Subscribe and receive the latest news from the industry.
Join 62,000+ members. Yes it's completely free.
Analysts say that more preparation is needed for telcos to take full advantage of what generative AI can offer. They must put the requisite data in order and ensure they have the skills and expertise to use the technology effectively.
January 15, 2024
Are telcos ready for generative AI? Yes, but only in small amounts at first.
That's the consensus from a recent webinar where Ben Wodecki, Deputy Editor for AI Business, hosted a panel of Omdia analysts that included Eden Zoller, Chief Analyst of Applied AI; Roz Rosenboro, Principal Analyst; Alexander Harrowell, Principal Analyst, Advanced Computing for AI; and Liyan Jye Su, Chief Analyst.
In the short term, telecoms will apply generative AI to customer service, improving chatbots by training them on customer data and making their interactions closer to a human experience. Gen AI can also improve the experience with human agents by providing them with scripts, strategies and recommendations for particular situations. But the gen AI revolution in telecom circles is just beginning.
"Telcos are still at a fairly early curve outside of customer experience management and customer care, but I think it's really useful sort of looking at what's happening in other Industries in other verticals," Su said.
Noting how some global telcos have become media and entertainment providers, he anticipated that generative AI would play a role in that part of the business. One example is automatically tagging video content to identify what would be restricted by compliance regulations in particular locations.
Another application that telecoms can copy from media companies is creating trailers tailored to individual interests. Su added that it can also improve accessibility by improving the accuracy of subtitles and dubbing into different languages.
Should telcos build their own generative AI models?
Zoller said telcos will need partners to move into Gen AI. "I think very few telcos can build their own foundation models from the ground up. A more viable option is to sort of fine-tune one of these existing foundation models and actually collaborating around fine-tuning." She added that "fine-tuning is a lot harder than it sounds."
The approach would have to be done in stages, starting with identifying the type of foundation model you aim to fine-tune. Therefore, it's essential to understand the various models and "their pros and cons" for a telco's needs. Zoller said, at this point, "the easiest route" is to use Chat GPT "to enhance existing applications."
Generative AI in the Telco Domain: Transformation and Disruption
Making use of unstructured data
What makes generative AI potentially so transformative, Rosenboro pointed out, is that it makes it possible to tap into unstructured data in a way that hasn't been possible before.
You used to have to get everything neatly set up in a database to work with it. "But now you can listen to all the customer care calls and figure out these situations were rectified, so there's a lot that's available now to help train those [AI] models," Roseboro said.
There are several ways to do that, as Wodecki explained. He said its possible to enable "querying data stored outside the model itself, which has a bunch of advantages, notably that rather than adapting it by retraining the model, we're going to adapt it by connecting it to a source of external data and use it as a query engine."
Additional generative AI use cases
Harrowell observed that it's possible to gain great value from simple AI applications. For example, computer vision could ascertain if all the parts of a manufacturing assembly line or repair facility are exactly where they should be.
Roseboro anticipates more value from intelligent process automation. She also sees great potential in synthetic data generation that will enable service providers to conduct "network simulations to help them figure out how to optimize performance."
AI can deliver "root cause analysis" for back office operations, Rosenboro pointed out. She considers that "the biggest near-term opportunity because it's not as reliant on getting boatloads of data into the model."
Pooling resources could accelerate generative AI adoption. Zoller noted, "Telco collaboration around generative AI can work quite well at the fine-tuning level." She mentioned that is the goal for the SKT and Deutsche Telekom partnership.
However, she added, "I think when you get deeper, and you're looking at collaborations to build a foundation model from the ground up, that's a whole other scenario altogether," which may not work out.
APAC market anomalies
Not all telcos are the same, and telcos in the APAC region are largely split into two camps regarding AI adoption, Su said. One is made up of "technology followers" who operate in smaller markets and would have to buy an AI solution from a vendor or hyperscalers. The other camp, which may be "unique to APAC," is whole countries as large as China, aspiring to "design and develop their own large language models."
China Mobile already launched a large language model this past March. That language model was "trained on a data set that features two trillion tokens, and that amount of depth of data alone is, I guess, unique to China because it is almost impossible to replicate that outside of the country," Su said.
He added that SKT also has its own AI hardware, and it has local competition that is developing the same. What's driving the innovation in that part of the world is not just the "language element" but also the fact that both serve government agencies requiring that AI technology used by those telcos be developed locally.
On the question of use cases, Su said the telco industry in APAC has to comply with the strict regulations governing the industry. This extends to defining "the quality of services" delivered. Also, those telcos believe that applying generative AI to streamline network operations can help their networks run more efficiently. That efficiency brings the added benefit in reducing energy consumption.
Recommendations for telcos to prepare for AI
Regardless of where telcos use AI, their networks will be used to deliver AI applications. That means they need to avoid latency problems that can occur from introducing AI applications. Su's recommendation was not to reinvent the wheel by building your tools when you can work with ones developed by specialists outside the telco domain.
Zoller highlighted the importance of responsible AI. She explained: "Responsible AI is complex" in that it encompasses various elements that need to be in place. Its goals extend not just to securing data privacy but also to be sure that AI models are "transparent and accountable" to assure that they are "free from bias."
She added: "If you embrace responsible AI, which everybody must do – not just telos – you have to do it with integrity. You've really got to mean it; otherwise, it's just more ethics washing."
Roseboro pointed out that any generative AI project has to be aligned with "a measurable business outcome with KPIs, so you can make sure that you're actually getting some return on that investment and not just doing technology for technology's sake."
Read more about:AI
You May Also Like
Rethinking AIOPs — It's All About the DataMar 12, 2024
SCTE® LiveLearning for Professionals Webinar™ Series: Fiddling with Fixed WirelessMar 21, 2024
SCTE® LiveLearning for Professionals Webinar™ Series: Cable and 5G: The Odd Couple?Apr 18, 2024
SCTE® LiveLearning for Professionals Webinar™ Series: Delivering the DAA DifferenceMay 16, 2024