## TL;DR
AI and blockchain converge at three frontiers: running AI services via smart contracts, using crypto-economic incentives to build decentralized compute networks that rival cloud providers in cost, and applying AI to detect fraud in blockchain systems. Together they form the infrastructure for trustless, decentralized intelligence.

## Core Explanation
Three convergence domains: (1) AI on blockchain -- smart contracts execute AI inference. Challenge: neural networks are too large for on-chain execution. Solutions: off-chain computation with on-chain verification via ZK-proofs (ZKML), oracle-based (off-chain AI -> on-chain result), and specialized lightweight models; (2) Blockchain for AI -- DePIN networks (Gensyn, Akash, Render, io.net) aggregate idle GPUs globally. Crypto-economic incentives: contributors earn tokens for verified computation. Benefits: democratizes GPU access (no cloud monopoly), reduces costs 40-70%, censorship-resistant (no single point of control); (3) AI for blockchain -- detecting DeFi exploits (MEV detection, rug pull prediction), smart contract vulnerability detection, and AML compliance (GNN-based transaction graph analysis).

## Detailed Analysis
ZKML (Zero-Knowledge Machine Learning): using ZK-SNARKs to prove that an ML inference was computed correctly without revealing the model weights or input data. This enables private AI services -- you can use an AI model without trusting the provider, because a ZK proof cryptographically guarantees correct computation. Current state: feasible for small models (<100M params), prohibitively expensive for LLMs (proving GPT-4 inference would cost thousands of dollars). ezkl and Modulus Labs provide tooling. Gensyn: task distribution (split training across nodes), verification (redundant computation + spot-checking), incentive (token rewards for verified work). The blockchain coordinates task assignment and payment without centralized orchestration. Applications: (1) Decentralized model marketplaces (buy/sell model access via smart contracts); (2) Data DAOs -- collectively own and monetize datasets; (3) Federated learning with blockchain coordination.