Value Alignment
How do we ensure AI decisions reflect human priorities — not just training data defaults? The ALIGN System steers decisions toward user-defined attributes like fairness, risk aversion, and moral deservingness.
Pluralistic AI
Whose values should AI represent? Different people hold different moral priorities. The ALIGN System personalizes decision-makers to diverse, fine-grained value profiles.
Cross-Model Alignment
Different LLM backbones can produce different default choices. The ALIGN System normalizes decisions toward the same target values, making aligned behavior more consistent across models.
Research Paper
ICML 2025 workshop paper on prompt-based attribute alignment for personalized LLM decision-making.
Blog Post
Kitware's overview of ethical and explainable AI contributions from DARPA ITM Phase 1.
Full Sandbox App
Interactive tool to compare aligned and unaligned AI decisions across scenarios and values.