Why I joined humans& and some (belated) reflections on pursuing AI research with meaning

A month and a half ago, I packed up six suitcases and moved to SF. I wanted to write about some initial reflections and learnings from this whirlwind and the journey that got me here.

The Japanese term ikigai refers to one’s “reason for being” and is thought to sit at the intersection of what you’re good at, what you love, what you can be paid for, and what the world needs. I have found ikigai to be a useful concept for thinking not just about life, but also about research, maybe in part because I, like a lot of researchers I know, derive a lot of fulfillment from my work.

On pivoting to research on AI tutoring

When I started my PhD in 2022,¹ I found myself starting to think a lot about why I was pursuing the research I was pursuing. I had many answers to why I was pursuing a career in academia,² but they were not specific to my particular research topics (at the time, LLM explainability). I think this is because I had been pursuing research driven primarily by my curiosity, not by the impact I wanted the research to have. In other words, I had found “what I love” but not “what the world needs.”

I have found that curiosity is not a scarce resource, at least for me. If you are in research, you are likely a curious person. And if you are curious, you are probably curious about many things. The hard thing is to find the right problem to be curious about – What is the curiosity in service of? What will be the impacts of the research, and what should they be?

After months of reflecting on some of these questions, I decided to pivot to working on AI tutoring.³ This was a space that excited my desire for both curiosity and impact: If researchers achieved our goals of building AI systems with personalized tutoring capabilities, that could enable meaningful impacts in education, some of which we’ve begun to see.⁴ And, at the same time, there are so many fundamental technical problems that need to be solved in order to achieve truly personalized tutors – clearly, since we were still far from models that exhibit human-level tutoring.⁵

1. This was two months before ChatGPT came out.
2. A whole post can be written about this alone so I won’t go into details for this blogpost!
3. This pivot was made possible by the support of my incredible advisor, Jacob Andreas, to whom I am forever indebted.
4. E.g. this study from Eedi and Google Deepmind
5. I also just think teaching is such a cool phenomenon! It probably helped that I grew up in a family of teachers and educators.

On joining humans&

Working on AI tutoring in my PhD has felt like living out a dream. The work is intellectually invigorating. I have found that working on applied problems can actually expose some of the most fundamental places where innovation is needed.⁶ It has also felt so motivating and inspiring to work in pursuit of a greater purpose. I have been deeply fortunate to work with incredible mentors, researchers, educators, and leaders in edtech. This all made pausing my PhD a really hard decision, particularly because I intend to pursue a career in academia. It was surprising, even to me at first, that I chose to put this all on pause, even temporarily.

But when the opportunity to join humans& arose, it felt like the natural next step. At this uniquely pivotal moment in time, joining humans& would allow me to help shape the future trajectory of AI for the better, with the team and resources necessary to make an impact at scale – all while working on the same types of technical problems that I have been working on throughout my PhD. It was clear that it was a once-in-a-lifetime opportunity – a chance to work with others driven in pursuit of a shared vision for the future. It is so rare to find a team of people who share the same vision as you, who work well together, and whom you believe in. I knew I’d be kicking myself down the line if I did not join them.

I, along with my coworkers, believe that building models that center humans (and collaborations between humans) is necessary to achieve a future with AI that enhances our lives. There are some fundamental innovations that are required to get there, and this is hugely exciting as a researcher. We are working on exactly the kinds of challenges that will enable AI to do the magical things that humans can do with each other at their peak, like teaching.

The past couple of months have already been an incredible adventure. It turns out that a company that centers human connection and collaboration will attract some of the best humans: My coworkers are brilliant, kind, and conscientious. As a bonus, our daily post-lunch walks (featuring very cool wildlife sightings!) and meetings filled with laughter make work that is innately fun feel even more fun. Most importantly though, I’ve found ikigai for this next chapter – shared with my coworkers – and the work we do every day feels like a step in the direction of what the world needs.

6. To achieve truly personalized tutors, we need AI models with better theory of mind; methods for simulating students to train and evaluate better tutor models at scale; better methods for clustering to find shared groups of student needs in a given population. Each of these is a whole research direction on its own.

---

A note

I wrote this in part because I hope it encourages other researchers to think about why they are doing what they are doing. What kind of future with AI do you want to work towards? If you solve all of the technical problems in your research, will the world be better? If any of this resonates with you, feel free to reach out :)