How AI Will Help You Stream Your Favourite Shows Faster

While streaming video may seem like cutting-edge technology, given how central it has become to people’s digital lives, there is still tremendous room for improvement. William Erbey, a serial entrepreneur and the founder of six multi-billion dollar companies, asserts that the video streaming industry faces strong headwinds due to the way growing demand negatively impacts streaming quality.

With these challenges in mind, Erbey invested in System73, a multi-CDN platform which seeks to tackle internet congestion by using AI-augmented, peer-to-peer, tree-based networks to reduce congestion and overall bandwidth consumption.

As more people connect online to stream their favorite shows, the burden on streaming infrastructure grows. Erbey asserts that the consumer’s appetite for bandwidth will grow exponentially in the future due to higher quality video formats and an ever-growing customer base.

Border gateway protocol is how the internet routes traffic. Its origins date back to the early days of the internet. As the internet grows, the need for network routing protocols increases and as it becomes more dynamic, we need more dynamic protocols. Today, we have border gateway protocol as the primary routing protocol.

Border gateway protocol (BGP) recognizes that it is routers that slow down internet traffic. Therefore, BGP selects data paths that traverse the fewest number of routers. On one level, this is extremely logical. The flaw in this approach is that BGP does not know if any of the routers along the path are congested resulting in poor Quality of Experience.

The current solution to congest is hardware-focused, mostly through adding more CDNs (Content Delivery Networks). More CDNs will spread out the distribution of data increasing the probability
that data will traverse an uncongested route.

According to Joshua Tate, the CEO of System73, more feasible, scalable solutions will leverage AI and other emerging technologies.

"We leverage AI to intelligently forecast traffic and create delivery paths that avoid overly congested routers. The benefit of this approach is that it can pinpoint network and delivery path issues, fixing or alleviating strain on these routes.”

Another benefit of this approach is that it can offload a tremendous amount of traffic through peer-to-peer connections, which grow in proportion to the number of users on the network. In other words, our approach intelligently routes content without adding additional strain on already-overloaded routers, unlike the traditional approach where each additional smartphone, computer, or tablet streaming video eats away at the total available bandwidth.

While optimizing content delivery can help reduce the strain on internet infrastructure, another solution is to simply shrink the amount of data that needs to be transmitted. Tate points out that one possible way of reducing the size of video streams is through context-aware encoding.

The goal of context-aware encoding is to use machine learning and AI to present the best video quality with the least amount of data. AI does this by recognizing that not every frame or scene in a video needs the same level of compression, which is the most common approach today. For example, a night scene in a movie can undergo more compression because so much of the screen real estate is black, leading to a huge decrease in the data required to be sent.

“When applied at a massive scale across millions of viewers, even a small reduction in the amount of data used can lead to huge performance gains for all parties,” Tate added.

While still in an early stage, AI and machine-learning solutions have huge implications on how networks function.

The current, hardware-centric solution is a band-aid approach with diminishing returns. This solution is not just expensive, but it also
can’t keep up with peoples’ growing data usage. Instead, AI and other emerging technologies need to be used to find scalable, sustainable solutions to problems that will grow as our lives become more digital.