Apple to Replace Core ML with 'Core AI'

Core ML, introduced at WWDC 2017, marked Apple's initial strategy for on-device machine learning, prioritizing user privacy by processing data locally. This framework allowed developers to integrate trained models into their apps, running on the CPU, GPU, or the Apple Neural Engine (ANE) for optimized performance. The ANE, a dedicated neural processing unit, first appeared in the A11 Bionic chip in 2017, accelerating AI tasks like Face ID. Subsequent updates, Core ML 2 in 2018 and Core ML 3 in 2019, expanded capabilities by enabling on-device model training and introducing tools like Create ML for easier custom model creation. This evolution underscored a consistent focus on a hybrid AI model that balances on-device processing for speed and privacy with secure cloud computations for more complex tasks. This strategy contrasts with the cloud-first approaches of competitors like Google and OpenAI. The shift to "Core AI" suggests a deeper integration of artificial intelligence across the operating system, moving beyond a developer-only framework. This aligns with the introduction of "Apple Intelligence," a system-wide AI layer for features like summarization and image generation, which heavily leverages on-device processing. The A17 Pro's Neural Engine, capable of 35 trillion operations per second, provides the necessary hardware foundation for these advanced, on-device AI experiences. The reported native support for third-party AI models in Core AI addresses a previous limitation and reflects a broader industry trend towards interoperability, exemplified by standards like ONNX. This move could simplify the process for developers accustomed to frameworks like TensorFlow and PyTorch to deploy their models within Apple's ecosystem, which has been a more complex task in the past. The integration with the Memory Compute Platform (MCP) points to a new architecture for how AI agents interact with applications and user data. MCP acts as a standardized protocol, a "USB-C for AI," enabling language models to connect with and control various tools and data sources. An open-source Apple Native Tools MCP Server already allows AI assistants to interact with macOS applications like Messages, Notes, and Calendar. This evolution suggests a future where AI workloads can be dynamically allocated between the device, edge, and a "Private Cloud Compute" environment. This hybrid approach ensures that sensitive data remains on the user's device while leveraging more powerful server-side models when necessary, without compromising Apple's core privacy principles.