PARIS—Some attendees at Upperside's 2019 MPLS + SDN + NFV World Congress in Paris—along with AI.net, a co-located conference on artificial intelligence (AI) and machine learning (ML) in networking—poked fun at the length of its name, but the conference has been going strong for 21 years, so kudos should be given to the organizers for continuing to adapt to changes in the industry. There's a core set of loyal followers among the more than 1,600 attendees on the ground, quite a few of whom have been attending for over 15 years.
There were a number of presentations on AI and ML, and some early work with ML has shown promising results at carriers. At the same time, the carriers are cautioning that we're still extremely early in the process. Colt, Orange and Telefonica all discussed ongoing efforts with AI and ML, but none were willing to speculate on when networks would be fully autonomous.
One of the areas of difficulty is in data gathering, data labeling, and training the models. ML models require large amounts of data to power the training, and the performance of the model ultimately depends on how it is trained. With carriers, even though the amount of data in aggregation across carriers is huge, it is hard to get legal permission to pull all that data together. The MEF, represented by Daniel Bar-Lev, discussed some of the challenges it had in asking for MEF carrier member data for running research trials for its members.
The other aspect is data labeling, data cleansing, and other preparations, which the carriers all agreed would take up at least 80% of the overall effort. The actual training and learning would be just a small fraction of the effort even though it's the most glamorous.
Autonomous networks and autonomous driving—Which comes first?
The conclusion of the carriers and AI/ML experts at the conference is that while early success in limited trials indicate that AI/ML has substantial promise, a fully autonomous network is a long ways away, and we'll likely have human oversight for the foreseeable future. Even though everyone agrees the scale of 5G makes it impossible for pure human operation, a fully AI/ML-run network with no guardrails appears unlikely at this point.
So we find ourselves in a world where cars might go fully autonomous before our networks do, which does seem strange. After all, if we're willing to trust the lives of people both within and outside the vehicle to AI/ML, isn't that so much more precious than any transport or mobile network? And isn't it ironic that as autonomous vehicles come to depend on 5G networks, we can't AI-automate those same networks?
I'd like to offer an alternate view though, one that was shared by some of the industry experts at the show, and that is that a massive network failure can impact more than a few lives. In a society that has become dependent on mobile and fixed networks as critical infrastructure, such a failure can prevent emergency services from getting to a location in time, can impact telemedicine, or cause sensor failures, throw control loops out of control, take down infrastructure at a national scale, causing much more mayhem and resulting in potentially more lives lost than an errant vehicle. And that is why we should take the AI/ML steps in the network carefully, sticking with safe use cases like revenue enhancement, churn reduction, predictive fault management and route optimization, and then going with an assistive model while we learn more about the impact of AI/ML.
SkyNet will rise, but not from networking
Many of us will remember the malicious sentient AI, "SkyNet," from the Terminator movies. Well, given the careful, measured steps networking carriers are advancing in AI/ML (for good reason), it looks like even though it has "Net" in its name, SkyNet is likely to be born in the sky, within the cloud players first, and not from us networking types.
[Author’s note: it’s very sad what happened to the Notre Dame Cathedral this week. Many of the conference participants passed by the beautiful structure last week on our way to dinner, not knowing it would be the last time we would see the original structure intact.]
Roy Chua is founder and principal at AvidThink, an independent research and advisory service formed in 2018 out of SDxCentral's research group. Prior to co-founding SDxCentral and running its research and product teams, Chua was a management consultant working with both Fortune 500 and startup technology companies on go-to-market and product consulting. As an early proponent of the software-defined infrastructure movement, Chua is a frequent speaker at technology events in the telco and cloud space and a regular contributor to major leading online publications. A graduate of UC Berkeley's electrical engineering and computer science program and MIT's Sloan School of Business, Chua has 20+ years of experience in telco and enterprise cloud computing, networking and security, including founding several Silicon Valley startups.
Industry Voices are opinion columns written by outside contributors—often industry experts or analysts—who are invited to the conversation by FierceWireless staff. They do not represent the opinions of FierceWireless.