We are pleased to announce that three papers from the BiConnect Lab have been accepted at AAMAS 2026, the International Conference on Autonomous Agents and Multi-Agent Systems—one of the leading international venues shaping research on AI interaction, coordination, and decision-making in complex socio-technical environments.
As AI algorithms and models become increasingly powerful and embedded in everyday life, the interaction between humans and AI systems is no longer peripheral—it is central. At BiConnect, we study how humans connect, perceive, and collaborate across Human–AI ecosystems, and how these interactions can be taught, audited, and regulated to support well-being, autonomy, and responsible innovation.
AAMAS 2026, the International Conference on Autonomous Agents and Multi-Agent Systems, is one of the leading international venues shaping research on autonomous agents, multi-agent interaction, learning, and decision-making in complex human–AI and socio-technical systems. The conference will be held in Cyprus, and our contributions span interactive learning for large language models, transparency in social media recommender systems, and the mitigation of problematic online behavior.
Accepted papers
Teaching LLMs Naturally: Pedagogical Strategies for Interactive Knowledge Acquisition (Full Paper)
Sabrina Patania, Luca Annese, Cansu Koyuturk, Dimitri Ognibene
This work advances the study of interactive learning for large language models, showing that explanation-driven, mixed-initiative dialogue enables more natural and adaptable knowledge acquisition. By framing learning as a pedagogical interaction rather than passive data exposure, the study contributes to the development of human-aligned, post-training learning paradigms for AI systems with wide applications outside labs and tech companies.
→ Link to OpenReview
Toward Recognizing Social Media Recommenders under Absent Recommendations (Extended Abstract)
Sabrina Guidotti, Dimitri Ognibene, Giuseppe Vizzari, Udo Kruschwitz, Gregor Donabauer, Davide Taibi
Social media recommender systems strongly shape collective attention while remaining largely opaque. This work formalizes the challenge of recommender identification under missing recommendations and introduces SM-ARR-G, a first approach based on graph neural network and simulated infospheres. The contribution advances algorithm auditing and transparency, supporting evidence-based assessment of AI influence in online social environments.
→ Link to OpenReview
Mitigating Problematic Social Media Use through Paired Recommender Systems with Contrasting Objectives(Extended Abstract)
Stefano Livella, Luca Bolis, Sabrina Patania, Matteo Papini, Dimitri Ognibene
This study explores how recommender systems can be designed to regulate engagement rather than maximize it indiscriminately. Through simulation, we show that pairing engagement-driven and regulation-oriented modules can reduce addiction-like behaviors while maintaining a sustainable engagement, pointing toward more responsible and value-aligned recommender architectures.
→ Link to OpenReview
Across these papers, we engage with Human–AI interaction not just as a societal challenge, but as a source of new algorithmic questions, models, and experimental setups. From implementing interactive learners to simulating recommender dynamics, multi-agent simulations, graph neural networks, and dual-objective formulations give us a playground where understanding AI behavior is as important as building systems—and often more fun.
