When you interact with AI in your daily life, you probably don’t give much thought to the humans and machines that make it work behind the scenes — and that’s part of the technology’s allure. It’s thrilling to sit back and watch ChatGPT write a cover letter or listen to Siri read the weather. But the excitement around AI often eclipses the very real physical, environmental, monetary, and human costs of the technology.

“AI seems like this intangible thing that doesn’t have any kind of physical presence or footprint,” says Dr. Sasha Luccioni, a Research Scientist and the Climate Lead at Hugging Face, an organization aimed at democratizing AI through open source and open science. “But actually, in order to create AI models or products, there’s a whole physical infrastructure that has to be put in place.”

As a leading ethical AI researcher, much of Dr. Luccioni’s work centers around the societal and environmental impacts of AI models — like calculating the carbon footprint of large language models and measuring the energy efficiency of neural networks. Practically speaking, the life of a researcher involves everything from writing code and analyzing data, to attending conferences and publishing papers (plus lots of reading). 

Dr. Luccioni’s expertise gives her a unique perspective on the AI craze, and insight into what’s missing from the conversation. Here’s how she first learned Python for a project, what it’s like working at Hugging Face, and her advice for combining AI with your own interests. 

What got me interested in the work

“Typically the people who profit most from AI progress are people who are already relatively well off — often the technology that’s made for the Global North is not applicable in countries where there’s no stable electricity or people can’t afford a smartphone. ‘AI for good’ tries to balance the power a little bit and give nonprofit organizations or people in developing countries access to technology. Essentially, it’s taking AI progress, adopting it, contextualizing it, and understanding whether this is something that AI should be used for.  

There’s this whole set of costs that we’re not really aware of, because AI is such a magical, ethereal tool that we’re using. For example, if you’re using Google Maps to navigate, you’re using a smartphone that’s created with rare metals, all sorts of plastics, and other components that have a tangible environmental footprint. 

Above and beyond that, when you’re asking Google Maps to find the best route given traffic conditions, it’s going to query an AI model that’s running on a cloud server. So the server is always responding to queries, and it’s consuming energy. These servers are also made out of metal, and often use a lot of water in order to cool them because they’re so big and generate so much heat. 

Part of my work is trying to make that more transparent, to connect the dots, and to raise awareness to the fact that it might seem like Siri is magical and responding to your instructions, but we’re using resources.”

How I learned to code

“Both my parents are mathematicians and computer scientists, so I wanted to do anything except computers and math. I always loved languages, so I went into linguistics for undergrad. Then I met a professor who was working in a field called natural language processing, which is using computers to analyze text. 

My first year I did an internship with my professor where we worked on analyzing Wikipedia using Python. I didn’t know how to code; I didn’t know any kind of data structures or anything. But I really, really liked it. I liked the tools and the opportunities that coding gave me compared to the more manual things we were doing in my other classes. I really like the fact that you could do things more scalably and get more insights from data. 

I took more formal coding classes in my masters and then in my PhD. The PhD program I did was really cool: It’s called cognitive computing, and the idea is to bring together people from cognitive science — things like linguistics, philosophy, psychology, all of those domains — and computer science. In my first year of PhD, I took a lot of more heavy, very structured coding classes, like Java. But until then, I was self-taught and did quick-and-dirty coding in order to get things done from a research perspective, not production-ready coding.”

What I actually do every day

“Hugging Face has a very refreshing stance on meetings — we typically don’t have more than one or two meetings a day, which is great. I read a lot of research. For example, I’m starting a new project to measure the energy efficiency of neural networks, a subset of AI models. I’ve been trying to figure out what people in parallel research groups and other domains have done. 

Right now, what I’m doing is writing a proof of concept, and figuring out how to measure the energy consumption of a laptop and a GPU. Once I make it run on my own local GPU, and on a couple of models that I’m looking at, I’ll write a more complex script to make it run on bigger GPUs and things like that. I’m trying to work my way up, and do experimental design around what I want to cover, and make sure I don’t leave anything out that can impact the final result. Since this is a relatively new subject, I’m trying to cover all my bases. 

I also am currently working on a more long-term project, which is developing a code of ethics for a major AI conference called NeurIPS. There hasn’t been an explicit set of guidelines about what people should and shouldn’t work on, or what kinds of things they should take into account in their work. The idea is to converge upon a set of principles that we all agree upon as a community.” 

Here’s what you need to get started 

No matter what stage of your coding journey you’re at, Dr. Luccioni suggests building projects around subjects or issues that you’re interested in — whether that’s music, sports, pictures of cats, or climate change. (Remember: Data doesn’t have to be fancy or have academic pedigree to be worth exploring.)

“It’s important for people to realize that they can use AI to further whatever convictions they have and contribute towards things that they care about,” Dr. Luccioni says. Need some project inspo? Take a look at our machine learning coding projects and see what resonates with you. 

For folks hoping to get involved with climate tech, Dr. Luccioni recommends checking out the nonprofit Climate Change AI. They organize workshops and community events for people who care about climate change and work in AI or are interested in AI. 

On an individual scale, there are tweaks you can make as a developer to design software that’s sustainable, meaning it requires fewer physical resources and uses energy efficiently. Read this blog to learn about the principles of sustainable software design and how to implement them. 

Energized to learn more about AI? We have lots of AI and machine learning courses that will help you develop the skills you need to make an impact with coding. In fact, we recently launched a brand new ChatGPT course that will take you through the ethics, risks, and limitations of AI. Be sure to read the blog for more tips and information about the diverse careers you can have in machine learning.