AI & softwareResearch findings
Same Behavior, Different Mechanism: Finite Experiments Can Certify Integrated Information
Original published 2026.11.06 03:55 KSTAdded to Blackdeck 2026.11.08 22:17 KST
What has changed?
Integrated information φ in stochastic networks of four and five units is identifiable from finite experimental data with known ground truth.
Key points
- Finite experiments can certify mechanism-level properties of AI systems.
- Identical behaviors can be distinguished via internal dynamics.
- φ exhibits sharp edges, changing 25–100% at critical boundaries and showing >20% misreporting in surrogate models.
Connect it to your work
Recurrent and feed-forward networks with identical input-output laws are distinguished by internal recordings.
What remains unproven
This research note is based on an abstract; check the original for test conditions and application scope.
Originals used for this note
Links open the original publisher in a new tab.