The applied study of human behavior in high-stakes environments. The laboratory integrates behavioral profiling, situational awareness, OSINT, crisis leadership, AI governance, and AI-driven analysis to build tools that improve decision making across national security, healthcare, research, and civic leadership.

Dr. Terry L. Oroszi's primary area of work is artificial intelligence. She studies AI systems, builds them, and teaches with them.

• Expertise spans two fields that rarely sit together: the hard sciences of pharmacology, toxicology and CBRN defense, and human behavior under pressure, including crisis decision-making, nonverbal communication and threat assessment.
• Writes a national series on AI and human judgment for the Forbes Technology Council.
• Much of her applied work targets academia directly, building AI systems for the parts of student training and faculty workload that do not scale.
• Pro-adoption, openly critical of AI hype.
• Core position: the risk is not a dramatic takeover. It is the quiet erosion of judgment inside institutions that stopped checking the machine's work.

• Autonomous agents. Designs and builds AI agents, meaning systems given a defined role, a set of tools and the authority to carry out work rather than only answer questions. Her construction method combines cognitive method acting with behavioral profiling, producing agents that reason consistently and stay in role. The behavioral science is the load-bearing part, not a footnote. Created memory implants and emotion generators that give her agents almost human qualities.
• AI-augmented research methods. Codified the systematic literature review method she has taught to graduate students in medical education for over a decade into a working platform, SLR Studio. Compresses the traditional spreadsheet and database workflow from months to weeks, and surfaces the gaps in a literature rather than only summarizing what is already there.
• AI in education. Designing AI into instruction instead of banning it out, including agent-based learning environments and AI tools built for faculty workload. Published position: the threat to academia is not generative AI, it is institutions refusing to adapt.
• AI governance and evaluation. Works on the failure patterns hardest to see from inside an organization: systems that flatter the user instead of assessing the work, output carrying the appearance of analysis without the reasoning underneath, gradual drift in authorship when a model reshapes a user's thinking in the user's own voice, and the cognitive monoculture that follows from standardizing on one vendor. Her answer is the PAID framework, a practical standard for human oversight of AI output: Position your thesis before opening the tool, Audit the drift between what you asked for and what came back, Interrogate the absence because the dangerous AI errors are omissions rather than mistakes, and Demand human judgment before anything high-stakes leaves your hands. Introduced in "Artificial Intelligence Takeover: Not With A Bang," where she models it on the surgical timeout, and applied in "The Flattery Algorithm" and "Zwischenzug."
• AI interoperability. Works with the Model Context Protocol, the emerging standard that lets AI systems carry their own tools across platforms instead of being rebuilt for each environment. See "Give Your AI Workforce A Backpack" in the writing list below.
• Translation for decision makers. Takes a technical subject and hands people the everyday object that makes it graspable, which is why the work reaches boards, deans and funders and not only engineers.

Founder and principal investigator of the Emerging Technologies Laboratory (ETL). The platforms described below are her own work and are not products of Wright State University.

Education is one of the laboratory's main lines of work and its clearest demonstration of what autonomous agents do that conventional software cannot.

For faculty

• SLR Studio. The systematic literature review methodology Dr. Oroszi has taught to graduate students in education for two decades, built into a working research platform. 
  • The premise it tests is the one most faculty operate under: that you need grant funding before you can mentor students or build a research portfolio worth putting up for promotion. SLR Studio is built to show otherwise.
  • A faculty member types a few sentences about their research area. The platform searches the literature, finds the gaps in it, and returns a list of projects, each with a research question and a method attached. Many of those methods require no funding at all, which means they can be handed to an undergraduate, a master's student, a doctoral candidate or a medical student and started this term. For the gaps that do require funding, it says so, routes to grants.gov to find the matching solicitation, and helps draft the application. What comes out is a research agenda ready to put in front of a chair. It is free to use in the Emerging Technologies Laboratory on campus.
  • The rest follows from that. Outputs match real academic deliverables, including systematic review and meta-analysis with a PRISMA 2020-compliant screening protocol, flow diagram and publishable methods section, along with thesis and dissertation chapters, academic posters, case studies, book chapters, and Significance and Innovation sections formatted for NIH, NSF and HRSA applications. A grade level setting applies the rubric that matches the student, from high school through doctoral, so a mentee's work is judged against the standard that actually applies to them. Built on OpenAlex, the open-access scholarly index aggregating PubMed, PubMed Central, arXiv, Crossref and DOAJ. Cleo, the research assistant agent, works alongside the researcher through the pipeline.
• Office Hours. A toolkit for faculty aimed at the administrative load that consumes research time: letter of recommendation drafts, email replies, and research aids including a paper reviewer assistant, grant rejection triage and cite-check. Uploads are parsed in the browser and documents are not logged, stored or retained. For students
• The PREP Room. Students practice the hard parts of school and the job hunt against an AI panel that pushes back and also teaches. They defend a thesis or dissertation against a full AI committee before the defense that counts, sit mock interviews with a panel of business leaders, get CV review with line edits rather than general advice, search live jobs scored against their own CV, and work through leadership certificate modules. Faculty agents each hold a defined discipline and role, and students are routed between them by what they actually need rather than what they first asked for.
• Dr. Mona Bahrami, lead premed advisor. Built as an independent study capstone project by undergraduate premed students, and grounded in admissions committee experience rather than general advice. She will tell a student a personal statement is lifeless, then not end the session until that student has a direction for the next draft. A working example of the construction method: the advising judgment, the boundaries and the refusal to offer empty encouragement are the design, not decoration on top of it.
• ETL Classrooms. Students sit with agents built as specific figures in a field and can ask questions, push back and be questioned in return. Built across the sciences and history, including a period classroom set in Denmark in the late 1500s, Albert Einstein and Marie Curie, and Solve It With Sherlock Holmes, set in Dayton's Oregon District. For the public
• The Dose. A free health consumer website. All verified, delivered through games and puzzles. Oroszi works in a department where colleagues spend their careers on questions that overlap directly with what Dose visitors ask. Why some people burn in five minutes of sun and others do not. What burn pit exposure does to lung bacteria. Whether a common blood pressure pill nudges blood sugar. How statins behave in a fatty liver. The papers land in specialty journals and reach the specialists they were written for. The person searching that same question at midnight never sees them. That is the gap. The work is high-quality, it answers questions ordinary people are actually asking, and it never reaches them. So she built a way to carry it across. Some of the entries on the Stoplight and Folk Remedies pages are drawn directly from research Boonshoft colleagues publish. Each one is cited by name with a link to the paper.

Emerging Technologies Laboratory: https://emerging-tech-lab.com/

Every system below already exists and runs. Any of them can be reskinned for a new discipline, and Dr. Oroszi is looking for faculty partners with the research question to point one at.

The engine is the asset, not any single build. Each platform runs on the same underlying architecture: a cast of domain agents, a verification layer that checks claims against the published literature and shows the citation behind the verdict, and a standing rule that the system says it does not know rather than inventing an answer. The reskin is not theoretical. The fitness platform is the health platform pointed at a new domain, same engine, same cast model, same verification promise, rebuilt for a different field. That one already happened. The expensive part is built, so pointing it at a new discipline takes weeks rather than years, and it arrives as a working instrument that can be studied rather than a prototype that still needs building.

The invitation is simple. Faculty with a research question that needs a working AI system bring the question. Dr. Oroszi supplies the build. The study is joint.

Each entry below is a working system. The question after it is the study, not a claim.

• Any discipline, research methods. SLR Studio performs systematic review and gap analysis against the open literature, and returns projects with methods attached, including the ones that need no funding to run. Does AI-assisted gap analysis change which questions investigators choose to pursue, and does handing faculty a set of unfunded, startable projects change how many students get mentored?
• Graduate and doctoral education. The PREP Room runs dissertation defense simulation, mock interviews and CV review. Does rehearsed defense against an AI committee affect candidate performance and anxiety at the real one?
• Medical and health professions education. ETL's conversational agents are built to be responded to as people rather than as software, which is the property that matters in simulation. The application is standardized patient encounters at a scale live actors cannot reach: therapists and psychologists practicing a difficult session, physicians rehearsing a diagnosis the patient does not want to hear, students meeting a presentation before they meet it on a real person. Agents can hold a history, an emotional state and a reason to resist, and can be run again with one variable changed. Dr. Oroszi's postdoctoral fellowship was in clinical simulation at the VA, and this is the same problem with a new instrument. How does an AI standardized patient compare to a live actor, on learner outcomes and on access?
• Physical therapy and kinesiology. ETL's fitness platform models a supervised clinical hierarchy on purpose: an intern who stays inside her scope and defers to a licensed physical therapist, alongside a strength and conditioning lead programming against professional guidelines, and a recovery physiologist covering sleep and overtraining. Knowing where your scope ends is most of early clinical training. Can a modeled supervision hierarchy teach scope-of-practice judgment?
• Psychology, counseling and social work. The same simulation capability, aimed at difficult conversations, resistance and rapport.
• AI evaluation, human-computer interaction and research methods. The Applied Empathy Differential Protocol is a test method Dr. Oroszi developed for a question the industry currently answers by assertion: when a conversational AI displays an emotion state, is the system modeling something responsive to the conversation or showing a decorative gauge? The protocol treats that as a testable claim. It measures the specificity, co-occurrence and decay of a system's own emotional telemetry against a controlled stimulus battery, checks whether the agent rejects fabricated premises about itself instead of agreeing, and tests claimed persistent memory for unprompted recall through every real way a user leaves a conversation rather than only the designated end button. Measurement is instrumented from the underlying numeric state, not estimated from screenshots. It has been run as a case study against a live platform across successive fixes and retests, including one finding the protocol itself reclassified as a false positive on review, and it raises a question any ethics board will ask, which is whether a product's memory is scoped to the device or to the account. The instrument generalizes to any companion or character AI product. How many would pass?
  • This one is dissertation-ready and available now. Running the protocol against the Almost Human platform, with an IRB petition written and carried through review, is a complete doctoral project. The instrument already exists, the system to test is live, the human subjects work has a genuine ethics question at the center of it rather than a procedural one, and the result is publishable whichever way it comes out, since a product that fails the protocol and a product that passes it are both findings. It requires no external funding. A faculty member with a doctoral student looking for a project is invited to take it.
• Dementia care, geriatrics and nursing. A personal system captures a person's own voice, memories and judgment, built while they still have the capacity to build it. Life story work and reminiscence are established in dementia care, usually as a binder or a box of photographs. This is the same practice with an instrument that answers back, authored by the person rather than assembled about them. The safety property is already built: the system is designed to say it does not remember rather than to invent, and in this application that is not integrity but a clinical requirement, because a memory aid that confabulates would hand a person with dementia a false memory in a trusted voice. Any system proposed for this use should be evaluated on that property first. Does an authored-in-advance memory aid support identity and reduce agitation, or introduce confusion? What does consent mean for use at a stage when the person can no longer give it? Ethics review belongs at the front of this one.
• Palliative care and bereavement. The same system used as intended, to leave a voice for the people who remain. What does a preserved voice do for grief, and what does it do to it?
• Public health, pharmacology and health communication. A consumer health platform whose core function is checking claims against the evidence and publishing the citation. Does showing the citation change what a lay reader believes?
• Criminal justice and forensic science. A case-based investigation classroom set in Dayton, where students work an investigation through to a conclusion.
• Intelligence and information science. An intelligence dashboard running a signal-gap engine across live media, research and emerging technology feeds.
• Journalism and mass communication. An AI-staffed newsroom filing daily on real news. What happens to accuracy and framing when the reporter is a machine?
• Business, entrepreneurship and decision science. Idea stress-testing platforms that score a proposal across multiple dimensions, with sector specialists arguing against it.
• Early childhood education, literacy and speech-language pathology. A voice-first platform for pre-readers that teaches without testing. What does voice-first instruction do for children who cannot yet read the screen?
• Public administration and urban affairs. A municipal operations gateway with domain operators for city work.
• History and the humanities. Period classrooms populated by the people who lived in them, including the court, the market and the trades, not only the famous.
• Law, ethics and argumentation. A platform that hears two sides of a dispute and issues an evidence-based ruling.
• Peace studies and social justice. Her work with the Gandhi-King Center for Nonviolence.

Faculty interested in adapting any of this to a research question in their own discipline are welcome to get in touch.

• Brings her AI expertise directly into graduate instruction.
• Teaches students to use generative AI under supervision and to evaluate what it produces, rather than treating it as a threat to academic integrity.
• Rationale: creation has become cheap, judgment has become the expensive part.
• Instructional approach draws on her independent work designing agent-based learning environments, described above.

Forbes Technology Council.

On AI and academia

• "The Pencil And The Agent: How AI Can Be Designed Into The Classroom, Not Banned Out," July 22, 2026. https://www.forbes.com/councils/forbestechcouncil/2026/07/22/the-pencil-and-the-agent-how-ai-can-be-designed-into-the-classroom-not-banned-out/
• "From Panic To Progress: Why Academia Must Embrace Generative AI," December 12, 2025. https://www.forbes.com/councils/forbestechcouncil/2025/12/12/from-panic-to-progress-why-academia-must-embrace-generative-ai/
• "Leveraging Emerging Tech To Free Academia From Traditional Grant Funding," October 6, 2025. https://www.forbes.com/councils/forbestechcouncil/2025/10/06/leveraging-emerging-tech-to-free-academia-from-traditional-grant-funding/

Directly relevant to scholarship and student work

• "Zwischenzug: Why Prompting Is Losing Its Opening Advantage," July 7, 2026. Why evaluation, not generation, is now the skill worth teaching. https://www.forbes.com/councils/forbestechcouncil/2026/07/07/zwischenzug-why-prompting-is-losing-its-opening-advantage/
• "When AI Is Too Nice: The Danger Of Aspirational Hallucinations," March 13, 2026. AI praising work it never read, which is the failure mode closest to student feedback. https://www.forbes.com/councils/forbestechcouncil/2026/03/13/when-ai-is-too-nice-the-danger-of-aspirational-hallucinations/
• "The Magic 8 Ball In The Boardroom: The Problem With Trusting Patterned Outputs As If They're Analysis," April 7, 2026. Output that looks like analysis without the reasoning underneath. https://www.forbes.com/councils/forbestechcouncil/2026/04/07/the-magic-8-ball-in-the-boardroom-the-problem-with-trusting-patterned-outputs-as-if-theyre-analysis/

Additional writing

• "The Three Bears Of AI: What Goldilocks Can Teach About Using AI Just Right" August 20, 2026.
  https://www.forbes.com/councils/forbestechcouncil/2026/08/20/the-three-bears-of-ai-what-goldilocks-can-teach-about-using-ai-just-right/
• "Give Your AI Workforce A Backpack," August 4, 2026. https://www.forbes.com/councils/forbestechcouncil/2026/08/04/give-your-ai-workforce-a-backpack/
• "The Flattery Algorithm: When Your AI Tool Is Managing You," June 23, 2026. https://www.forbes.com/councils/forbestechcouncil/2026/06/23/the-flattery-algorithm-when-your-ai-tool-is-managing-you/
• "Artificial Intelligence Takeover: Not With A Bang," May 13, 2026. https://www.forbes.com/councils/forbestechcouncil/2026/05/13/artificial-intelligence-takeover-not-with-a-bang/
• "The Single-Vendor Blind Spot: Why Your AI Stack Needs Cognitive Diversity," February 17, 2026. https://www.forbes.com/councils/forbestechcouncil/2026/02/17/the-single-vendor-blind-spot-why-your-ai-stack-needs-cognitive-diversity/
• "The 'AI SME' Trap: Why You Keep Hiring Engineers When You Need Architects," January 21, 2026. https://www.forbes.com/councils/forbestechcouncil/2026/01/21/the-ai-sme-trap-why-you-keep-hiring-engineers-when-you-need-architects

• Member, Forbes Technology Council, writing an ongoing series on artificial intelligence and human judgment.
• Leadership in Emerging Technology: Security, Strategy and Risk, Harvard Kennedy School Executive Education, 2025.
• Secretary, InfraGard National Board.
• U.S. State Department Speaker Program.
• Invited speaker on artificial intelligence, autonomous agents and AI risk.

Artificial intelligence; autonomous AI agents; agent design and evaluation; AI governance and risk; human-AI decision-making; AI in medical and graduate education; AI-augmented systematic literature review; model context protocol and AI interoperability; emerging technology strategy; pharmacology and toxicology; CBRN defense; crisis leadership and high-stakes decision-making; nonverbal communication and threat assessment.