Beyond the Black Box
IEEE TCSS — accepted
The foundational programmable-AI paper: lifting neural outputs into inspectable symbolic rules, then lowering them back as verifiable constraints on what the system may conclude.
For defense analysts, information-operations planners, and national-security research teams
Decisions that hold up under adversarial scrutiny.
Adversaries spoof signals and data shifts daily. GoPX produces sealed explanations — tamper-evident records that tie each conclusion to source evidence, authoritative doctrine, and the exact rules applied — so results can be audited, challenged, reproduced, and trusted or rejected quickly.
Why black-box AI fails here
Contested environments break confident answers.
- 01
The inputs are adversarial.
Signals are spoofed and narratives are seeded deliberately. A system that cannot show where its conclusion came from cannot show whether it was manipulated.
- 02
The ground shifts daily.
Yesterday's model behavior is not evidence about today's. Conclusions must be bound to the data, assumptions, and model versions active when they were computed — or they cannot be re-examined.
- 03
Every decision faces review.
Operational decision-makers need speed plus decisions that are defensible under scrutiny. A fluent answer with no reasoning trail becomes a liability the moment anyone asks how it was reached.
The sealed explanation
Every conclusion carries a tamper-evident receipt.
The output of the reasoning process is not a paragraph — it is a sealed record tracing evidence to applied rules to intermediate claims to conclusion, bound to the references and model versions active at computation time.
- 01
Ground in evidence
Every input is registered with its source and provenance. Nothing enters the trace without a citable origin.
- 02
Bind to doctrine
Conclusions attach by explicit reference to the authoritative knowledge that governs them — doctrine, policy, and standing rules.
- 03
Apply explicit rules
Reasoning runs through inspectable rules, not free-form generation. The rule applied is part of the record.
- 04
Pin every version
Model versions, assumptions, and reference states are pinned into the record, so the result can be reproduced and re-examined later.
- 05
Seal or flag
If every link holds, the conclusion is sealed. If any link fails — missing provenance, conflicting references, broken rule application — the seal breaks and the output is flagged for human review.
If any link fails, the seal breaks.
A sealed explanation makes the AI output a primary mission artifact: auditable, challengeable, reproducible. The system never converts uncertainty into confidence — it grounds the conclusion or flags the gap. That is the discipline adversarial conditions demand.
Research lineage
Built on funded, peer-reviewed cognitive-defense research.
The programmable-AI foundations that run GoPX's legal and healthcare surfaces began in defense-funded social intelligence research at ASU's CIPS-AI Lab.
NARRA-SCALE
IEEE ICTAI 2025
Narrative detection and scaling in social networks — mapping users and messaging across contested issues with an explainable structure rather than a black-box score.
Two decades of funded work
DoD Minerva · ARTIS MAGI · NSF
Sociocultural modeling and multilingual social-intelligence research, recognized with the 2011 HSCB Exceptional Scientific Achievement Award from the Office of the Assistant Secretary of Defense.
Current status
GoPX is engaging research partners, mission teams, and program offices on cognitive-defense decision support. This is expert-supervised analytical infrastructure — every output remains subject to human review and command authority.
Discuss a defense engagementBuild AI decisions that can survive review.
A 45-minute session. We run your contracts, policies, or regulations through GoPX live and return structured logic, a decision walkthrough, and an honest read on fit. No pitch deck.
- Run on your documents, not a generic demo
- Structured logic + decision walkthrough
- Honest read on fit — no pitch deck