Apple Computer

Contribution to AI

Most of the field's history happened in laboratories; the contribution here was distribution, and the constraints that came with it. The Newton's handwriting recogniser in 1993 was one of the first attempts to sell statistical pattern recognition to ordinary consumers, and its public failure was an early lesson in how little error a shipped product is forgiven. Siri mattered more. Bought in April 2010 from a company spun out of SRI's DARPA-funded CALO project, it arrived on the iPhone 4S in October 2011 as a built-in feature rather than an app, and within a few years every major platform had an assistant of its own; talking to a machine stopped being remarkable. The deeper change was architectural. A commitment to keeping user data on the handset pushed inference to the edge: the local differential privacy rolled out in 2016 was among the first large-scale commercial deployments of that technique, Core ML in 2017 let outside developers run trained models on a phone, and the A11's Neural Engine that September put dedicated neural-network silicon into a mass-market handset, as every A- and M-series chip has since. Efficiency thereby acquired a commercial customer — quantisation, distillation and small models had somewhere to go. Apple Intelligence kept the split, pairing a roughly three-billion-parameter on-device model with a server tier designed to be unable to retain what it processes.

Drafted with AI and edited by hand (claude-opus-5, reviewed 2026-08).