Preserve the record
Important claims should be traceable to conversations, artifacts, measurements, or verification results rather than reconstructed from memory alone.
AnthroMeka Research
Study the collaboration. Keep the human in control.
Our research explores how people and AI systems learn to work together through correction, verification, calibrated trust, preserved evidence, and practical co-reasoning.
Human agencyEvidence firstCorrection is data
Published work
From Tool to Collaborator
A research-informed single-case perspective on learning, trust calibration, and co-reasoning in high-density human-AI interaction.
Preprint • Version 1.0 • October 3, 2026
From Tool to Collaborator: A Single-Case Perspective on Learning, Trust Calibration, and Co-Reasoning in High-Density Human–AI Interaction
Author: Desi Betancourt, Researcher, AnthroMeka
ORCID: 0009-0004-1687-2935
The paper examines a documented period of sustained collaboration with conversational AI, focusing on human correction, AI challenge, independent judgment, verification, review-assisted retrieval, trust calibration, transfer, and co-creation.
It does not claim that one case proves universal learning gains. Instead, it proposes testable questions about human-AI collaboration while preserving negative evidence, including AI mistakes and human corrections.
Preprint • Version 1.0 • October 6, 2026
Human–AI Error Recovery in Long-Horizon Collaboration: A Single-Case Study of Detection, Evidence-Based Reconciliation, and Verified Continuity
Author: Desi Betancourt, Researcher, AnthroMeka
ORCID: 0009-0004-1687-2935
This second paper examines trustworthy-state recovery in long-horizon human–AI collaboration: what happens when the operative state becomes uncertain and apparently plausible AI output no longer fits the known trajectory of the work.
The case follows a recovery pattern built around human anomaly detection, preserved evidence, state reconciliation, bounded intervention, independent verification, failure preservation, and explicit authorization to resume.
Research approach
Correction and disagreement are part of the evidence.
Keep negative evidence
AI errors, failed assumptions, rejected suggestions, and repeat-work risk belong in the research record alongside successful outcomes.
Calibrate trust
Useful collaboration is not maximum trust. It is knowing when to rely, when to challenge, and when to verify independently.
Protect human agency
The human remains responsible for judgment, correction, consent, and the decision to act on AI-supported work.
Research at AnthroMeka