This told essay is based on a conversation with Natalie Gilbert, a 30-year-old data scientist at AT&T. Natalie Gilbert’s father was a researcher at the company’s Bell Labs division. The interview has been edited for length and clarity.
As a kid, I was very naive about what AT&T was.
What I knew about this company came through the lens of my father, who researched speech recognition. He collaborated with people such as Yann LeCun, who was developing the ability to detect handwriting and convert it into text, and Dennis Ritchie, who created the C programming language.
My father’s work in speech recognition and synthesis laid the foundation for what I do today with generative AI. Everything I’ve built here has the same foundation that he was working on: convolutional neural networks that allow computers to process inputs such as images and audio. It’s really amazing to see how that foundation has evolved.
Natalie Gilbert and her father, Mazin Gilbert. AT&T
Their early discoveries enabled us to work with AI agents and make them more autonomous.
As a child, I remember being in my father’s office almost every day after school and watching him and his co-workers have heated discussions and draw funny diagrams on the whiteboard.
That inspired me to start drawing my own decision trees and other pretty nonsense stuff, but the experience taught me how to be creative and analytical.
One of the side projects my dad and I worked on together was called Dr Bot, which was an early iteration of a large-scale language model that could assess symptoms and tell you where to seek treatment.
From whiteboarding to coding and vice versa
What I do with my AI agent actually boils down to a bunch of decision trees that reason about how to get from point A to point B. That’s something I learned from my father from an early age.
Human interaction is becoming increasingly important when building AI technology.
AT&T’s Chief Data Office is working on a project to change the way people think about using HR technology within the company. By having an AI agent identify the policies and procedures relevant to an individual’s situation, we are essentially eliminating the question of where to turn to resolve HR issues. In an organization as large and complex as AT&T, this is no small problem.
Natalie Gilbert and her colleagues at AT&T. AT&T
In my own work, I use coding co-pilots (digital assistants), which speeds things up significantly, but those developing AI tools still need to understand the underlying technology of LLM and machine learning models.
New AI tools are incredibly powerful, but they can’t do everything
As these co-pilots grow in popularity, problems can arise if you don’t understand how these technologies fundamentally work.
For example, AI tools are useless if you don’t know how your code actually handles edge case scenarios.
At the same time, people feel like they need to learn something new every two months.
What I see changing in large-scale language models is that they are more focused on natural language rather than coded. I mean, I actually spend most of my time doing prompt engineering, which is not coding. Let machines understand us using natural language.
It’s kind of ironic because it’s a different version of what my dad did 30 years ago.
AI has changed a lot in my lifetime, but now I feel like I represent him and represent his legacy. It feels surreal to continue the work he’s been doing.