PhD Student

Lina Lukyantseva

PhD ’23
Lina Lukyantseva
Lina Lukyantseva
We can develop solutions that are not perfect, but practical, and offer guarantees for those solutions.
May 19, 2023
By

Lina Lukyantseva grew up in Novosibirsk, Siberia, cross-country skiing with her family on the weekends and taking advanced math classes in school during the week. She attended college in Moscow and went to Mexico as an exchange student, where she began doing economics research. It was there, she says, that she decided to pursue a doctorate in economics.

That aspiration will soon be a reality. Lukyantseva will graduate from Stanford this spring with a PhD in economics and computer science; her focus now is on the development of artificial intelligence. “We now have the kind of technology that used to be in science fiction,” she says, adding that how the technology is developed can put us on a path toward an amazing future or a catastrophe. As she departs GSB for an AI startup based in San Francisco, Lukyantseva’s mind is on the biggest question of all: how to make AI useful and safe?

Tell us about your experience in Siberia, and how it compares to California.

I was lucky to attend a very good high school in Siberia — there were only 20 students in my class, and many of them live all around the world now, including one classmate who ended up landing a job in Palo Alto. I’ve been surrounded by brilliant people — both in Siberia and at Stanford. As for the differences, California has much better weather! But even though I don’t miss the cold, I do miss the snow — fortunately, we are close to Lake Tahoe, so I spend a lot of my weekends in the winter there. I grew up cross-country skiing with my family; in California, I got into downhill skiing and backcountry touring.

You started your PhD in economics but then added a minor in computer science. What was the origin of that new interest?

In my first quarter here, I was taking a class on market design, where [economics] professor Mohammad Akbarpour presented a paper that involved the notion of computational complexity. He said something that changed my course of study: “I think every PhD student should take at least one class on algorithms.” So I said, “Okay!”

The next quarter I signed up for an algorithms class within the computer science department. I fell in love with the field right away, and I realized I wanted my research to be on the intersection between economics and computer science.

You’ve mentioned before that you appreciate computer science for its beauty and elegance, and how simple tools can address complex problems. Could you expand on these two ideas?

Many problems in the world, even if they sound simple, are too hard to solve exactly right. Even if it’s possible, it takes a prohibitively long time to do so. Instead, we can develop solutions that are not perfect, but rather practical, and offer guarantees for the performance of those solutions. For instance, we can say, “We are not able to solve your problem exactly, but here is a simple solution that is at least 95% of the optimum.” This approach was immediately appealing to me.

How did you combine your fascination with algorithms with economic thinking?

As an economist, I’m used to thinking about problems through the lens of incentives and strategic considerations. The natural question to me became, how should we think about algorithmic problems when those algorithms are being applied to humans, and humans can react differently?

What’s an example?

There is a class of algorithms called multi-armed bandits which are commonly used in recommender systems, for example, to populate your news feed on social media or to suggest new content on streaming platforms. These algorithms help resolve the “exploration vs exploitation” trade-off: this means to decide between suggesting content that the user is likely to consume and exploring a new option that has the potential of being better than the first one but is not guaranteed to do so. A common assumption made in the existing algorithms is that the users stay on the platform for a fixed number of periods. However, in reality, users may leave the platform sooner or later depending on how much they enjoy the recommendations. This observation motivated me to design a new algorithm that takes this human response into account.

Quote
“If AI technology continues to develop at extreme rates it is likely to lead to extreme outcomes.”

And what motivated you to work on AI?

Throughout my PhD, I was looking for a topic to focus on in the long term, something that would be not only intellectually engaging but also impactful for other people. I was searching for answers to the question: What will determine the development of society and what kind of challenges are we going to face in the next 5, 10, or 15 years? In recent years it became more and more obvious that the answer, or at least a big part of the answer, to that question is artificial intelligence.

The first time this became apparent to me was when AlphaFold was released. AlphaFold is an algorithm that offers highly accurate predictions for the three-dimensional shape of proteins based on their sequence of amino acids. Though I was not familiar with the details, I knew that the protein-folding problem was one of the most important unsolved challenges in modern science. So when I saw the news, I seriously thought for the first time that AI is going to change the world. And then, of course, there was code generation, image generation, large language models that are capable of producing text indistinguishable from humans. I believe this is just the beginning.

How do you see the future of AI?

There is a wide range of scenarios. In the positive ones, the future is unbelievably good. AI-empowered scientific progress leads to a solution to problems like cancer and climate change. Most humans don’t have to work and can instead spend time with family and friends developing meaningful connections. Distributing the technology and its benefits around the globe would alleviate inequality by providing access to education, and medical or legal advice at a low cost — just imagine a personalized tutor, doctor, or lawyer ready to answer any of your questions. Even smaller things, like entertainment, are going to get better: today, the set of content is fixed, and we are trying to find the best option out of the available ones; in the future, new content may be instantaneously produced and fully personalized to your current mood and needs.

But there are also very negative scenarios. Economically, it will be challenging to distribute the benefits of the technology. If we fail to do so and all the benefits and power which they bring are concentrated in the hands of a few, it will exacerbate inequality, not alleviate it. More importantly, advanced AI is associated with safety risks. In a world in which it is impossible to distinguish if information, of any form, is coming from a human or an AI, we will encounter trust issues and massive misinformation campaigns. AI may be used to facilitate cyberattacks on the infrastructure of national importance or software controlling powerful weapons. Finally, there could even be an existential risk to humanity if at some point we deploy a system that we are no longer able to fully control.

It sounds like either a utopia or a dystopia.

I think that if the technology continues to develop at extreme rates it is likely to lead to extreme outcomes. But there is some chance that the rate of progress slows down significantly, in which case we won’t see much more advanced AI.

Can you share some thoughts on your experience at Stanford?

At Stanford, I had a chance to learn from world-class economists, world-class computer scientists, and world-class industry innovators. There are many great academic institutions, but Stanford is truly unique because it is located in the center of the technological world.

I’m very thankful to my PhD program — it gives the students a lot of freedom, and it fully trusts us to find the right things to work on. It is a challenging process, but in the end, it is very rewarding. The faculty are amazing and always open to discussing ideas. I was particularly fortunate to be advised by Andrzej Skrzypacz who was extremely supportive and encouraged me to work on the ideas I feel passionate about.

What’s next for you?

I’m going to join an AI startup as a research scientist. I hope to contribute to steering AI toward the brighter future!

Photos by Saul Bromberger & Sandra Hoover Photography

Lina Lukyantseva
Lina Lukyantseva
PhD ’23
Hometown
Novosibirsk, Russia
Education
BA, Economics, New Economic School & Higher School of Economics, Russia
Field of Study
Economic Analysis & Policy
Current Profile