From Input to Understanding
Artificial intelligence systems are typically designed to respond to specific commands. The Trustworthy Algorithmic Delegate (TAD) project, led by Parallax Advanced Research, takes this further by developing systems that align to user priorities like mission and quality of life, based on their responses to example situations. The goal is to enable AI to make autonomous decisions that align to human priorities, even when ethics and tradeoffs are unclear, and experts disagree. This is intended to increase the trustworthiness of AI decisions across a range of applications.
On July 6, 2023, Parallax Advanced Research was awarded a $4.067 million contract from the Defense Advanced Research Projects Agency (DARPA) for the In the Moment (ITM) initiative. The Parallax effort is focused on creating trustworthy AI systems that can make complex, real-time decisions in high-stakes situations— where existing AI systems may make unethical decisions or fail to decide.
Parallax worked in close collaboration with Drexel University and Knexus Research Corporation to advance human-aligned decision-making in medical triage and mass casualty care environments.
“DARPA was looking for AI systems that could make human-aligned decisions in difficult environments—decisions that reflect human priorities like quality of life, fairness, or mission focus,” said Matt Molineaux, director of Al/Autonomy, Program Manager ACT3 and PRIDE and research scientist on the TAD team. “Our approach with TAD was to develop a solution that could reason in those terms, even with limited data, using a case-based, highly explainable framework.”
After months of foundational development, the TAD team has reached a significant technical milestone: transitioning from exploratory research to functional prototypes. This shift signals a pivotal moment—not only for the team, but for the future of natural human-AI collaboration.
The TAD system leverages breakthroughs in natural language processing, symbolic reasoning, and explainable case-based reasoning to create agents capable of recognizing and applying human priorities. Unlike conventional AI models that rely on statistical regularities to generate middle-of-the-road responses, TAD analyzes the context and meaning of a unique situation in terms of human viewpoints and considers what a human decision-maker would likely prioritize.
At the core of this achievement is a modular framework that translates ambiguous language into structured meaning representations. These are then grounded in real-world tasks using an expanding set of inference engines and domain-specific interpreters.
“We designed the system to be highly explainable—something that people could interpret just by looking at what it’s doing,” Molineaux said. “It needed to mimic the kind of characteristics humans show in decision-making and be able to trace its choices back to trusted human examples,”
From Concept to Capability
Over the last quarter, the ITM team successfully demonstrated TAD capabilities in human-subject experiments, meeting or exceeding all DARPA metrics for success. The system demonstrated the ability to:
Make decisions in multiple domains
Acquire representations of human characteristics underlying their decisions
Encourage willingness to delegate in subjects with medical expertise
This prototype has been validated in user-facing trials, showing notable gains in accuracy and trust when compared to decision systems that do not model human decision makers.
What makes TAD especially transformative is its potential to redefine how humans and machines collaborate under pressure. By enabling AI systems to reason about goals, constraints, and preferences, TAD supports applications ranging from mission planning and decision support to critical care and education.
Future versions of the TAD system are expected to dramatically improve emergency medical decision-making—offering essential support in both military and civilian contexts. It could change how care is delivered when time, resources, and clarity are limited.
Project Closeout
As the TAD project concludes, the team reflects on its transition from idea to prototype and the valuable collaborations forged with government, academic, and commercial partners. These partnerships have played a pivotal role in demonstrating ITM’s potential to scale and influence the future of intelligent, trustworthy systems.
“With transparency as a guiding principle, we embedded mechanisms for after-action review and designed subsystems to reflect how people reason—predictively, diagnostically, and within bounded rationality,” said Molineaux.
While the active research phase has ended, TAD leaves behind a flexible architecture and a robust foundation validated through real-world engagement. For those continuing the pursuit of AI that advances from mere responsiveness to genuine understanding and alignment with human values, TAD offers a compelling blueprint for what comes next.
###
About Parallax Advanced Research & The Ohio Aerospace Institute (OAI)
Parallax is a 501(c)(3) private nonprofit research institute that tackles global challenges through strategic partnerships with government, industry, and academia. It accelerates innovation, addresses critical global issues, and develops groundbreaking ideas with its partners. With offices in Ohio and Virginia, Parallax aims to deliver new solutions and speed them to market. In 2023, Parallax and OAI formed a collaborative affiliation to drive innovation and technological advancements in Ohio and for the nation. OAI plays a pivotal role in advancing the aerospace industry in Ohio and the nation by fostering collaborations between universities, aerospace industries, and government organizations, and managing aerospace research, education, and workforce development projects.
