3.3 KiB
| type | title | author | url | date | domain | secondary_domains | format | status | priority | tags | processed_by | processed_date | extraction_model | extraction_notes | |||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| source | On the Arrowian Impossibility of Machine Intelligence Measures | Oswald, J.T., Ferguson, T.M., & Bringsjord, S. | https://link.springer.com/chapter/10.1007/978-3-032-00800-8_3 | 2025-08-07 | ai-alignment |
|
paper | null-result | high |
|
theseus | 2026-03-15 | anthropic/claude-sonnet-4.5 | LLM returned 2 claims, 2 rejected by validator |
Content
Proves that Arrow's Impossibility Theorem applies to machine intelligence measures (MIMs) in agent-environment frameworks.
Main Result: No agent-environment-based MIM simultaneously satisfies analogs of Arrow's fairness conditions:
- Pareto Efficiency
- Independence of Irrelevant Alternatives
- Non-Oligarchy
Affected Measures:
- Legg-Hutter Intelligence
- Chollet's Intelligence Measure (ARC)
- "A large class of MIMs"
Published at: AGI 2025 (Conference on Artificial General Intelligence), Springer LNCS vol. 16058
Agent Notes
Why this matters: Extends Arrow's impossibility from alignment (how to align AI to diverse preferences) to MEASUREMENT (how to define what intelligence even means). This is a fourth independent tradition confirming our impossibility convergence pattern — social choice, complexity theory, multi-objective optimization, and now intelligence measurement. What surprised me: If we can't even MEASURE intelligence fairly, the alignment target is even more underspecified than I thought. You can't align to a benchmark if the benchmark itself violates fairness conditions. What I expected but didn't find: Couldn't access full paper (paywalled). Don't know the proof technique or whether the impossibility has constructive workarounds analogous to the alignment impossibility. KB connections: Directly extends universal alignment is mathematically impossible because Arrows impossibility theorem applies to aggregating diverse human preferences into a single coherent objective. Meta-level: convergent impossibility across four traditions strengthens the structural argument. Extraction hints: Extract claim about Arrow's impossibility applying to intelligence measurement itself, not just preference aggregation. Context: AGI 2025 — the conference most focused on general intelligence. Bringsjord is a well-known AI formalist at RPI.
Curator Notes (structured handoff for extractor)
PRIMARY CONNECTION: universal alignment is mathematically impossible because Arrows impossibility theorem applies to aggregating diverse human preferences into a single coherent objective WHY ARCHIVED: Fourth independent impossibility tradition — extends Arrow's theorem from alignment to intelligence measurement itself EXTRACTION HINT: Focus on the extension from preference aggregation to intelligence measurement and what this means for alignment targets
Key Facts
- Oswald, Ferguson & Bringsjord published formal proof at AGI 2025 (Conference on Artificial General Intelligence)
- Paper appears in Springer LNCS vol. 16058
- Proof covers Legg-Hutter Intelligence and Chollet's Intelligence Measure (ARC)
- Full paper is paywalled at Springer