Joe Madden: The Next Killer App is AI Video

The world of general artificial intelligence started with words.

ChatGPT’s first release was remarkably fluent in answering questions of all kinds. Two year later, we now see applications using analysis of images. If you like the shoes in a photo, just circle them, and your AI Assistant can look up options to purchase them.

We moved from simple text on ChatGPT to image-based AI on smartphones in about 22 months. Within the next two years, we’ll likely have widespread AI applications that analyze video from your smartphone. But if a picture is worth a thousand words, a video is worth a thousand pictures—the complexity of video data to analyze is far greater than that of an image.

Predicting a person’s movements, or a car’s direction of travel, or a golf ball’s trajectory will require more information, and in some cases will surpass the processing power available on the phone. That’s when things will get interesting for our nation’s wireless networks.

In 2003, 2G and early 3G networks struggled to keep up with the explosion in Blackberry email traffic. In 2009, the iPhone 3GS was introduced and 3G network operators were surprised by the sheer weight of data downloaded. These stories are famous in Silicon Valley, as they gave the term “killer app” a double meaning:  it’s a new application that establishes a new key platform, but it can also crash the network.

AI video analytics will drive the next “killer app” for 5G devices—in both senses of the word. Apple, X, Meta, Google, and Samsung are all circling this technology space, working to establish a leading position. Some players are developing consumer concepts such as Augmented Reality (try on the outfit virtually!  See how this tattoo would look on your arm!)   Other players are developing enterprise applications (an app from H&R Block uses video to capture a shoe box full of receipts, filling in tax forms for you).   

The use of video in the AI engine will have a significant impact on the network. Instead of processing simpler images on the phone itself, video analytics will increasingly require data to be sent up to the cloud.  

Today’s 5G networks are not ready for the onslaught of heavy video uploads. Benchmarking with mobile operators and chip vendors, I recently modeled potential AI traffic placed on U.S. mobile networks and found that starting in 2028, networks won’t be able to keep up with demand.

The 5G network has been engineered for download-heavy applications like video streaming, not for a substantial rise in uploads needed for cloud-based AI. What’s worse, operators are stuck without clear access to new spectrum.

The mobile network crashing from Blackberry emails may seem quaint now, but if we don’t learn from history, we are doomed to repeat it. Thankfully, the wireless industry has time to adapt. But the government has a role in this too; we need Congress, the Administration, and the FCC to work together to clear and license spectrum for more network capacity. AI technology is moving quickly, so we need to shore up our nation’s wireless infrastructure just as quickly.

Joe Madden is the founder and chief analyst at Mobile Experts. Over the past 30 years, Madden has predicted digital predistortion, remote radio heads, small cells, 5G fixed wireless, as well as accurate growth forecasts for IoT, virtual RAN, and open RAN. He has worked as a radio frequency engineer and managed product lines in mobile infrastructure and in handset components. This Expert Opinion is exclusive to Broadband Breakfast.

Broadband Breakfast accepts commentary from informed observers of the broadband scene. Please send pieces to commentary@breakfast.media. The views expressed in Expert Opinion pieces do not necessarily reflect the views of Broadband Breakfast and Breakfast Media LLC.