Inside Siri: Luc Julia’s Journey from HCI to AI’s Future

Playing a PC

As a child, Luc Julia first played with electronic devices and, around age 15, discovered his first personal computer. He began programming on it using assembly language, a habit that never left him.

Human‑Computer Interaction

He became intrigued by whether computers could converse with humans, leading him to explore HCI and signal processing to enable machines to understand spoken signals. This work evolved into a focus on voice, gesture, and ultimately artificial intelligence.

After earning a PhD in France and completing post‑doctoral research at MIT and Stanford, his team decided to spin out a company, giving rise to SRI (the Stanford Research Institute) as a derivative factory.

He later investigated multimodal interfaces, aiming to combine various modalities—starting with voice because microphones were easier to integrate than cameras.

He notes that speech recognition remains challenging, especially moving from word recognition to understanding meaning, i.e., natural language.

Internet as a Playground

At SRI, they created the Computer‑Human Interaction Center (CHIC), an applied research lab that produced many products. By 1999‑2000, the internet was still early, and testing products online with real users was straightforward.

The Birth of Siri

In the late 1990s, the idea of a voice assistant emerged when searching the web by voice seemed promising. After years of research, Adam Cheyer founded Siri in 2007, coinciding with the launch of the iPhone, which provided a high‑quality microphone for voice interaction.

Building Siri’s Backend

When Cheyer approached Luc Julia, they rebuilt Siri’s backend from scratch, scaling from a modest user base of 150,000 to hundreds of millions. The system required high availability, scalability, and performance, comprising roughly 17 services at launch and growing over time.

The backend involved purchasing servers, constructing data centers, and integrating numerous services such as iTunes storage and extensive compute resources.

Introducing Agile at HP

At Hewlett‑Packard, Luc Julia led a team of 250 engineers and introduced agile development methods to replace the traditional waterfall approach, which required six‑month advance planning. He also helped develop ePrint, the first network‑connected printer, selling about 80 million units in its first year.

Opening Samsung’s AI Lab in Paris

In June 2018, Samsung opened an AI laboratory in Paris, focusing on the Internet‑of‑Things cloud that connects Samsung and non‑Samsung devices, a highly heterogeneous environment requiring five years of work to ensure interoperability.

He praises French engineers for their quality and the dynamic startup ecosystem that encourages flexibility and innovation.

AI’s Ecological Cost

AI training often throws massive amounts of data at CPUs, GPUs, or other processing units, consuming significant storage and compute resources. The hidden energy consumption in data centers is rarely noticed, especially when using public clouds like AWS or Azure.

He warns that the ease of scaling AI can obscure its environmental impact, calling for greater awareness and regulation.

The Future of AI

While early voice assistants appeared in the 1990s, the future will see many specialized assistants across domains such as healthcare, transportation, and genomics, rather than a single general AI.

Emerging technologies like quantum computing and biocomputing are still in their infancy, and building truly human‑level intelligence may require breakthroughs beyond current mathematical approaches.