Quantum machine learning has become a catch-all term that encompasses everything from running classical algorithms on quantum hardware to genuinely quantum approaches. The confusion stems from the field's rapid evolution since Google and NASA established their Quantum Artificial Intelligence Lab in 2013. Today, QML spans applications that accelerate machine learning, classical algorithms inspired by quantum physics, and familiar ML workflows on quantum hardware.

At its core, quantum machine learning becomes quantum when quantum information serves as the computational substrate. This manifests in three key ways: data represented as quantum states using complex amplitudes and superposition, models built from parameterized quantum circuits that apply unitary transformations, and measurement processes that are probabilistic and destructive. Unlike classical ML where reading outputs is trivial, QML's learning process is fundamentally tied to measurement statistics and sampling noise.

Current quantum hardware remains noisy and resource-constrained, making general quantum advantage for ML tasks unlikely in the near term. Data loading and noise often dominate performance. However, QML remains valuable not for immediate speedups but for expanding our understanding of what learning means in quantum systems. The field forces us to reconsider foundational questions about learning from quantum data, noise effects on optimization, and model classes unique to quantum systems.