William Erbey Says We Need to Use AI for Better Streaming Video
A growing number of influential investors and tech leaders have started to utilize recent and rapid advances in AI for consumer-level applications. One such individual is Bill Erbey, a serial entrepreneur who has founded 6 multi-billion dollar cap companies.
I had a chance to speak with Erbey who has a background in informatics and is the founder of System73, an intelligent content distribution solution. Erbey has recognized that streaming video’s convenience and universal presence have become its biggest problem, adversely affecting the number of concurrent viewers that can view a quality stream.
Streaming video may seem like a concept on the cutting edge, given its central place in our digital existence, but there is nearly infinite room for improvement. “With a compound annual growth rate of nearly 20%, streaming video faces strong headwinds. Most notable is how collective demand affects streaming quality. As more people connect to watch and increasingly expect video quality to keep up with the latest formats like 4k, we place a significant burden on streaming infrastructure,” Erbey says.
“Viewers are largely unaware of this, but their immense appetite for more bandwidth may ultimately negatively impact end-user experience,” he said.
Scaling is Where Smart AI Shines
“There are many applications for artificial intelligence to help scale streaming video while maintaining high video quality,’’ according to Erbey.
The “lowest-hanging fruit” for AI is context-aware encoding. Not every scene in a video clip or streamed TV show requires the same level of compression. For example, though it is preferable to have the least compression possible (and therefore the highest quality), some content can be compressed more than other parts without sacrificing the user experience. Companies which depend on the attention span of their users, like Netflix, would find it otherwise impossible to identify the degree of compression that can be applied without context-aware AI.
Erbey points out that the next generation of Artificial Intelligence assisted scaling will be AI augmented tree-based peer-to-peer networks that support smarter content distribution and reduce strain at peak times without sacrificing quality. Currently, the internet’s routers which direct traffic operate largely independent of each other. Erbey notes that “this disorganization creates congestion. Instead of cooperating to create better pathways, traffic hubs operate without considering their counterparts, leading to overlap, waste and heavier congestion.”
Today’s internet is largely hardware centric. This approach can only take us so far. Simply adding more CDNs to push more data through congested routers has its limitations. Instead, AI and other emerging technologies should be embraced to find creative and sustainable solutions to problems that will only grow worse as our lives continue to migrate online.
As opposed to the diminishing returns of hardware-based solutions, AI augmented tree-based networks can scale exponentially. Each smartphone, tablet and computer streaming video is additive to the overlay network, increasing capacity and quality. The network only gets stronger with increased utilization. Furthermore, instead of randomly routing content and hoping for the best, the technology can predict and forecast traffic to make better decisions about the optimal pathways for delivery, by passing congested routers and links. AI can pinpoint problems within networks and across delivery pathways to both fix and prevent problems from occurring.
An important secondary benefit of AI augmented tree-based networks is their ability to significantly off-load traffic. And, the proportion of traffic off-loaded expands to 90+% as the number of concurrent viewers grows, further enhancing scalability.
On the horizon William Erbey predicts that AI supported tree-based networks will be used in the open internet to forecast traffic and CDN capacity utilization so that traffic can be proactively re-routed to underutilized CDNs. The spray and pray method of adding more CDNs to manage capacity spikes will be a thing of the past. Today, the standard is to switch CDNs after they become congested causing interruptions resulting in poor video quality for affected users. With AI, congestion can be anticipated so that viewer traffic can be optimized before failure or loss of quality occurs.
Overall, it will be Artificial Intelligence and software, not hardware, that will enable the growth of video streaming to continue with ever increasing resolution and without annoying rebuffering. Someday, large scale live and linear streaming with tens, if not, hundreds of millions of concurrent viewers will consume 4K events with the reliability of today’s cable service.
