Scientist

I am an AI systems researcher and engineer who builds intelligent systems and gets them into production. Thirty years of leading technical teams from concept through build, test, and delivery: the U.S. Navy nuclear submarine force, industry engineering organizations, a tenured computer science professorship, twelve years of industrial research at PARC (the Palo Alto Research Center), and startup leadership.

Today I am Chief Science Officer (i.e., Chief Scientist) at Filuta AI, where I lead invention from research through shipped product for a document-intelligence platform that turns large, heterogeneous document collections into searchable knowledge, deployed in production. The lab's automated-planning research has produced peer-reviewed publications at KR, ICAPS, and IJCAI, plus two patent applications.

My career focus is interactive artificial intelligence, human-machine collaboration, and automated planning, delivered as systems used by real people. My current research interest is how people and teams change as AI shifts from tool to collaborator, and how to mentor people through that transition. Across every setting, my work has been building the conditions for research to leave the lab: partnerships, multi-institution programs, and commercialization. 

The record: 115 publications (h-index 24, ~3,200 citations), 16 patent families with 9 U.S. patents granted, more than $36.6M in externally sponsored research as PI or Co-PI from DARPA, NSF, NIH, the Naval Research Laboratory, the Czech TWIST Programme and industry as well as 5 graduated PhD and MS thesis students, with doctoral graduates now at OpenAI, Sandia National Labs, and Salesforce.

Highlights along the way: led PARC's Fittle behavior-change platform from concept to commercialization; Co-PI on five DARPA programs and technical lead on four including Explainable AI; tenured associate professor at UNC Charlotte, where I co-founded the Game Design & Development programs and mentored 11 senior design and studio teams; Chief Scientist of the NSF-funded MavHome smart home; 3G base station software still in service from my Motorola years; and eight years enlisted in the Navy's nuclear program, finishing as an Engine Room Supervisor and lead shipboard instructor.

Away from the keyboard I serve on the board of a rural water association near Cloudcroft, volunteer as a projectionist at southern New Mexico's only art-house cinema, and work as a furnace glass artist, trained in Murano, Italy.

I study artificial intelligence (AI) in contact with the people who have to live with it. For thirty years my research has followed one rule: a system is not finished until it survives its users. That rule has taken me from a smart home that learned its inhabitants’ routines, to mobile health coaching that ran pilot studies with more than a thousand people and shipped as a commercial product, to autonomous drones that had to explain themselves to the people responsible for them, to a document-intelligence platform now in production with European government agencies. The questions have changed over three decades. The stance has not: build intelligent systems with people in the loop, deploy them for real, and study what actually happens. Today the people in the loop are no longer only users; they are teammates, and that shift is what my work now follows. 

The record

My research identity was formed on deployed systems. As Chief Scientist of the National Science Foundation (NSF)-funded MavHome project at UT Arlington, I turned data-driven generation of hierarchical partially observable Markov decision processes into a working, learning smart home, coordinating seven to eight researchers across teams led by five professors. As a professor at UNC Charlotte, I directed the Game Intelligence Group in interactive AI for games and simulation: behavior-based control, spatial reasoning and navigation representations, and machine learning from player data, producing 66 peer-reviewed papers with a group whose core was six to eight students. At PARC (the Palo Alto Research Center), I led the Fittle behavior-change platform from concept through more than fifty month-long build cycles and eight pilot studies with 1,200+ users to commercialization, and served as Co-Principal Investigator (Co-PI) and technical lead on four DARPA programs. The eXplainable AI (XAI) program’s COGLE project remains the deepest of these: we built systems that explained a drone’s decisions to the humans accountable for its missions, work reported in Applied AI Letters and carried into a 2026 U.S. patent on coordinated agent learning and explanation. Sixteen patent families, nine granted U.S. patents, and more than $36M in sponsored research as PI or Co-PI trace the same arc: research that leaves the lab.

Today, as Chief Science Officer at Filuta AI, I direct the research behind a document-intelligence platform that turns large, heterogeneous document collections into searchable knowledge with full audit provenance. Our automated-planning research runs against real products: planning-based regression testing of video games, learning planning action models from state traces, and evaluating large language models as translators from natural-language goals to the Planning Domain Definition Language (PDDL), published at KR, ICAPS, and IJCAI venues with the Filuta team.

The question I want to spend the next decade on

What happens to scientists, engineers, artists, scholars, and education when AI stops being a tool and becomes a collaborator? I watch this transition daily from inside a working AI lab, and I study it from wherever I can see it most widely; a role that serves research across a whole university sees it in every discipline at once. The interesting problems are not in the models. They are in the humans and the teams around the models: how work divides when one teammate is a machine, how trust gets calibrated, who verifies what, and what a student must now learn that their curriculum does not yet teach. I intend to study this transition where it is happening, with the people it is happening to, and my research program has three connected threads.

Teams with AI teammates. Student teams are an ideal observatory for this transition: real projects, real deadlines, and members encountering AI collaboration for the first time, with their practices still forming. I want to instrument that encounter. How do teams divide labor with AI agents in the loop? When do people accept generated work, and what makes them check it? Which roles disappear, and which new ones appear? The output is empirical: observed practices, measurable trust and verification behaviors, and curriculum that teaches judgment for an era when generation is cheap and responsibility is not. My studio-teaching and capstone-mentorship record, eleven interdisciplinary teams shipping working systems through public demonstration days, is the methodological foundation for this thread, and the same questions extend to faculty and research teams across every college.

Formal scaffolding between human intent and machine execution. When an AI collaborator does the building, the specification becomes the contract. My planning research points directly at this: representations like PDDL, and the neurosymbolic and composite-AI architectures behind Filuta AI products, are exactly the kind of formal middle ground where human intent becomes checkable machine work. I will continue this line, extending action-model learning and language-to-specification translation, because it is both a live research area and the infrastructure the first thread needs: teams collaborate best with machines through artifacts both can verify. This work is publishable in many venues and it generates systems that students at every level can build, break, and study.

Explanation as an artifact. COGLE taught me that explanation is not a courtesy feature; it is the evidence an accountable human needs before signing their name. As AI moves into all kinds of teams, explanation becomes code review, becomes test evidence, becomes safety culture. What must an AI collaborator show the person accountable for the work to justify accepting it? I come to this question with a specific formation: eight years in the Navy’s nuclear program, where the documentation is the operation and no one signs what they cannot defend. Bringing that qualification culture to human-AI work is, I believe, the most consequential thread of the three.

How I work, and with whom

My method is to build real systems, deploy them with real users, and measure, combining systems research with the formal user-experience and product evaluation practice I ran in industry. Students are co-investigators at every level: I have supervised 26 NSF Research Experiences for Undergraduates students, six industrial interns, two postdocs, and graduated doctoral students now at OpenAI, Sandia National Laboratories, and Salesforce. My final doctoral student completed a nationally competitive dissertation while employed full-time, and advising students on their terms remains the model I value. I publish where the threads live: planning and knowledge-representation venues, interactive-AI venues, and computing-education venues for the curriculum results.

On funding, my record and my intentions are the same. I have been PI or Co-PI on more than $36.6M in sponsored research from DARPA, NSF, the National Institutes of Health, the Naval Research Laboratory, the Czech TWIST Programme, and industry, most often as the technical execution lead who delivers a multi-institution program against its milestones. I partner well with colleagues who own a program vision; I bring the delivery record and the industry network that sponsors trust, and my sponsor-side experience means I know what keeps an industrial partner engaged with a research team. Industry-sponsored projects, multi-institution collaborations, and education-focused programs are where I do my best work, and where I can help others do theirs.

Thirty years ago I maintained the robots in the lab I would later help lead. The through-line from there to here is short: intelligent systems are interesting to me exactly where they meet the people who must rely on them. The next decade of that meeting happens in teams and across whole enterprises, and that is where I want to do the work.

Seventeen externally funded research projects, 2004–2027, totaling approximately $36.6M (USD equivalent): $32,866,937 at PARC / Xerox / SRI, $1,035,542 at UNC Charlotte, $755,500 at UT Arlington, and ∼$1,977,963 at Filuta AI (42.57M Kč total project cost). Sponsors span DARPA (I2O, DSO, IPTO), NSF, NIH, NRL, the Czech TWIST Programme, and industry. Roles: PI on four awards; Co-PI and technical execution lead on twelve; senior researcher and project manager on the current TWIST award. 

Projects are listed newest first. Amounts are as awarded; two PARC awards include the listed institutional match.

Self-Service AI Planning Agents (“Samoobslužní AI plánovací agenti”)

• • Role: Senior Researcher & Project Manager

  • PI: Filip Dvorák (Filuta AI) and Roman Barták (Charles University)

  • Team: Filuta AI CZ s.r.o. (lead applicant) and the Faculty of Mathematics and Physics, Charles University (MFF UK), Prague

  • Sponsor: TWIST Programme, Czech Republic

  • Amount: 42,570,710 Kč  total project cost (∼USD $1,977,963), 69.25% grant-funded — Filuta AI 36,790,550 Kč  (∼$1,709,399); MFF UK 5,780,160 Kč  (∼$268,564)

  • Period: 9/2025 – 8/2027 (Filuta AI)

DARPA Knowledge Management at Scale and Speed (KMASS)  Retrieving, Organizing, and Creating Knowledge for Execution on Tasks (ROCKET)

• Role: Co-PI, Initial Technical Manager, Corpus Development & Evaluation Lead 

• PI: Peter Patel-Schneider then Leora Morgenstern then Richard Crouch with me as interim at times 

• Team: PARC/SRI

• Sponsor: DARPA DSO (PMs: Ted Senator, Matthew Marge) 

• Amount: $7,329,800 

• Period: 9/2022 – 9/2025 (PARC / SRI International) 

DARPA Computational Cultural Understanding (CCU) — MAKSIMUS: Mixed Architectures and Knowledge Structures for Inferring, Mastering, and Understanding Sociocultural Norms

• • Role: Co-PI

  • PI: Leora Morgenstern (PARC)

  • Team: Co-PIs: Matthew Shreve, Michael Youngblood (PARC); Dan Goldwasser (Purdue); Yair Neuman (Ben-Gurion)

  • Sponsor: DARPA I2O (PM: William Corvey)

  • Amount: $5,997,259

  • Period: 3/2022 – 2/2026 (PARC)

DARPA Perceptually-enabled Task Guidance (PTG) — AMIGOS: Autonomous Multimodal Ingestion for Goal-Oriented Support

• • Role: Co-PI

  • PI: Charlie Ortiz (PARC)

  • Team: Co-PIs: Matthew Shreve, Bob Price, Ed Stabler, Michael Youngblood (PARC); Xifeng Yan (UCSB); Kristina Yordanova (U. Rostock); Patrick O’Shaughnessy (Patched Reality)

  • Sponsor: DARPA I2O (PM: Bruce Draper)

  • Amount: $5,852,523 + $1,300,000 PARC match = $7,152,523

  • Period: 11/2021 – 10/2025 (PARC)

DARPA AIE COnstructive Machine-learning Battles with Adversary Tactics (COMBAT) — CHARGE: Conflict-resolving Hierarchical Adversarial planneR Guided by Extracted text

• • Role: Co-PI

  • PI: Roni Stern (PARC)

  • Team: Co-PIs: Leora Morgenstern, Michael Youngblood, Peter Patel-Schneider (PARC)

  • Sponsor: DARPA DSO (PM: Paul Zablocky)

  • Amount: $999,845

  • Period: 9/2020 – 3/2022 (PARC)

DARPA Explainable AI (XAI) — COGLE: COmmon Ground Learning and Explanation

• • Role: Co-PI

  • PI: Mark Stefik (PARC)

  • Team: Co-PIs: Michael Youngblood (PARC); Peter Pirolli (IHMC); Christian Lebiere (CMU); Ram Ramamoorthy (Edinburgh); Honglak Lee (U. Michigan)

  • Sponsor: DARPA I2O (PMs: Dave Gunning, Matt Turek)

  • Amount: $7,331,193 + $800,000 PARC match = $8,131,193

  • Period: 5/2016 – 3/2021 (PARC)

Personalized Health Behavior System to Promote Well-Being in Older Adults (NIH R01)

• • Role: Co-I

  • PI: Sara Czaja (University of Miami Miller School of Medicine)

  • Team: Co-Is: Peter Pirolli, Michael Youngblood (PARC)

  • Sponsor: National Institutes of Health (NIH)

  • Amount: $2,025,247

  • Period: 3/2016 – 2/2020 (PARC)

SCH: INT: Collaborative Research — FITTLE+: Theory and Models for Smartphone Ecological Momentary Intervention

• • Role: Co-PI

  • PI: Peter Pirolli (PARC)

  • Team: Collaborative institution: Carnegie Mellon University (Bob Kraut)

  • Sponsor: NSF–NIH Smart and Connected Health

  • Amount: $1,231,070

  • Period: 10/2013 – 9/2017 (PARC)

TUES: Building BRIDGES Within the Undergraduate Major in Computer Science

• • Role: Co-PI

  • PI: K.R. Subramanian (UNC Charlotte)

  • Team: Co-PIs: Michael Youngblood, Robert Kosara, Jamie Payton, Paula Goolkasian

  • Sponsor: National Science Foundation (NSF)

  • Amount: $250,000

  • Period: 6/2013 – 8/2015 (UNC Charlotte)

Growing Greener

• • Role: Co-PI

  • PI: Tiffany Barnes (UNC Charlotte)

  • Team: Co-PIs: Michael Youngblood, Heather Lipford

  • Sponsor: Corporate sponsor (NDA restricts disclosure)

  • Amount: $178,000

  • Period: 2/2011 – 8/2011 (UNC Charlotte)

Automated Navigation Mesh Creation in Procedurally-Generated Worlds

• • Role: PI

  • PI: Michael Youngblood (UNC Charlotte)

  • Team: Summer international research embedded with Gamr7 in Roanne, France, with doctoral student D. Hunter Hale

  • Sponsor: NSF PIRE (through Florida International University)

  • Amount: $14,000

  • Period: 5/2011 – 8/2011 (UNC Charlotte)

Digital Kennedy LoRez Wall

• • Role: PI

  • PI: Michael Youngblood (UNC Charlotte)

  • Sponsor: UNC Charlotte Office of Information Technology (CIO Jay Dominick)

  • Amount: $28,000

  • Period: 5/2010 – 5/2011 (UNC Charlotte)

DARPA Computer Science Study Group, Phase 2

• • Role: PI

  • PI: Michael Youngblood (UNC Charlotte)

  • Sponsor: DARPA DSO (PM: Ben Mann)

  • Amount: $488,862

  • Period: 5/2008 – 5/2010 (UNC Charlotte)

DARPA Computer Science Study Group, Phase 1

• • Role: PI

  • PI: Michael Youngblood (UNC Charlotte)

  • Sponsor: DARPA DSO (PM: Ben Mann)

  • Amount: $76,680

  • Period: 4/2007 – 4/2008 (UNC Charlotte)

Transfer Learning in Integrated Cognitive Systems

• • Role: Co-PI

  • PI: Larry Holder (UT Arlington)

  • Team: Co-PIs: Michael Youngblood, Diane Cook, Manfred Huber

  • Sponsor: DARPA Transfer Learning Program (IPTO, PM: Ted Senator), subcontract via the Institute for the Study of

  • Learning and Expertise (ISLE)

  • Amount: $620,000

  • Period: 10/2005 – 10/2006 (UT Arlington)

Autonomous Vehicles Lab Activity Support Grant

• • Role: Co-PI

  • PI: Atilla Dogan (UT Arlington)

  • Team: Co-PIs: Brian Huff, Arthur Reyes, Kamesh Subbarao, Michael Youngblood

  • Sponsor: Bell Helicopter–Textron

  • Amount: $60,000

  • Period: 1/2006 – 12/2006 (UT Arlington)

Integration of a Cognitive Architecture and an Urban Warfare Simulator for the Evaluation of AI Methods

• • Role: Co-PI

  • PI: Larry Holder (UT Arlington)

  • Team: Co-PI: Michael Youngblood

  • Sponsor: Naval Research Laboratory (BAA 55-03-02, PM: David Aha)

  • Amount: $75,500

  • Period: 9/2004 – 8/2005 (UT Arlington)