AI and the future of telecom with use cases

Do Telecoms use Artificial Intelligence?

Artificial Intelligence (AI) has a long history in the telecom industry. AT&T likes to say that they’ve been using AI for nigh on 22 years now (since 1996), but let’s be honest and admit that AI has changed significantly in the last quarter century.
In the beginning, it certainly wasn’t very smart. It was largely a reactive system, doing a limited number of fairly useful things, in response to a limited number of inputs. Anything more difficult than interpreting tones and redirecting calls or recording a message often left it gaping like a spent fish that had accidentally jumped onto land.

In the Beginning

Our modern AIs are much more intelligent than those of 20 years ago. Their ability to decipher content and interpret meaning in real-time is, in many cases, absolutely astonishing.
The simple models that we play with on a daily basis, such as Alexa, Siri, Cortana, or Lyra, are not artificial intelligence per se, but rather reactive systems that respond to certain fixed stimuli. They have a wide range of responses but aren’t too much different from the very first program like that.
ELIZA was that computer program, written in the mid60s (you may have seen it referenced recently in a TV show called Young Sheldon, based on the youth of the Sheldon character from the Big Bang Theory), which demonstrated how superficial communication was between machines and people.
The creator, Joseph Weizenbaum, knew and claimed that it was not an AI, and was amazed at how easily people fell into the idea that there was a real person operating the other end of the teletype (monitor screens were new and rare—most people interacted with a computer via a teletype printing on physical paper).
No matter how often he explained that it was an algorithm that simply turned the question around and gave it back to you (like a non-directive psychologist would), people still insisted that it had a real personality and emotional understanding. They saw it as a “friend.” The program had created one of the most important aspects of human-machine communication: Empathy.
Effectively, it was the first computer program that could pass the Turing Test, meaning that people couldn’t distinguish if it was a machine or person. Even Weizenbaum’s secretary used to ask him to leave the room so she could talk privately to her “friend.” We humans do seek to have interactions with others, even if it is only a simulation of a human being.

AIs and Telecoms Need each other

With millions of customers using all of the telecoms’ services, at nearly full-capacity, on a continuous basis, AI just can’t arrive too soon. There are trillions of possibilities at any given moment—well-beyond the capacity of humans to resolve—and so we have relied on sophisticated automation to manage these complexities. Even our ultra-fast switching is being taxed by the current demands.
But wading through all those possibilities, relying on brute force computing, could still take minutes, hours, or weeks. Fast as computers are, they’re still too slow to handle the sheer volume.

Understanding the Question

AI steps it up to a whole different level. Yes, it can or will be able to talk directly to customers in many languages; it can type in an online chat in their native language if it’s properly cross-indexed with other languages or one of the 103 currently listed on the Google translator. If it can’t do it already, it won’t be long before it can listen, interpret, and speak any language for which it is programmed; it’s completely unfazed by accents unless someone elects to speak Vietnamese with an Italian accent, or some other unusual combination.
Where it excels, however, is in its ability to interpret the question, classify it according to the most likely applicable answer, and eliminate all of the irrelevant categories. Through the use of Neural Networks, Machine Learning, Deep Learning, as well as Prediction & Modelling, it thinks in a way similar to how humans think. That means it doesn’t have to slog its way through 99 percent of the database, and an answer might be forthcoming in just 500 milliseconds, instead of minutes, or even more.

Real-time Speech

Half a second (500 ms) pauses would be almost undetectable in ordinary speech. Further, if we put any significant effort into it, our synthesized voices could sound exactly like a human being, with all the properly changing tones and inflections.
Sounding like a real person and emulating empathy makes the customer feel better as if they have been heard and understood. Why should it matter if it’s an AI, provided the problem is solved satisfactorily?
Using the SQuAD (Stanford Question Answering Dataset) to create a test shows that two companies (Microsoft™ and Alibaba™) have now exceeded human question answering capability for the first time. This demonstrates a deep understanding of the meaning of the questions.
It could even understand the Plumber and Chemist quandary properly because it would understand the context in which the word was used. The original quandary was: “Seeing two people, and knowing one is a plumber and one is a chemist, what single question can you ask both to be certain of their professions?” The answer, of course, is to ask them to pronounce the word unionized. The plumber would say “yoon-yun-ized” whereas the chemist would say “un-ionized.” AIs can make that distinction contextually.

Invisible Changes

Infrastructure is being tasked to carry far more information than has ever been required of it in the past. We’re also demanding a higher level of reliability when that information is transferred. The worse the traffic level becomes, the worse the network should be—if we rely strictly on logic—but that is not the case.
AIs are doing a much better job of getting the information where it needs to go even when the conditions are not ideal. Using Network Function Virtualization (NFV) AIs can understand information transmission conditions; they are now driving Self Optimizing Networks (SON), which means they can autonomously select the best channel for a given type, quality, or quantity of data.
Most people who play MMORPGs (Massively Multiplayer Online Role-Playing Games) through the Telecommunications Network don’t even notice if the occasional information packet gets dropped (lost). The games are designed internally to cope with lost packets because they were originally designed to run on a much lower quality dial-up telephone network. As long as the network response time stays under one second, most competitive games are still quite playable.
People watching HQ High-Def video will notice when information is lost because portions of the screen may freeze, video artifacts may appear on the screen, and sound quality may deteriorate. On the whole, these sorts of events tend to pull you out of the experience, making it very difficult to enjoy.


NFV works best by performing switching electronically instead of with physical hardware. This requires a Software Defined Network (SDN) on which to create the virtualization. This greatly improves efficiency, integrity, and speed so that complaints about these types of issues are decreasing. If you haven’t noticed a substantial improvement in your data quality recently, your service provider probably isn’t using an SDN/NFV/SON combination yet.
Of those that have, AIs are learning to detect decreasing quality in a network. In physics, entropy means that decay is inevitable. That also means it can be monitored and assessed. If certain configurations cause faster failure, the AI overseer can alter the parameters. When it becomes a serious threat to the network integrity, it can work its way around the difficulty while creating a maintenance request to correct the problem.
A self-maintaining network will pay for itself many times over. Fewer complaints, and happier customers are of incalculable value to any business.

The Takeaway

AI is going to continue to be an integral part of the telecommunications industry. We have already been using it for a number of years, witnessing its evolution.
There should be very little reluctance or resistance to adopting even more substantial and capable iterations. We understand what early versions of AI have done for our industry, so we’re in the best position to understand its potential; moreover, we are poised to garner the greatest benefit because of our history with AI, and knowing how to exploit its capabilities.
Our industry is outrageously lucky compared to most others because of our historical experience with AI; it means that the learning curve is going to be among the shallowest of all. When others are still learning to walk, we will already be airborne—at least for those that don’t have to be dragged, kicking and screaming into the future.
Don’t squander the incredible lead we have over other industries by hesitating, or by over-thinking your participation in modern AI. If you want to learn more, we provide webinars to show you where the future is going and how you can get there. Don’t shortchange yourself. Join us today!