Multimodal learning, autonomous agents, multi-agent systems and distributed AI.
• Build trustworthy AI for real-world use: Explainability, fairness, governance and human oversight embedded into system design.
• Translate AI research into meaningful impact: Working with clinicians, government and industry to tackle problems where AI augments rather than replaces human expertise.

Research themes

Human-centric perception and multimodal intelligence

We study multimodal fusion of vision, language, sensor and contextual signals so that models support interpretation, not only recognition, inferring what observed evidence means for a human decision-maker.

Agentic AI and collective intelligence

We study agentic and multi-agent systems in which specialised agents, humans and digital services coordinate on planning and reasoning. Focus is placed on LLM-based and neuro-symbolic agents that cooperate with people on tasks where current models remain unreliable.

Trustworthy, explainable and responsible AI

We develop methods for interpretability, fairness, auditability and uncertainty quantification, treating these properties as architectural requirements rather than post-hoc add-ons to otherwise opaque models.

Distributed, federated and edge intelligence

We develop federated, edge and distributed learning methods that enable collaboration without centralising data, addressing privacy, latency and governance constraints in healthcare, smart cities and autonomous systems.

AI for health and human wellbeing

We develop clinically supervised models for patient modelling and decision support, designed to augment clinical judgement rather than replace it.

From fundamental AI to real-world impact

HCAIR's methods are applied across five key application domains, each providing a demanding real-world setting for testing and refining our research:

• Healthcare: Clinical AI, precision and personalised care, connected to the GRI in Health and Care Technologies.
• Education and lifelong learning: AI-supported learning and adaptive educational tools.
• Smart Cities and Transportation: Multi-agent coordination and digital twins for urban environments.
• Creative industries and UI/UX: Human-centred design and multimodal interaction for creative tools and interfaces.
• Energy and Smart Grids: AI for energy forecasting and smart-grid management.

Our approach

A defining characteristic of HCAIR is co-creation: we work closely with the people who understand the problem — clinicians, patients, researchers, policy makers and industry partners — incorporating domain expertise from the start. We combine fundamental AI research with real-world datasets, prototypes and deployments, testing not just whether an algorithm performs well in isolation, but whether it remains useful and understandable in the environment it was designed for.

Selected current projects