Agentic AI for Responsible Tourism Recommendations

How can AI-powered recommender systems move beyond optimising for individual preferences to support more responsible tourism? We begin by examining the evolution from traditional recommender systems to LLM-powered and agentic recommendation, exploring how LLMs enable richer contextual understanding, conversational interaction, and reasoning, and how agentic AI can integrate multiple objectives and stakeholders into the recommendation process. Grounding this transition in practical research, we spotlight two recent projects: Collab-REC, a multi-agent framework that brings together personalisation, popularity, and sustainability objectives, and TRACE, an interactive conversational system designed to help users understand and consider more sustainable travel choices.Using Gemini and Google’s Agent Development Kit (ADK), these projects illustrate how emerging agentic architectures can translate research ideas into practical AI systems. Looking ahead, as agentic recommenders become more autonomous and personalised, new questions emerge around privacy, transparency, evaluation, accountability, and the balance between user preferences and societal and environmental impact. The talk explores these open challenges and opportunities for building AI systems that are not only more capable but also more responsible. Who should attend? The underlying architecture of recommender systems solves a universal bottleneck in AI for Science: how to build autonomous agent systems that negotiate complex, competing objectives rather than optimising for a single metric. Whether you are balancing therapeutic efficacy against toxicity in healthcare regimens, trading off mechanical performance against lifecycle carbon footprints in materials discovery, or navigating ecological constraints in geoscientific resource modelling, the challenge is identical. Attendees will walk away with practical workflows for orchestrating multi-agent consensus, grounding generative agents in strict domain constraints, and designing transparent user-in-the-loop reasoning. If your research demands AI workflows that reason across multi-stakeholder trade-offs, ethics, and sustainability, this session provides an actionable blueprint applicable across scientific discovery and engineering.