## TL;DR
AI and quantum computing form a symbiosis: AI optimizes quantum circuits and corrects qubit errors, while quantum computers may eventually accelerate certain ML computations. DeepMind's AlphaQubit uses RL for quantum error correction -- a critical capability for building fault-tolerant quantum computers.

## Core Explanation
AI for quantum: (1) Quantum error correction -- qubits are fragile (decoherence). Surface codes encode logical qubits across many physical qubits, but decoding (identifying which physical qubits errored) is hard. AI decoders (AlphaQubit, neural network decoders) outperform classical minimum-weight perfect matching; (2) Circuit compilation -- ML reduces quantum circuit depth, maps logical to physical qubits respecting connectivity constraints; (3) Algorithm discovery -- AI proposes novel quantum circuits for VQE, QAOA, and quantum chemistry.

## Detailed Analysis
AlphaQubit (DeepMind, Nature 2024): RL-based decoding of surface codes. Trained on simulated quantum error data. Achieves lower logical error rate than prior decoders at same physical error rate. This directly translates to fewer physical qubits needed per logical qubit -- critical for scaling. Quantum ML: NISQ devices (100-1000 qubits, noisy) have not demonstrated practical advantage over classical ML for any real-world task. The "quantum winter" concern (2023-2025): venture investment shifted from quantum software to hardware. Google Willow (2024): 105 qubits, demonstrated exponential error reduction with increased qubit count -- a milestone toward fault-tolerance. IBM roadmap: 100K+ qubits by 2033. Timeline: practical QML advantage likely requires >1000 logical qubits, expected post-2035.