Thomas Higgenbotham

Thomas Higginbotham is on a mission to make artificial intelligence more human. The third generation in his family to serve as an officer in the U.S. Air Force, he sees a future where AI assists human decision-making and reduces catastrophic equipment failures in the military through predictive maintenance.

Higginbotham’s Air Force experience helped build expertise in using the electromagnetic spectrum as a means of disrupting enemy operations, such as jamming communications. It also gave him a deeper understanding of the barriers to more widespread adoption of AI.

Intrigued by Professor Fei-Fei Li’s work at the Stanford Institute for Human-Centered AI, Higginbotham applied to one MBA program: Stanford Graduate School of Business. He left active duty for that program and, as a student, served as research assistant to Jennifer Aaker, professor of marketing, and Fei-Fei Li, professor (by courtesy) of operations, information, and technology. While at Stanford, he published seven case studies on human-centered AI.

Higginbotham then applied his skills at Alphabet’s X, the moonshot factory, on a product vision related to multi-modal computing and AI. He’s now Senior Director of Defense at C3 AI, a company that builds generative AI for enterprise, where he leads a portfolio of technical teams and programs across the Department of Defense and Intelligence Community.

Higginbotham lives in Idaho with his wife and enjoys photography, including astrophotography, time-lapse, and aerial photography.

What aspect of AI excites you the most?

Based on the focus that I have at C3 and beyond, it’s the opportunity to improve human decision-making and support that in one way or another. I think there’s a lot of future potential to reduce burden on certain mundane decisions and improve the inputs on complex decisions. Humans are fundamentally good at very complex decision-making. Artificial intelligence is not necessarily great at that, but it can support certain aspects of that.

What kind of decisions would benefit from AI?

Our biggest program is for Air Force predictive maintenance, determining when to service certain parts in an aircraft, and ideally doing that before some type of catastrophic failure. Using all different sensor inputs from adjacent systems to service a part or repair an aircraft is complex. There are a lot of different factors in there, and some of them are related to the actual health of a system or a component. But what machine learning can do is provide insights.

We do something called sensor-based algorithms that give you a precise view of system health based on the sensor readings from different systems. Typically, the way a human would see those is if the sensor rating tripped a threshold, so went above a certain value, and then that would drive certain maintenance procedures. But now, with machine learning, deep learning, and our approach to sensor-based algorithms, you can actually have a precise view of system health over a period of flights and be able to monitor even in real time what the health of that system is. That can inform proactive maintenance. It can form very different procedures when it comes to the maintenance of aircraft and reducing catastrophic failures. You’re now able to rip or replace parts more proactively, so it saves a lot of downtime.

Is this in lieu of human decision-making or as a supplement?

It’s to help humans do things that humans are good at and reduce the burden on things that they don’t need to be focused on. Also, in some other dimension, machine learning is not necessarily replacing other decisions that humans don’t need to make, but it’s surfacing insights that they can’t see without machine learning. There’s no automatic triggering of a maintenance procedure, but it helps them see something they never could have seen before.

Can you share a lesson from the Air Force that you’re now applying to your work in defense tech and AI?

My last year in the Air Force, I was an exec officer on the Air Force Scientific Advisory Board, which advises the Secretary of the Air Force on emerging technologies and particularly the impact thereof on Air Force operations and strategy. We did a year-long study, and one of the things that we heard time and time again when we went to all these different units was that they had artificial intelligence or automated capabilities. But they did not use them and didn’t train to them because they were afraid that if they trusted them, they would start World War III.

Everything we do at C3 is around how we create a user-centric software application that enables their workflow but then, at the appropriate place, injects a little bit of artificial intelligence to help them do their job. And so they’re essentially able to use a software application for their entire flow that they’re familiar with, and then maybe in one part of the screen, you’re able to show a machine learning-based system health score that is just reducing the leap to actually adopt this technology. We’re really focused on adoption. That came a lot from my time in the Air Force.

Any memorable moments from your Stanford experience?

I always had trouble sleeping during my time in the Air Force. Come to find out, a lot of it was stress and PTSD related. I would wake up every sleep cycle, every 90 minutes. That was just how I slept for at least six years.

And then, at Stanford, I gave a speech where I talked about a failure in a mission. It was the first time I spoke publicly about it. When I was by myself preparing for the speech, and I was walking through my notes, I found myself shaking uncontrollably. It was a physiological response to this memory. It caused a lot of anxiety. And when I gave the speech, it was a huge relief. It was a lifting of that moment on the mission.

That night, I slept completely through. The following night, I slept the entire night through. It was the first time in six years that I slept two straight nights without waking up, and since that very moment, I have not had sleep issues.

Where do you see human-centered AI headed in the future?

Artificial intelligence needs to be designed and thought of with the human in mind first, and that impacts all sorts of decisions in terms of data you use and preserving privacy. It determines the types of algorithms you use. It impacts the application, whether you put this into a software application or whether this is an input into some type of human on the loop.

All of that is impacted by a worldview of human-centered AI, and I think that’s one of the things Fei-Fei Li and the Human-Centered AI Institute have been championing and something that the industry has really taken to heart. There was just an explosion of both research in the area and product approaches towards explainability and a human-centered application of this. So much of that came from Stanford.

What’s your advice to incoming students?

I got great advice from some alumni to optimize for something. You can’t optimize for everything. You can’t do everything at Stanford. If you don’t have something that you’re focused on, that you’re optimizing for, you’ll get distracted by every shiny object. You’ll spend your entire Stanford experience getting an inch deep in everything and not really getting deep into understanding anything substantially.

But if you choose the one thing you will optimize for, you will get something good out of it. I optimized for trying to go deep in this field of artificial intelligence and human-centered AI. That led me to doing some case studies with Fei-Fei Li and Jennifer Aaker and being able to really be at the epicenter of thought leadership in the space, which was an opportunity I would not have had anywhere else.

Photos by Elena Zhukova