San Francisco Bay Area, California. Jungwon Byun is the COO and co-founder of Ought, a research lab that builds tools applying machine learning to human reasoning and decision-making. Prior to that, she spent four years leading growth at Upstart, an online lending company that underwrites personal loans to expand access to affordable credit. Saffron Huang, on Interact Communications, caught up with Jungwon in December.
Hey Jungwon, can you tell us about what Ought does and what your role is?
Ought is a machine learning research lab. Our mission is to automate and scale open-ended reasoning. We want good thinking about events in the future to become abundantly available. Answers to questions like, “When will this project be done?” should be as easy to get as “What will the weather be like next week?” — the parts of our life where we already have plentiful good predictions. It’s clear how useful forecasts are, but they should be available across many, many more domains.
We want AI to be useful for those questions, so we build tools that partially automate the workflow for people making predictions. We get them to start thinking out loud in the tool and want to introduce more and more automation to the process until we can do automated reasoning end-to-end, so someone just types in a question and gets an automated answer back.
I’m the COO, so I do everything that isn’t research and engineering. I do lots of product and product strategy: what exactly should we build, what does our roadmap look like, how do we prioritize? How do we think about our value proposition? Who are our users?
Tell me more about how you decided to found Ought.
At the beginning of 2019 I was thinking about starting my own company. I was interested in how people process stress because I had a friend who had really bad anxiety and depression. There were a few times I was very seriously concerned about her.
I noticed patterns in how we worked through problems. And I thought, is there a way to productize this so that people who don’t have the resources to sort through this giant stress cloud can still get that kind of help? How do we break down the anxiety bubble into things that feel more manageable and controllable?
I got connected to Andreas, my co-founder, who had been working on a very similar mechanism for about a year, but his interest in breaking down problems was from an automation perspective. The idea is, we need to teach AI to be useful for helping humanity with complex, open-ended problems. But AI is not good at that today, and that type of reasoning — while very impactful in society — is really hard to scale. Maybe one way to automate it is breaking hard problems into smaller pieces that are more discrete, reusable and therefore more automated.
The shared mission and passion to help people by helping them with reasoning and decision-making was the most important factor for me joining him. We had a lot of values and fit overlap; also, his skill-set was very complimentary to mine. On top of that, I felt really excited about being part of the bleeding edge of AI technology.
What are your views on using AI and how they could interact with humans in the future?
I recently had the thought that humans should do the things that no other human can do. The extreme version of the idea is that if multiple humans can do it, no human should do it.
Concretely, Ought builds tools for forecasters and policy analysts thinking critically about the future, which is currently a very human-intensive process. People have to do a lot of research and literature review, and build their own world models.
Trying to make these forecasts is a lot of work, and work that I don’t uniquely need to do. Anyone could have found the introductory contexts, datasets or benchmarks. Where I want to participate is at the point of the assumptions or beliefs that only I hold, or at the most critical assumptions, where there’s a lot of disagreement.
The tremendous redundancy makes me less interested in making forecasts, even though having good predictions is really useful for the world, e.g. when will we get a vaccine, when will artificial general intelligence happen.
So one way that AI and humans can reason interactively is, the things that are generically true across all people should largely be automated. People should focus on the layer of reasoning where we don’t have a lot of data and there’s a lot of uncertainty, and at the level of your fundamental beliefs or values. That’s also the type of reasoning that people want to do; it’s not that interesting to go and find the dataset.
That’s fascinating. Let’s take a step back: how did you end up here? You grew up in New Jersey, right?
I was born in Korea, and moved to New Jersey when I was six. I grew up there, went to Yale, lived in New York for a year and then moved out to the Bay Area.
In New York you were working in consulting, then you joined Upstart, an alternative lending startup. How did you make that industry transition?
I wanted to be a consultant since I was in high school, which is… no one feels like that, but I was sincerely very passionate about management consulting. I often felt like it was one of the best jobs, and on paper it was, but something always felt missing.
It was a very consumerist career, where you are the product. Everything revolves around you, your skills and your career path, which is great as a new grad. But it wasn’t a place where I could put my heart and soul behind the mission. I could clearly see myself seven years from now, and the hoops I’d need to jump to get there. But at the end of it, all I would have built was my own career. I wouldn’t have built a product or a company or a team.
I’d been passionate about fintech and alternative credit models for a while and was interested in startups since college, so when the Upstart opportunity came along it was the perfect fit.
Switching to a startup, it felt like I was cramming five consulting projects into a week. At Upstart I had to make decisions at a similar level and of a similar type as consulting, but at a much higher velocity. I also needed to not just provide considerations for someone else to make decisions, but actually have my own proposals to advocate for.
The first year at Upstart was really hard, but even on days when I cried, I always knew that that was exactly where I needed to be.
That sounds like an intense but meaningful experience. You were there for four or five years; you must have seen it grow a lot.
Yeah, it’s crazy. When I joined, the company had just come off its first phase of hyper-growth. After I joined, the industry contracted in a really intense way and we essentially plateaued for a year. I was on the growth team, so there was intense pressure to deliver on goals and nothing was working. It was very, very stressful.
The next year we went back to hyper-growth, so I got to see many different phases of a business condensed into just a few years, which I guess is the value proposition of a startup.
What did you learn from being stuck in that plateau?
In general, I am much less afraid now. So many times I felt like I had just ruined everything; then we survived and things got better. So I have a better way to contextualize my fear response. When I feel really stressed or afraid, I know to trust that a bit less because I’ve felt it before and things were fine afterwards. I feel more resilient.
What do you do for fun? Any new lockdown hobbies?
I started working with a writing coach! I originally wanted to get feedback on my business-related writing. But my coach would much prefer to talk about deeply vulnerable, intimate raw things so it ended up just becoming a creative writing relationship. So I’ve been working on a short story with him. It’s so much fun. And it’s actually half therapy.
At some point I want to publish my stuff. I have a secret dream of being a writer someday.
Last thing is, how has Interact impacted you over the last few years?
I viscerally remember the day of the retreat back in 2016. I had just moved to the Bay, and I didn’t really know anyone. It ended up being incredible. Some of my closest friends are from Interact.
Every experience I have with someone at Interact is like, if I’d met that person in any other context, I would have been like, wow. You are one of my special people. This is how I feel when I meet someone and our souls connect. And Interact is hundreds of those — it’s like, how did all of those people gather in this one community?
Every time a cool company comes out and I’m thinking, that sounds amazing, it turns out an Interacter founded it, which blows my mind. It’s an absurdly impressive group of people and they’ve been so helpful to me professionally. I think that combination of being both very practically useful for work and being these super thoughtful people that I connect with on a much more soulful level has been one of the most magical things about Interact and I really, really appreciate that.
Interact is a community of mission-driven technologists. Applications for the 2021 Fellowship are now open at joininteract.com.