Big shots are joining the Edge AI community

It’s no secret that Edge AI is a hot topic, arguably the main theme of last week's Embedded World in Nuremberg, Germany.

The biggest buzz on the conference floor? Edge Impulse's acquisition by Qualcomm! Would this mean that the technology is moving beyond its infancy and into industry adoption?

The Edge AI space is moving fast

“This year’s insane!” cried Zach Shelby, reacting to the sheer number of real Edge AI projects on display. While Edge Impulse may be ahead of the curve, it’s clear that the Edge AI space is moving fast. A year ago, a conversation about “the state of Edge AI” might have felt like an echo chamber where people could only talk about a handful of Edge AI applications which barely moved beyond its POC stage. This year is different. It seemed like every exhibitor at Embedded World had a real Edge AI story to share, primarily focused on computer vision, built without powerful, energy-draining and expensive GPUs. Instead, exhibitors showcased low-resolution cameras connected to inexpensive, battery-powered microcontrollers, running ultra-fast models with minimal latency and energy consumption

Real data, real sensors

The Edge AI community is flourishing, with more developers than ever gaining access to free AI models. But the real bottleneck to adoption isn’t the models, it’s the quality of the data.

Companies have been collecting data for decades, yet much of it can be completely thrown away. Many data sets are sampled, aggregated or lack clear context on what happened when that data was collected, making them useless for training AI models. “We need good data from real sensor as close to the real deployments as possible.” Shelby underscores the need for best practices to enable people to collect high quality data and more collaboration centered around data; we need to find better ways to crowdsource, collect and share data.

We don’t need more chatbots!

The threshold to getting started might seem low, but we’re mostly talking to the same group of embedded developers who’re actually putting Edge AI to use. “It’s an engineering party, hosted in an ivory tower” as Fabio Violante, CEO of Arduino put it. We need to open up, bringing in more creativity.

“What’s the hello world of AI?” asked Brandon Satrom from Blues. “It’s chatbots!” And if there’s anything we don’t need more of, it’s chatbots, which won’t work on MCU-based devices anyway. Put that Raspberry Pi away and stop running Ollama. Look beyond chatbots and start implementing sensor-driven applications which solve real problems. As Violante nicely put it “Edge AI is there to solve problems, not the other way around”

Community, unite!

The industry is at a turning point with big shots investing heavily. Now is the time to define best practices, share experiences and crowdsource high quality data. Could this be the moment for the EDGE AI FOUNDATION to take the lead?

Let’s stop messing around and start solving real problems using a technology that has proven itself.

Laurens Slats

Based on the conversation with Pete Bernard, Zach Shelby, Brandon Satrom, and Fabio Violante.