Designing Intelligent Systems for Symbiotic Trust

This talk traces a path from a critique of anthropomorphism to a constructive design principle for intelligent systems. I begin with a historical argument developed in my earlier work on the history of intelligent machines: from ancient automata to contemporary AI, machine design has been persistently anthropocentric. It is built to imitate humans, extend human dominance over the environment, or ease human labor. I will show why this lineage is not a neutral design choice but carries real costs. It distorts self-understanding, encourages passive over-reliance, and widens the empathy gap between people and the living systems and environments they depend on. Out of this critique, I proposed a set of ecocentric design principles as an alternative orientation; one that does not center human interests by default, but treats intelligent systems as embedded participants in a wider ecology of human, machine, and environmental actors. 

I will then turn to how this orientation has shaped my more recent work in multi-robot systems, where my lab has developed symbiosis as a concrete coordination mechanism rather than a metaphor: heterogeneous robots that reshape each other's reward structures and reallocate tasks under uncertainty, coordinating as interdependent agents rather than through fixed hierarchies. I will walk through examples from our recent papers on neuroplasticity-inspired task allocation and reward-decomposition-based coordination, showing what symbiotic behavior looks like when implemented and evaluated in real multi-robot teams. 

Finally, I will lay out where this work is heading. Extending symbiotic coordination frameworks to explicitly include humans as actors within the multi-agent system, rather than as external operators the system merely assists or imitates. This reframes the design question for anthropomorphic and intelligent assistive systems. It moves the focus to not how convincingly a system can perform humanlikeness, but how genuinely it can participate as a symbiotic partner alongside the people, robots, and environments it shares.

Reducing Traffic Accidents Involving Older Drivers: Driver Agent Research 

Japan is facing a super-aging society, where many older adults continue to rely on private cars for daily mobility. To promote safer driving, we have been developing a Driver Agent, an anthropomorphic in-vehicle agent that provides real-time driving support and post-driving reflection support. Rather than directly instructing drivers, the agent is designed as a companion-like presence that helps drivers recognize their own driving behavior and voluntarily adopt safer driving practices.

In this talk, I will introduce the concept and development of the Driver Agent and present findings from driving simulator studies, public-road experiments, and a nationwide long-term field study. I will also discuss recent applications of this approach to public transportation, including robot conductors for autonomous buses and robot-mediated interactions for encouraging safer passenger behavior.