Americas

  • United States

Asia

What is Darwin AI and how could Apple use its new tech?

news analysis
Mar 15, 20246 mins
AppleArtificial IntelligenceGenerative AI

What's interesting about the purchase of Canada’s Darwin AI is that the company was focused on machine vision intelligence, smart manufacturing, improved machine learning efficiency and edge-based intelligence.

Apple park with Siri logo

Apple has quietly acquired yet another AI startup, Canada’s Darwin AI, a company  focused on machine vision intelligence, smart manufacturing, improved machine learning efficiency and edge-based intelligence.

All of these seem critical to Apple’s future plans.

Who is Darwin AI?

As reported by Bloomberg, several members of the Darwin AI team have now joined Apple. Ostensibly a visual quality inspection company, Darwin AI was developed to provide electronics manufacturers with a tool to improve product quality and production efficiency.

A note on Arm.com suggests DarwinAI’s patented Explainable AI (XAI) platform was in use across a number of Fortune 500 companies, including Audi, BMW, Honeywell, and Arm. That site also explains that the technology was used in the development of Covid-Net, an open source system to diagnose COVID-19 via chest x-rays. (The company was also declared a cool vendor in Gartner’s October 2019 Cool Vendors in Enterprise AI Governance and Ethical Response report.

While Darwin AI’s Twitter/X and YouTube feeds have been deleted,  a four-year old video providing some insight into how the tech works is still available via Arm Software Developers/YouTube. Still, four years is a very, very long time in AI tech innovation — particularly for a company whose co-founder, Alexander Wong, was in 2021 ranked as one of the world’s leading scientists by Stanford University.

A tool for smart industries

In July 2023, CEO Sheldon Fernandez explained to SMT007 Magazine (p.16) that the company was, “leveraging our latest tools in deep learning to do visual inspection of PCBs in a compelling way.”

There are challenges to systems of this kind. Because they lean deeply on imagery, “The AI can be great, but if the images are blurry, they’re not high resolution, or it doesn’t work with their workflow, it won’t be an effective product,” he said.

Fernandez pointed out that the AI his company developed could be trained to have a full understanding of a new PCB design in a couple of minutes. It can then take its seat on the production line, accurately monitoring product quality and detecting faults in real time.

He also noted that the system becomes more accurate over time: “Sometimes we’ll bring up a system for a client that’s 93% accurate, but within a couple of months we’re at 97%,” he said.

At the time, he argued that true robotic factories are in reach, saying the industry now anticipates AGI (Artificial General Intelligence) will appear around 2026. (Think ChatGPT on steroids.)

It is also worth noting that Darwin AI’s tech also has implications for healthcare tech, with funder Alexander Wong being the Canada Research Chair in Medical Imaging and Artificial Intelligence. So, there’s a sea of possibilities to unlock in that as well.

Robots that teach each other

Another strand to the company’s bow was revealed in an August 2023 Manufacturing Automation piece, which reported Darwin AI’s involvement in “federated learning” research.

This explored how robots can learn from each other without sharing their training data. The report’s focus was on how robots could federate their learning without leaking company secrets. In incredibly over-simplistic terms, it works a little like this:

  • Robots equipped with cameras learn to recognize different items and use a suitable picking method.
  • With thousands of items to learn, building enough data for the AI is time-consuming.
  • By working with lots of robots across multiple organizations, the data pool expands exponentially,
  • As each robot learns, this information is share with a central machine learning server.
  • That data can then be shared across robots across multiple companies, with no secrets revealed.

Why this matters is that it in theory enables much faster creation of effective AI models while maintaining data privacy and security — which meshes well with Apple’s overall approach to this technology.

“By using distributed learning, also known as federated learning, we are able to strike the right balance between having a wide range of data available and keeping data secure in the industrial environment,” said Jonathan Auberle, of the Institute of Materials Handling and Logistics (IFL) at the Karlsruhe Institute of Technology (KIT). KIT led the research.

Data that flies? Send less data

Yet another facet to Darwin AI’s work lay in developing an AI fit for edge computing using 5G. A report from 5G Innovation Lab explained why networks are so important,

“In the Industry 4.0-enabled smart factory, computers are connected and talk to each other. Lines are fully automated and controlled by robots. Autonomous vehicles transport goods to and from work cells.”

All of this generates data that must be moved from the edge where it is created to the central server where it is analyzed.

Indeed, “the news comes as use of AI in manufacturing begins to proliferate, even as mobile networks  build wireless technologies (network slicing, SD-WAN, etc) designed to support large numbers of connected devices across private networks built for resilience against hacking and also to deliver high QoS levels,” as I noted elsewhere.

What Apple might do

With so much focus on manufacturing on the Darwin AI resume, it seems plausible Apple might deploy some of the tech on its own production lines. Doing so could help the company on its quest to build a circular manufacturing system and drive efficiencies on its increasingly global network of iPhone production lines.

But the capacity to build effective machine-learning models using data in a private way will also be useful for a company with over one billion actively used devices in circulation. And the ability to minimize the data that must be transported on the network also has significant effect, particularly as Apple is thought to be developing AI models that perform almost entirely at the edge.

The capacity to share insights between machines while maintaining user privacy and security might help generate the kind of high-quality data the very best AI systems are going to need. It’s hard not to connect the dots and it’s really hard not to see a synergy between these systems and Vision Pro.

After all, with automated systems ready to spot faults on the production line, it’s an obvious next move for a chief engineer to use the Vision Pro to take a closer look before despatching any incident response team.

That’s just one of many implications, of course. But as the torrent of AI-related Apple news turns into a flood, you’d better begin placing your bets on exciting AI-related news announcements as Apple prepares for WWDC. After all, the internet of things already exists — now they just need to work better together.

Please follow me on Mastodon, or join me in the AppleHolic’s bar & grill and Apple Discussions groups on MeWe.