The Fifth Elephant 2025 Annual Conference CfP

Abstract

As AI systems become increasingly agentic, safety becomes less about control and more about co-evolution. What would it mean to design AI not just for accuracy or alignment, but for mutual adaptability and scaling trust over time? This talk introduces relational alignment, a systems-level reframe that treats human-AI interaction as a living, co-evolving relationship. Drawing from behavioural systems design, coaching frameworks, and industrial process safety, I’ll offer real-world analogies and design principles that help build AI agents not just to get it right, but to be safe while still learning. We’ll explore how affordances, friction and memory can be designed to enable trustworthy interaction in motion, especially under drift, uncertainty or competing values. Instead of asking whether AI is obedient or interpretable, I ask, can it adapt with us, without breaking us?

Audience for this talk

This talk is for engineers, researchers and builders working on AI agents that act autonomously, learn from interaction, or make decisions in real-world systems. If you’re working on alignment, safety, or human AI interaction, especially in dynamic or uncertain environments, this talk will offer a new perspective on how to design agents that evolve well with people.

Key takeaways

• Why control-based safety methods fall short for adaptive, agentic systems
• How to design agents that build trust through interaction over time
• Simple principles for using friction, memory and feedback to improve alignment
• A live demo showing why capability isn’t enough, fit matters.

Speaker bio

Priyanka Bharadwaj is an author, educator and behavioural systems designer whose work lives in the intersection of AI, trust and ethical complexity. She teaches neuroscience-informed human behaviour courses at IIT Madras, and previously built India’s first matchmaking advisory platform using behavioural science and values alignment. Former VP at Delhivery and head of strategic supply chain for media categories at Amazon UK, Priyanka has designed systems that move packages, people and preferences, at scale. Her current work focuses on developing tools, courses and narratives around alignment in socio-technical systems, including writing her debut non-fiction on millennial matchmaking. She also writes a regular newsletter on Substack, called Shapely Gal.