AiThority Interview with Dr. Mark Austin, VP Data Science at AT&T

Hi, Dr. Austin. Welcome to the AiThority Interview Series. Please tell us about your role at AT&T and how you arrived here.

I manage a team of about 250 data scientists and developers in the Chief Data Office applying AI, machine learning and automation with applications across many areas of the company such as Fraud, Field Operations, Consumer, Security, Supply Chain, and Corporate functions. I also oversee our new generative AI tool Ask AT&T, which is an internal AI assistant for employees. I have a bachelor’s degree in electrical engineering from the University of Maryland and a doctor of philosophy degree in electrical engineering from Georgia Tech.

Could you tell us about the major challenges that enterprise information management teams face in the post-COVID era?

One thing that COVID drove was massive virtual work, and as such there was a major need to help employees find enterprise information online, as one didn’t have the luxury of turning to one’s office mate and asking them (although the rise of chat attempted to fill the gap). Nevertheless, this can be done more effectively now with the rise of Gen AI and its ability to summarize content via an internal secure enterprise Question and Answer system, and this is a bit of what we have enabled at AT&T with our “Ask AT&T” platform.

How does a powerful concept such as Generative AI solve the challenges that you mentioned above?

Expanding on the above, “Ask AT&T”, is an enterprise-wide secure large language model system, which has both a MS Teams and a web interface, and the ability to upload internal AT&T documents to “teach it about AT&T” so that it can succinctly answer AT&T questions.  While the teaching it is just beginning, we see massive uptake and use across the company and appreciation for its help in productivity gains.

Which departments are most likely to benefit the most from using Gen AI tools?

More than 19,000 users have been onboarded to Ask AT&T since early June.

We started by enabling the technical users across the enterprise, such as the coders, as we saw early-on that 35-40% of all questions on the platform were about explaining, fixing, and generating code and we wanted to help enable those insights. AT&T, like many companies, has been modernizing our software and moving to the cloud and we saw some great benefits from   using Ask AT&T to help these efforts. Additionally, we have some early success in assisting our care representatives and improved our customers’ ability to find answers on our website.

What triggered the idea of building your own ChatGPT tool? Could you tell us more about “Ask AT&T”?

As generative AI started to accelerate last year and into 2023, we saw this could become a game changer for productivity and how we operate. We started testing and coding what would become Ask AT&T in January and that led to an expanded, cross functional POC to determine our efficiency improvement metrics in February. For instance, we ran several A/B tests with coders who reported between a 20-50% efficiency gain (speed) on generating or translating code depending on the coding task and experience of the coder.

We’ve worked with Microsoft to make Ask AT&T secure and safe for our employees and our corporate data. It runs in an AT&T-dedicated Azure tenant that’s been pressure tested for leakage. AT&T employees can bring company data and information into Ask AT&T without worrying about that material leaking into the public domain. The first iteration enables a secure version of the ChatGPT LLM, however, our internal platform is flexible and allows other LLMs which we are testing for some use cases.

What has changed in the programming landscape since the launch of ChatGPT? Why are companies aggressively chasing the ChatGPT benchmarks in their existing technology stacks?

ChatGPT certainly has received massive awareness and publicity in just a few months. But it’s important to note that the LLM is only an enabler, and to make it real, secure, and useful for internal enterprise use takes additional efforts to stand up those platforms, and we think that will be the big needs for companies in the next few years Additionally, the whole Gen AI tech area is an extremely fast-moving space right now, and we purposefully designed Ask AT&T to be interoperable with new LLM’s, and the various pipelines that help bring content to the LLM to teach it to answer and solve internal enterprise questions. It’s also important that companies working with generative AI do so responsibly.

Our AI Policy and AI Guiding Principles ground our employees in both our overall AI strategy and the concrete do’s-and-don’t’s for creating, deploying and executing AI at AT&T. Issues like bias and accuracy have to remain at the forefront even as development accelerates.

Tell us what’s in store for AI research at AT&T—which industries are you currently focusing on, and why?

We’re focused on responsibly driving business value. We have literally an avalanche of proposed use cases across the company ranging from Question and Answer needs, Summarization, Classification, and various “multi-step” language problems for solving complex interactions or flows, which interact with databases and structured content as well. We have some early versions of “Ask AT&T Databases” working which unlocks a whole new world for people to interact with data also that we expect to be very fruitful for many roles in the company. It’s incredibly exciting to democratize its use across for all business units (Care, Tech, Supply Chain, Network, HR, Finance, … you name it….)  and see the myriad of use cases enabled for helping everyone do their job and help the customer in new and innovative ways.

Thank you, Dr. Austin! That was fun and we hope to see you back on AiThority.com soon.