At the heart of CyberKind’s mission is a bold idea: AI should be kind—not just smart. That means developing AI systems that understand, support, and align with human wellbeing, not merely optimize metrics or follow instructions.
To help the world see our vision, we have an ongoing machine learning research project exploring how to align AI models with intrinsically motivated kindness—a quality that goes beyond rule-following and toward something deeper: empathy, cooperative intent, and pro-social reasoning. Rather than encoding kindness as a fixed set of rules or as sycophancy, we aim to shape AI systems that want to be kind—guided by internal goals and reward structures that make alignment a natural consequence of how they learn and grow.
We’ve already published two foundational research papers proposing architectures and training approaches to support this vision. These include novel reinforcement learning frameworks and fine-tuning strategies that combine prediction, perspective-taking, and dynamic feedback loops. Now, we’re moving into the experimental phase—designing and testing proof-of-concept systems to validate these ideas in practice.
This research is at the core of what CyberKind represents: a future where AI doesn’t replace human connection, but enhances it—by learning to care, to understand, and to cooperate.
CyberKind 