In June, AWS announced a $100 million investment to develop a center to help enterprises use generative AI (GenAI), the technology behind OpenAI’s ChatGPT chatbot. The AI innovation center will offer generative AI testing and training services and will be staffed by data scientists, engineers and solutions architects. To find out more about the company’s broader approach to generative AI and how telcos might use it, RCR Wireless News spoke with Ishwar Parulkar, the chief technology officer for the telco industry at AWS.
While Parulkar acknowledged that generative AI does indeed represent a “big transformation,” he also pointed out that AWS is no stranger to AI and has, in fact, been working in this space for the past 15 to 20 years, both internally in a customer-facing capacity via offerings like the code generator tool Amazon CodeWhisperer and Amazon Bedrock, which makes it easier for developers to build foundational generative AI models.
Looking ahead, though, Parulkar said that GenAI will “touch pretty much every industry” in one way or another and that nearly every customer the company works with is interested in finding ways to put this tool to use. For telcos, specifically, Parulkar sees the value of GenAI unfolding across the following three phases:
The foundational phase
The first phase involves the the foundational AI models and capabilities that exist today, such as text summarization and generation, as well as image generation. “So we’re looking at things like chatbots that can be used for customer care,” he explained further, adding that this of particular importance to telcos because customer churn is a top metrics they use to track success. Therefore, they are looking at using AI to improve the customer experience, as well as business applications like revenue assurance.
Additionally, the existing manuals that field technicians use to install equipment and troubleshoot network equipment can be used to train models that can create a more real-time interactive interface to guide them through the installation and troubleshooting processes.
For the most part, all of these capabilities are available today, according to Parulkar.
The training phase
The second phase, shared Parulkar, is about training foundational models to do “new types of activities.” One of those standout activities for telcos is configuring networks. However, Parulkar said that in order for this to be possible, the existing foundational models must be tuned with new data, and a lot of it, and so this phase will take some time to come to fruition.
The network-based phase
The final phase is “a little way out,” Parulkar confided. These would be foundational models that are “network-based” and “geared for the network.” Such models, he continued, could be used for several network applications “from the designing of network to maintenance of networks to all operational aspects of networks.”
In closing, Parulkar warned that GenAI still faces notable challenges, particularly around security, data privacy and ethics and that it is crucial for telcos — or anyone looking to use this tool, really — to remain vigilant and only operate in environments where the AI can be managed with responsibility. “This will only become more and more important because GenAI is based on large amounts of data from multiple sources,” he added.
