Consensus is (almost) never what you need
The article argues for a paradigm shift in decentralized systems: moving from deterministic consensus models to probabilistic approaches. The author contends that consensus mechanisms and time synchronization are limiting innovation in distributed systems.
Problems with Consensus
- Natural systems (ant colonies, neural networks) function without explicit consensus
- Consensus algorithms provide an “illusion of certainty” with probabilistic guarantees
- Requires bounded systems, limiting flexibility and scalability
Problems with Time Synchronization
- Impossible to achieve precise agreement across many entities
- Natural systems work without universally agreed time
- Attempts are inherently bounded and imprecise
Proposed Probabilistic Alternatives
- Range-Based Search: Query range of nodes using distance-based metrics, monitor response consistency, detect malicious data naturally
- Dynamic Validation: Validate parent transactions based on complexity without global consensus
Benefits of Probabilistic Approach
- Higher concurrency
- Faster transaction processing
- Natural conflict resolution
- More aligned with natural/evolutionary processes The author advocates moving beyond consensus limitations to unlock innovation in decentralized, robust, and scalable systems.