Make your own smart home that doesn’t send everything to the cloud

Owen Williams
Photo: NVIDIA Jetson Nano 2GB Developer Kit/NVIDIA

Artificial intelligence has been heralded as the next wave of computing for years, but learning or even tinkering with it has required access to expensive hardware with powerful GPUs capable of crunching massive data sets.

That’s starting to change with the debut of cheap all-in-one A.I. computers from companies like Nvidia, which introduced its latest Jetson Nano A.I. developer kit this week — for just $59. The Jetson is a full computer in a tiny package, similar to a Raspberry Pi, that allows hacking on projects or learning from home, while making A.I. accessible to a much broader audience.

The debut of the Raspberry Pi in 2012 was a watershed moment for computing because it made computers accessible in a tiny, all-in-one package, for just $35. It meant that hobbyists like me could buy a full computer and hack around with ideas, like building a magic mirror or DIY smart screen. Before the Raspberry Pi, there were few ways to tinker on these kinds of ideas without investing hundreds of dollars in specialized hardware or cannibalizing an old laptop for the job.

Cheap, integrated computers like the Jetson and Google’s “Coral” development board are poised to do the same for A.I. Oversimplifying it, A.I. (which is often referred to as machine learning) is basically processing vast quantities of data to “teach” a computer to find a pattern in that data. Doing so relies on GPUs, which are adept at processing large sets of data quickly and creating inferences.

The problem, until recently, has been that the GPUs required to perform machine learning tasks are expensive, large, and require building an entire computer. That meant that those interested in the field generally had a choice: build a powerful desktop computer, or pay Google, Amazon, or Microsoft for their cloud platforms to do it for you.

While these cloud platforms are useful, the raw processing power required for A.I. meant that it was difficult to build in smarts to home devices without depending on an internet connection to send data to that cloud. Ring’s smart doorbells, for example, perform facial recognition in the cloud because there’s historically been no reasonable, affordable way to put the smarts required into a device that’s glued to your door.

Nvidia’s Jetson computer, and others like it, change the game by making tiny, powerful chips that let people access the power of A.I. without needing to rely on the power of the cloud. Instead of building a smart doorbell that relies on a cloud server to perform facial recognition, Jetson makes it simple to run that algorithm locally, without sending data to the cloud for processing. Want to build a self-driving robot that doesn’t crash into objects for under $100? That’s now a reality.

Cheap A.I.-enabled computers level the playing field.

This is an enormous leap because it means it’s now possible to build privacy-first smart devices that don’t rely on the cloud to crunch petabytes of your personal data in order to be useful. It would allow device makers like Roomba, for example, to build vacuum cleaners that learn about your home’s layout without sending that data to the cloud, or companies like Ubiquiti to build smart doorbells that detect faces in footage without sending it outside your home. More importantly, it allows you to do the same without spending thousands of dollars.

The process of crunching this data on-device is referred to as “Edge” A.I., which means that the processing is done near the source of data. While this technology has been used in smartphones in recent years for things like processing “Hey Siri” on the iPhone and Google’s Call Screen technology, it hasn’t been as common in other smart devices because it’s cheaper to just fire that data off to the cloud than to include an expensive custom chip to process information locally. That comes at the cost of your privacy — and a reliance on the company keeping those servers online so your device continues to function.

Dependence on the cloud gives companies like Ring, which is owned by Amazon, an advantage over any new entrants to the space because it has access to practically unlimited resources from its parent company to build A.I.-enabled devices for free. Cheap A.I.-enabled computers level the playing field: Anyone can now play with fully fledged A.I. for a onetime cost of $59 and build devices that compete with the industry’s giants without paying endless cash to the cloud platforms to get access to the smarts they require.

More importantly, devices like Jetson open the floodgates for a new generation to play around with their A.I.-powered ideas at home, building their own voice assistants or self-flying drones without requiring an internet connection or an expensive investment in high-end hardware.

Like the Raspberry Pi leveled the playing field for hobbyists and students a few short years ago, Nvidia’s new Jetson computer will usher in a new wave of A.I.-based projects that anyone can build for themselves simply because it’s so accessible — and some of those ideas might turn into the next big thing.