Experienced AI Credibility Researcher
Eric Gladstone
Behavioral + computational scientist
I study how information is distributed, interpreted, and transformed through networks, and how communication structure shapes judgment, coordination, and collective behavior across human, organizational, and AI systems.
The behavior of an interacting system is not determined by its components alone.
Credibility lens
Controller
Predictability
experiment operates as specified
Reproducibility
portable, rerunnable experimental code
Transparency
complete interaction records
Specified in code
Representative communication structures
Credibility may require evidence at the level of the interacting system, not only the individual model. Relevant properties may include information flow, dependence, failure propagation, provenance, traceability, and the conditions under which these alter system behavior.
The empirical question is which system-level properties, if any, remain informative across changes in models, tasks, capability levels, and operating conditions.
Properties that prove sufficiently stable may support reproducible tests, common evaluation criteria, rubrics, benchmarks, and assurance procedures, allowing credibility evidence to accumulate across studies and model generations rather than being reconstructed for each new system.
Presenting