Designing destiny: Roanoke alum now working in leading-edge tech

Thomas Lux ’16 jokes that he’s a Roanoke College fanatic, but it’s hard to resist gushing about a place where he discovered a passion for research and community service, a penchant for leadership and – most importantly – the woman of his dreams, his wife Alex (Grant) Lux ’16.

He has since parlayed that experience into a Ph.D. in computer science from Virginia Tech and a career as a research scientist for Meta, where he specializes in designing artificial intelligence and machine learning technologies. 

Knowing that Lux would be the perfect person to help us envision the future of artificial intelligence, we called him at his home in Redwood City, California, to pick his brain.

Q: What inspired you to pursue this path that you're on?  

It started at Roanoke College, before the industry was so captivated by AI. When I first started doing research, I went into robotics but realized very quickly that it wasn’t the hardware that blocked you from making cool things – intelligence was the limiter. So I considered what I could build that would have the most value for everyone and AI was an obvious answer, if not a little bit of a pie-in-the-sky one. It's very motivating because if you solve this problem, there is no other problem that you need to solve. It is the penultimate research goal. 

Q: What else about your experience at Roanoke got you started on this path?  

Oh, a lot. I wasn't just like an academic. I was a resident advisor. I was in leadership for student government for two years, and I loved to explore different topics while getting my core competencies. The unique thing that Roanoke did for me is that it blended my ability to do great work in three areas: working with people, being creative while building community, and doing great research one-on-one with professors, independent studies and summer research projects.  

Q: What do you think is the most misunderstood issue about AI?  

People think AI is coming to steal the jobs of the world – kind of the doom-and-gloom scenarios. They are worried about AI taking over the world. Frankly, I think that's a big misunderstanding. At the end of the day, it’s just like almost every piece of technology we've invented over the last few 100 years. Yes, it is a tool more powerful than all the ones before it, so obviously with power comes risk. But there’s no inherent danger to it. It's all up to how we use it and how we adapt to what our life can be with it around. I think people should embrace it – with hesitancy, of course.  

Also, it is technically a little bit easier to build safeguards into AI that make it harder for people to do harm with it than some other tools. So, I don't think it's as dangerous as a lot of the experts talk about it being. And I think a lot about this stuff. I'm not just on the sidelines.  

Q: What are some ways you think AI will improve our lives in the near future?   

I think one of the biggest ones for everyone in the world will be personalized health care. There's still some work that needs to be done in safety and making sure it doesn't make bad decisions, but we're on a very near-term trajectory to these language models being able to look at all your medical records holistically and give you very personalized, accurate, helpful, and cheap medical advice. Google has recently shown that they can perform better than a lot of general practice doctors at looking at the cumulative biometric data that's collected by watches and phones and assessing your risk for health consequences. 

Also, I think the new AI tools that are being created in the realms of genetics have the potential to create whole new classes of cures for diseases that couldn't exist before, because it's hard to predict how DNA will cause proteins to fold, and if you can predict that accurately, you can design drugs that bind to very specific parts of the body or cells or structures. I think that is probably one of the best paths to fighting some of our hardest-fought battles in all of medicine, like cancer and heart disease. 

Another big area for AI is automated transportation and self driving vehicles. When that becomes widely available it will return a lot of time to people that is currently lost. It will be up to us all to make the best of that newly found time though!

I'm not personally super excited by all the social media stuff – ironic, I know. That's where people are probably going to see it the most, but I don't think that's necessarily going to be the one that changes people's lives for the better; that's just going to be entertainment. On the entertainment front, I think it's going to be a huge windfall to the amount of entertainment material that can be created, but we'll see about quality. We’ll see if people end up drowning in pools of content they’re not interested in.  

Q: How does machine learning help solve real-world problems? 

In general, all of machine learning is about making a prediction. Anywhere you can make a prediction, it can be helpful. Artificial intelligence is already being shown to compete with some of the best weather models that have been developed for decades. As I just mentioned, Google has done a lot of projects that have been publicly talked about, such as a system called AlphaFold, so they can predict how molecules will interact with each other and that helps you design new drugs for medical treatments. There's a really cool rocket company right now, Relativity Space, that uses machine learning end-to-end to design the whole rocket on the fly, and then they 3D print the rocket custom built for a specific application. So there's an example of how machine learning can help us build a physical object and get satellites into space. In the realm of business, you can use AI to figure out how much supply to send to certain parts of the country so that you distribute your product best and you have the best margins. In finance almost all high frequency trading is powered by AI and new research is showing that language models are better than most stock analysts at predicting which companies will be profitable in the future. Machine learning can help with all of that and more.  

Q: Can’t humans do anything that AI can do?  

Yes, but AI can often just do it 1,000 to a million times faster or cheaper, so that's really what makes it valuable. Take self-driving cars, for example. Whatever your opinions are of the automotive sector or Tesla, enabling cheap transit that doesn't need human drivers is a big economy unlock. The North Star goal of most AI work is to achieve parity with humans on all tasks and eventually exceed it, so to become as good or better than a human at anything. I don't see people jumping to sort data in spreadsheets right now or jumping to manually operate elevators instead of using a push button. People naturally move to the tasks that they enjoy more, and I think AI will change the landscape of what tasks are fulfilling and valuable.  

Q: Will we ever have AI that can do laundry and dishes? 

Every person that works in robotics and AI loves the goal of the house robot, but it’s remarkably difficult to engineer. For example, how many times has anyone seen a robot sit down in a chair? Almost never, because the action of bending your knees in the right way and not tipping the chair over and sitting down – that’s a very hard balancing act. Still, I think the odds are north of 75% that within two years, we will see our first commercial bipedal walking house robot that is reasonably affordable. I don't know what reasonable is – maybe $10,000, which is really expensive, but that's still technically affordable. It’s less than a car.  

Q: What else excites you about the future of AI?  

My personal vision – and I've thought about this for more than 10 years now – is for us to get to the point where AI can be a caretaker for humanity, but also for life. In the sense that we could create a rocket, put the necessary computers and robot arms on it that had this caretaker AI, and it would be fully capable of reestablishing life on another world. That is, to me, the ultimate sci-fi, incredible universe that we can build. In this scenario, humanity would look a lot like it does now, but digital life will be everywhere too. It'll just be another type of life. And to be honest, that's kind of the most robust form, because it's a lot easier for it to survive. But now we're talking about a sci fi book. 

Q: Yes, great book! That can be what you do in your off time.  

I’ll have Chat GPT help write it for me. [Laughs] 

Q: Any current projects you are really proud of?  

It takes a small data center months to create Chat GPT, but I’m working on a new algorithm for AI that could make it so efficient that you can get something smarter than a chat bot that not only runs on your laptop or phone, but is trained on your laptop or phone. It could be something that's been with you your whole life, like a lifelong digital assistant. I know the chances that I create something that revolutionizes the industry aren't very high, but it's nice to have big dreams.