AI & Intelligent Systems

What can intelligent systems reliably do—and what can be verified, controlled, or delegated?

Sponte studies the gap between what an intelligent system appears able to do and what can be established about its capability, reliability, objectives, and legitimate authority.

• Alignment, goal specification, and proxy failure

• Evaluation, construct validity, and predictive validity

• Prediction, planning, optimization, and control

• Verification, formal guarantees, and operating envelopes

• Capability limits and safety–capability tradeoffs

• Robustness under distribution shift and adversarial pressure

• Human oversight under capability gaps

• Agentic systems and long-horizon action

• Multi-agent emergence, collusion, cascades, and instability

• Interpretability where it supports decision-relevant verification

• Agent identity, provenance, commitments, and accountability

• Social AI, companionship, persuasion, and dependency

• Open and closed models and infrastructure

• The boundary between assistance, automation, and authority