What’s Next is a biweekly conversation between SDxCentral CEO Matt Palmer and a senior-level executive from the technology industry. In each video, Matt has an informal but in-depth video chat with a fellow thought leader to uncover what the future holds for the enterprise IT and telecom markets — the hook is each guest is a long-term acquaintance of Matt’s, so expect a lively conversation.

This time out, Palmer spoke with Andrew Coward, GM of software-defined networking at IBM. Coward has 25 years of experience in building, growing and managing product, presales, sales and marketing groups from start-ups to multi-billion dollar operations.

Editor’s note: The following is a summary of what Coward shared in their conversation, edited for length. To hear the full conversation, be sure to watch the video.

Telco challenges

Andrew Coward: There's a huge amount of excitement and hype around what AI (artificial intelligence) can do. And the power of AI to long-term reduce the workforce in telco through to improving efficiency of networks you name it. There's there's some kind of AI thing that's gonna help, but I think telcos are kind of challenged in how they're going to get there, because there's some kind of fundamentals not yet in place. So I'm describing AI as a three-legged stool from a telco perspective. And one of those legs is, they don't have necessarily the right data today. So if you're going to make decisions about something about the network, well, do you actually have the data?

Let me give an example. Let's just say that you wanted to type in a question like, why was Matt's Facebook experience terrible yesterday? That would be a perfect use case for AI. But do you have the data to be able to answer that question? Do you have it from the radio tower? Do you know that Facebook was working properly in the Bay Area when Matt was using it? All through to all the different components of the network. You have to have that real amount of information, regardless of AI or not. That's kind of fundamental.

Then there's the action part of it. That's the other leg of the stool. So if I want to increase capacity, let's just say I typed a command that said to increase the capacity to the 49 stadium for 20,000 spectators. Do you have the capacity and the network tools to be able to go do that?

So you can see that while having the kind of AI chat, almost this omnipotent, godlike way of controlling your network — which is what we all want to get to — I would argue that the investment in orchestration and the investment in observability has not been made in order for us to realize any of those things. And that's before we get to the model conversation or the lack of models.

Data disparity

Coward: We're spending a lot of time on disparate data sets and bringing them together. What's the next step? Data is a good example of bringing together these kinds of things that you might not think have an immediate association. Not necessarily just from the network as well. Can you take traffic information that's coming out of the traffic sensors? Because that's going to tell you information about why something might be congested, or where users are and start bringing those into a place where you can draw those those correlations.

Or you would say, well, I want to deploy my next cell towers. Where should I place them? That might even be future like how a city might be planned, or a new housing estate might be going, or how a tool building might impact the reception of your radio signals. There are some really interesting dynamics about how this can come together.

But there's a kind of fundamental that is missing from the talk environment today that does speak to the foundation models. I don't think the industry has good foundation models around time series data and topology. And these are two very important things that haven't been solved.

Vision recognition is a really good example. If you think about how much it costs to train a foundation model, it suggests that not one company or one telco is going to resolve that just with their data set. Billions and billions of data points are necessary. One of the things we're interested in at IBM is, should we be coming together with other interesting players and with telcos to say, is there a foundation model that we need to jointly invest in that gets us to the right level. That helps us bring the union of all these data sets together and train to provide the outcomes we're looking for.

How to get started

Coward: Take a specific domain. Take an area of your focus and start looking at the AI use cases. We've identified something like 40-50 different used cases where AI can make a practical difference today across that network infrastructure before we get to this advanced future. So pick the pain points, go after them whether it's with IBM or with other vendors and start getting the data assembled you're going to need start getting the orchestration you're going to need.

There's a there's a fundamental shift of Capex that is going to have to be made if telcos are serious about getting the benefits of this. So for the billions that are pouring into 5G, the questions the telco should be asking are, how much of this Capex should we now be investing in AI, if we plan to get to the savings and the network benefits we all have promised the street in 3 to 5 years time. This is that journey.

Watch the full video for the rest of the conversation between these old friends and colleagues, who also happen to be tech visionaries.