The end of a PhD feels like a good time to reflect on all that has led me to this point. PhDs are interestingly cumulative, in that one does not just “start” in a PhD program: high school interests, college interests, and beyond shape the path in a way that’s more linear than other careers. To that end, this is a story that starts in high school, and leads up to Vael Gates, age 27, doctorate in May. I found it useful to write a retrospective that linked a bunch of themes through my childhood and adulthood, though I have missed the story threads that didn’t weave nicely into the narrative. Hope you enjoy. (Length: ~15k words)
High School, Edina High School, MN
I’ve always been a happy, cheerful person. Earnest, hard-working, a touch autistic, with a tendency towards anxiety and concerned with adult feedback. Innocently self-absorbed and an avid reader; I was widely considered “weird”, but in a harmless kind of way. I grew up in an upper middle class family, in an upper middle class to upper class neighborhood, at schools where being good at academics was cool. Our school had three cheerleading teams and pep rallies: a weirdly classic American TV childhood. My parents were utterly devoted to my and my sisters’ education and well-being, and our house was filled with books. I wanted to be a writer (still do in some sense, still love personal writing), but was fine to leave that as a future goal, after I’d had my expected career.
I liked school. I liked sports as well, and swimming played an enormous role in my life, both in that it was my most formative social environment and that I spent between 10-20 hours of my life a week doing it for many years. But school was good, and I had a great group of school friends to be competitive with, and I went through AP classes like Pokémon cards: gotta catch ‘em all. In high school, I was also learning about how to “do social” (see: innocently self-absorbed, a touch autistic). Good thing I wasn’t looking for deep friendships or dating, because I spent a lot of high school learning how to do things like hold eye contact correctly. (One year I was scolded for not doing enough, then the next year my peers informed me I was staring, then the next year I was scolded for not enough again: stressful and confusing, social interactions are). Luckily, I didn’t care that much, and was still surprisingly popular (I happened to be attractive and smart, which both smoothed my way). So mostly I just did school and sports and worked very hard at both.
College, Wellesley College, MA; classes at MIT and research at my advisor’s lab (Harvard Med School)
It’s funny how college worked out. I and most of the people around me were expecting me to do a lot better at college admissions than I did, but it still turned out great. I had 2-3 schools to choose from, and in fact I told everyone I was going to one of them, but Wellesley sold me on visit day. Small classrooms, great community environment, and they’d let me swim. It was also an all-women’s college, which I expect contributed to its great community environment, but did have the issue that I’d never learn how to talk to boys. (In a stroke of good fortune, it turns out that talking to boys is very similar to talking to girls, so that my gradually developing skill of “talking to people” generalized and this didn’t end up being a problem.)
It took me a long time to get used to college. I swam my first two years, and loved the athlete identity. Like in high school, I had school friends and sports friends, and found my community mostly in my sports teams. I also threw myself into school again, asking for academic exceptions (e.g. permission for extra classes) in… huh, yes, even back in my first year. The pace was really fast, and I didn’t have the support structure I’d had at home, so first year, first semester was one of my roughest periods. But I didn’t feel like I had any other options, so I continued on, pushing as hard as I could tolerate. By my sophomore spring, I had a sense that Wellesley was home, that this is where I belonged.
When I went to college, my parents had set clear goals for me: I was to major in math or science. Well, I’m certainly not majoring in math, I thought, and took the intro biology, chemistry, biochemistry, and neuroscience courses to see what stuck. Neuroscience won by a landslide, probably because it contained psychology (I liked psychology, despite my internalized bias towards the hard sciences), the teacher challenged me in a very satisfying way (very similar to my mother, in retrospect), and it was interdisciplinary. I use “interdisciplinary” as a kind of non-word catchphrase these days– “and I will bring a [positive] interdisciplinary perspective to this work”– where the influence is positive because I already happen to be doing it. But looking at my history, in what classes I choose to take and what I choose to do, I think I’ve always had a draw towards academic generalism. I want to know the basics of what everyone else knows, I like keeping options open, and I think I like the novelty, the stretch of it, the interestingness. (I was limited to the sciences due to prioritization, unfortunately, but I did do most of the pre-medical curriculum for the completionism of it. At some point in college I got to take two writing classes as well, and that felt like a luxury. Thank goodness for university distribution requirements (which often require that students take classes of many different types) and their implicit allowances.)
I ended up taking math every year, to my bemusement. I thought I’d be done after my first year, and then the year after that, and then the year after that, and then it kept on being useful to take math. I’m well-rounded academically, but math was always my hardest subject, and the “well-roundedness” was somewhat a function of me pouring in way more effort into math classes. I also started programming my sophomore year, and thought that was great. Coding really appeals to me, including aesthetically, in a way few other things do. Being a neuroscience major turned out really nicely because I got to take computer science classes, biology classes, and psychology classes for the major. The psychology courses were always significantly easier than my other courses, partly because some of the material got covered in my neuroscience courses, but I enjoyed them. During college I’d renewed my efforts to figure out social interactions, and really liked topics like “theory of mind”: thinking about other people’s thinking.
By junior year, I’d decided that Wellesley was getting a wee bit small, and those were some pretty awesome science classes being taught at MIT, eh? Mayhaps I should consider taking some courses over there through the MIT-Wellesley cross-registration program. The main kicker was when I learned that MIT had a class on “computational cognitive science”, and in fact there was a whole existing field of “computational cognitive science”. Cocosci: you could study SOCIAL PSYCHOLOGY in an ACADEMIC SETTING with CODING and [the ever so virtuous] MATH, how LEGIT, what could be better than that? And it happened to be taught by arguably the leading expert in the field, what even.
Well, if I wanted to take an upper-division class… you know the drill. If you wanna take a class, you gotta do the pre-requisites, and Wellesley’s statistics class didn’t contain Bayesian statistics, so I needed to take that one at MIT, and if I was already taking classes at MIT, why not stack in enough to make it efficient… in short, I spent a lot of time commuting between Wellesley and MIT my junior spring and senior year. 30 minutes one-way if you had class at a good time, 90 minutes if you left during rush hour. I had a lot of feelings about that bus, and even now recall that queasy feeling of trying to do work when you’ve been speeding and slowing through traffic for an hour.
Throughout it all, I had really excellent teachers and mentors, had friends and many mealtime conversations, seemed to stop having crushes (what? I had these in high school, huh?), worked really hard, and also did a lot of research. Remember that first year neuroscience teacher who I liked a lot? He liked me too, and I did research with him from my sophomore fall to the summer after I graduated. We had our success stories and misadventures: my friends were always hearing about the story arcs of my time with him. He was incredibly generous with the opportunities he created for me, and I learned so much under him. The people in his lab made up my third set of people (fourth? I also had hard-won MIT friends). I could talk at length about my development as a scientist and person under his and the lab’s tutelage.
All right, we’re to the end of college. I’d wanted to become a professor because I wanted to teach people, but somewhere along the way I’d been informed that to become a professor who wanted to teach people, you had to be hella good at this unrelated skill called research. I was very confused about that at the time, and still don’t think it’s a great system. But I’m good at jumping through hoops, so I did research from my first year summer through my last year summer, and was still planning to pursue a PhD once senior year rolled around. I was still mainly motivated by a desire to teach college students– which I hadn’t even done yet due to time crunches from doing research– but I figured I would probably like teaching, since I liked things like tutoring and mentoring and making documents describing how to do research tasks. Plus, I wanted a 10-year plan for my career, a clearly structured road ahead of me. That meant a PhD, JD (…no, I am not suited to law), or MD (ruled that out definitively sophomore year, but knew quite a while before then. I just… ugh, bodies, they’re so biological, plus interacting with people all the time, and moreover with sick people? It didn’t seem like a good idea). By process of elimination, and preparation, PhD it was.
I was told that I wasn’t ready to apply to PhD programs straight out of college. I don’t know if that was true; half of my eventual PhD class came straight from undergrad, and I had a lot of research experience under my belt, albeit not a high-author paper. I did buy the argument that to pursue computational cognitive science, I should probably have more math and computer science knowledge. But the point is somewhat moot since I was far too busy to apply to PhD programs my senior fall (easily my worst semester since freshman fall). I was really well settled socially and loved my classes, but I was also taking like five of them, half of them at another school (with the requisite commuting, and also teachers I really wanted to impress), doing research at yet another location, dealing with exciting small events like “one of the postdocs in this class is low-key stalking me”, plus doing applications for research positions and maintaining a minimal social life. I periodically talked to myself in a soothing and inspirational voice. I even pulled my first academic all nighter during that finals period! (I’m careful about sleep, so I had never done that before. And while I was definitely sleep deprived in college as a general rule, I was falling asleep when talking to people after the all-nighter, which was a new one for me.) As a reward for that semester, I only took three classes in the spring, which was amazing and gave me time to struggle through Introduction to Machine Learning. (That was the second time I’d ever gotten less than 50% on an exam! AP Statistics and Intro to Machine Learning, such times.)
Requisite segment about money: US schooling is expensive, but the top schools do need-based (not merit-based) scholarships and loans. At some point my parents were paying more than they could afford, given that they were about to send their second child to college, so they talked to the school and got more scholarships my junior / senior years? I never interfaced with money in college or when I was younger, at least in terms of the costs of attending school. In college I was responsible for earning any spending money, which I did through being paid for research. During the school year, the standard route is to take research courses for credit (…little known fact, you can use this to boost your GPA, though that’s not actually how it feels at the time), but sometimes you can apply for scholarships, or very rarely get paid through your lab. Summer research usually pays stipends, which gives you enough for housing and food and then usually a decent amount leftover.
(As a flag, you should consider the way I’m writing this post to be biased in the following way: I generally focus on the positives and elide negative experiences, research or otherwise. This is a deliberate choice for this post, because it takes a lot of time and effort and thought and words to describe negative experiences in a fair way, especially since the people mentioned in this post are anonymized to different degrees. My sincere apologies, but for the sake of brevity (ha) and actually-writing-this I shall forge on!)
Master’s, Cambridge University, UK
Anyway, during my ridiculous fall semester I applied for a research assistant / lab tech position (two years by default, done post-BA and before PhD), and funding for a research-only Masters at Cambridge University (one year). I didn’t get the funding for the Masters, but I did get accepted to Cambridge. If I could get funding, I figured why not: I wanted to go to grad school in a year, I was worried about my lack of computer science and math experience, I’d always wanted to take a year abroad, plus it was Cambridge. Family resources came to my aid again– I received very generous donations from both sets of grandparents, one of them explicitly for academic spending, and filled in some by applying for a Wellesley fellowship and otherwise wiping out any savings I’d received from research scholarships over the years. (I do not generally recommend doing a Masters for post-grad students, because of the expense. I believe it makes much more sense financially if you’re from Europe and getting a degree in Europe, but tuition in the UK was very expensive for US students. If you’re a lab tech, then you earn money rather than spending it! And once you’re in a PhD, if it’s a science PhD at a top program, they’ll pay you to do research / teach. (I’m not sure how funding works across programs, and it seems to depend a lot on the wealth of the department, so I’m just going to say what I know for sure.))
My Cambridge experience, let me tell you, was far less work than undergrad. I had evenings off. I had weekends off. I had time to go to the pub with my lab members, to do round one of the Tinder Adventures, to apply to grad school, to figure out how to make a website, to have tea and scones outside in the Cambridge spring, to go punting, to attend formal dinners at various of Cambridge’s colleges, to compete in rowing on the Cam and then on the Thames. I ran so many participants through experiments, met bunches of great people, and traveled to a bunch of neighboring countries. Compared to undergrad, my Masters experience was luxurious in how much more spacious time felt, especially after I’d gotten through grad school applications in the fall. I’d done plenty of extracurricular-like things in college and adventures with friends, but in Cambridge, I wasn’t constantly worried about opportunity costs and all the homework I could have been doing.
Finally, PhD applications. I applied to eight PhD programs that fall, and got into 4.5 of them. (I award half a point because I didn’t get into one of them, but I think I could’ve if I’d wanted it more.) I was pretty insecure about applying while I was doing it, because I wasn’t sure if I was qualified in terms of math/computer science. However, several professors sent encouraging responses to my inquiry emails, and that encouraged me to apply to labs I wouldn’t have otherwise. It came down to two schools in the end, Berkeley and UPenn. UPenn would have been great, but I count myself very lucky that I ended up here at Berkeley, which turned out to be amazing for reasons I couldn’t have known beforehand.
That was a chunk of my spring: doing interview visits, flying back to the US and feeling keenly how much I’d missed having a culture I knew around me, having an accent that blended in. (Some amusing anecdotes: 1. My lab was highly international, and I was the youngest by a decent margin, so it was amusing how bad we were at figuring out songs that we all knew (nationality / age / music taste) when we went to karaoke together once. 2. When I landed in the US, someone told me my English was excellent. Apparently I’d picked up enough British intonation and vocabulary that layered on an American accent I just sounded kind of weird. 3. It’s also probably worth noting that this was when the Trump vs Clinton election campaigning was happening, and Brexit happened that summer! An exciting time, in a really international community.)
And now onto my Berkeley interview visit. First things first: I loved Berkeley. I loved Berkeley as soon as I stepped off the subway, into its sunny skies in late January, where homeless people lined the streets thronging with students. I’d just come from my interview visit at UCLA, and was flushed with success, 8 hours jetlagged, and sporting a bum ankle from when I’d fallen down my lab’s stairs the week before. California has a feel to it– dry and light, eternally the cool end of summer, with mountains rising in the distance and shorts-clad cool kids jabbering, strolling. I still catch glimpses of what it felt like back then.
I’d wanted Berkeley to work out, because the PhD advisor of my dreams was at Berkeley, but I had a checklist to go through. Were the students happy? Yes, far happier than any place else I’d visited. Would my dream advisor Tom take me? Uh… pending, but maybe if I begged him some more. Did I like the city, the feel of the campus? Hell yes I did. Was this worth betting on, despite Tom’s uncertainty about whether he’d be able to take students? I still consider this the riskiest deliberate choice I’ve ever made, career-wise (I had a far more certain and great situation at UPenn), but I went for it.
Tom wasn’t supposed to take any students that year; he took three (and a fourth, dual advised). I assume we were all skilled in the art of begging!
PhD, Berkeley, CA
Year 1-2, PhD
Grad school was both all I’d ever dreamed of, and surprisingly stark: I felt empty in a way I hadn’t in years. Despite the bustle and newness of the program, there was a thought that hit in the quiet moments, when I was working on research: is this it? I’d been working towards grad school for so long, that it was astonishing to have made it… and what it meant when you made it was that you did research. What. Come on. I’d already done this for years, where was the anticipated sudden sense of fulfillment?
Despite my uneasiness about my chosen career path, which perhaps could have been expected given that I’d chosen this route because I wanted to teach and liked ten-year plans, Berkeley was lively. My first year was crazy, in a greater number of directions than it’d been previously crazy. (Previously, I’d maxed out my work axis, which was plenty stimulating and did not require anything else to induce feeling overwhelmed. Grad school, like my Master’s program, never got close to how hard I regularly worked in undergrad, and so I branched out to more life pursuits.) I’ll attempt to detail the gist of what happened that first semester, but I first want to lay out some groundwork.
Something to know about Vael, especially Vael from before year 4-5 of the PhD, is that I was perpetually insecure about my computer science, math, and especially computer science + math knowledge. I was mildly insecure in high school, when I was on a FIRST Robotics programming team but couldn’t program. I was mildly insecure in college when I attempted to teach myself MATLAB, and did okay at it. (I then took two more classes in computer science and enjoyed that a lot. Classes were good, and computer science was way easier than straight math to me.) I was very worried by the time I’d decided to try for computational cognitive science, around junior or senior year of college, and had been trying to cram in math and computer science courses as fast as possible before I graduated. I was insecure about it when I was deciding where to apply (such that I agreed I shouldn’t apply straight to PhD programs), and I was insecure about it when I was applying, to the degree that I almost didn’t apply to labs like Tom’s. I’d decided computer science and math were so cool that I wanted to date computer scientists (I’d also decided I wanted to date athletes, and that didn’t happen) whenever I got around to the dating thing, which I’d planned would happen around the last years of my PhD. I was generally insecure about the math thing, and then the math + computer science thing, in a way that had been getting worse as I got older because I’d resolved to work in a field where this was the thing you did. I’d in fact picked this field partly because I thought it was so hard, and therefore cool.
My self-assessment here wasn’t entirely inaccurate. I did find math and computer science challenging, and I was not as good at it as some of my peers. However, the amount of angst and anxiety about it was disproportionate to the actual difference in skill or ability to learn. In retrospect, I think this was because I knew I didn’t like learning about the math + CS combination, and so it took a lot of emotional urgency to learn it anyway. In any case, I was completely blown away by the people who did it despite how arduous it felt to learn, or who even (god forbid) liked it. So, coming into my PhD, this was one of my dominant narratives for looking at the world: I need to be better at math + CS, if people look too closely they’ll see I’m not near where I need to be, yes I kind of hate it but this is what the virtuous people do, oh god I want to be a cool kid by these metrics, I hope people don’t see that I’m bad at my job and that some part of me doesn’t even want to get better.
In other news about first-year Vael mindsets: 1. I still strongly wanted to be told what to do by people who both cared about me, and would also push me towards my goals. 2. I thought I wasn’t a proper researcher– I didn’t care about learning for its own sake– but someday I hoped I’d be curious enough to pursue knowledge for its own sake instead of pursuing goals, and then I’d be amazing. 3. I’d decided social interaction was still confusing, but that I thought I’d gotten a handle on what the rules were. I still definitely liked studying social interaction (had been doing it for years), and thought I’d gotten ridiculously lucky in my PhD topic (computational cognitive models of social interaction). 4. Everyone kept telling me to date. Whatever, I’d tried some Tinder dates last year, I thought I was pretty good for now. 5. People also kept telling me to try alcohol and caffeine, and I was still a no on those. 6. I couldn’t believe I was in the long-anticipated PhD program! I’d made it, and Berkeley was amazing!
Okay! Cue first year of grad school. My first step, beyond anything that was happening, was to get Tom to take me on as a grad student.
Step 1: My dream advisor, Tom, had sort of agreed to take me, but he hadn’t committed yet. I’d entered into a neuroscience program, though I planned to do computational psychology, so I had “rotations” with different labs. (“Rotations” are when you spend a trial period in a few different labs to determine fit, and they’re common in neuroscience programs but not in psychology programs.) To impress Tom, I thought I should do well in these different labs, especially since he’d basically set them up for me. One of the rotation labs he suggested made clear sense (computational decision-making in neuroscience), but the other one was in human-robot collaboration. What? I pushed back on that, but Tom insisted it was relevant to my interests. Well, okay, Tom– you’re something like a minor deity in my mind, so I’m happy to do whatever you think is best. (He was right, by the way, and I ended up working with that collaborating advisor for the rest of my PhD.)
I thus spent my first rotation (computational decision-making in neuroscience) attempting to pretend I knew how reinforcement learning worked and make some progress there, which wasn’t so bad. In fact, I do pretty well at research when it comes down to it. Not spectacular, but I meet expectations. I’d been dealing with my insecurity about computational stuff for years now, so it wasn’t any trouble to make progress on projects in computational cognitive science, especially with help. Thus, that 10-week rotation, and the second rotation in Tom’s lab, passed pretty smoothly. I was both surprised at how much initiative they expected me to take and also comforted by the mentorship that was given. All went very well overall, and at some point Tom agreed to take me, so that was research taken care of!
Step 2: make friends. I’d entered into a fantastic PhD program, and very much enjoyed getting to know the eight other students in my cohort. I ended up being lasting friends with two of them– the first one took a while to warm up to me though, since they thought I was weird. The second was a computational neuroscientist who did math + CS, found me fascinating, used computer science concepts when describing normal real life things, lived in a decked-out car, and was ready to teach me and cheerfully challenge me constantly. Uh oh, I thought. I wasn’t supposed to meet The One until the end of grad school.
Step 3: Oh God I Need To Learn More CS. Berkeley had a new center based in their computer science department, called the Center for Human-Compatible Artificial Intelligence (CHAI). They had weekly meetings open to students affiliated with CHAI, which my advisor Tom was, so I showed up to try to figure out what all the cool machine learning / computer scientists were talking about. I indeed had no clue what they were talking about, but after a year of listening understood way more than I did when I showed up. Turns out, these people were worried that unless we developed artificial intelligence well, we were going to eventually cause humans not to exist through one mechanism or another. (The book Superintelligence had come out in 2014, and it was 2016 at the time). Huh, I thought. That could be a problem.
Back to dating. After consulting various friends through Skype, I determined that I should in fact give the potential One a chance, even if he was a bit ahead of schedule. Though that turned out to be a mess in various ways and only lasted a ~month, I learned a number of things. I learned about the existence of the Effective Altruism and Rationalist communities in Berkeley, I learned to not freak out when my hand was held, and I learned a pile of things about relationships that I did not know how to process at the time.
Back to CHAI. At the end of fall semester, around December, one of the senior people at CHAI took an interest in me. He told me to attend CFAR, a retreat about improving one’s thinking, with people from the Effective Altruism and Rationalist communities.
Back to research. For my third rotation, with the human-robot advisor, I had a choice about what project I wanted to do. Based on my understanding of the work that was happening at CHAI, I chose a project that was aimed at reducing risks from AI using an empirical psychology study.
Then I attended CFAR, a five-day retreat with ~40 attendees and 20 staff, aimed at thinking about one’s strategies and life. And that, dear readers, was destabilizing.
CFAR was like: hey, Vael, what are you doing with your life?
Vael: …going to be a professor?
CFAR: Is that what you want to be doing with your life?
Vael: uh, it’s fine? I’ve had this goal for a long time? Is that the wrong answer?
CFAR: PhDs can sometimes be a sub-optimal use of time, if one wants to focus on the world’s problems. Did you know you have the potential to do something more useful to help the world?
Vael: …I…I like when people tell me I have potential. I like when people mentor me and tell me what it’s good for me to do
CFAR: We think AI safety is a real problem. We’re concerned that if we don’t have enough people thinking about how to make AI safe, there will be large-scale negative impacts for the world. There are many other concerns one could also think about as well. We’re concerned about the increasing power of synthetic biology and how that could be dangerous, like engineered pandemics. We’re concerned about nuclear war. I see you do research that’s in the AI safety vein. Have you thought about doing AI safety research?
CFAR: Just give it a thought. You might be well-placed to do something about this problem that may affect millions or even billions of people, since risks that cause humanity to go extinct prevent future lives.
Vael: …I! You… think the best thing for me to do is to do AI safety research? Like, the computer science + math stuff, without any psychology?
CFAR: That is what we’re talking about, yeah.
Vael: You really think math + CS is the best thing for me to do in my life? Like, morally? And you think I specifically can make a big difference here, so I specifically should work on math + CS? For moral reasons?
CFAR: …Uh, yes, if you think it’s good? You’ve gotten into a PhD program, so you’ve shown you have potential for it, so you could drop out and work on the problem if you thought it was worthwhile.
Vael: AHHHHHHHHHHHHH (burst into tears)
Uh, and that episode describes the first time I tried to quit my PhD. Luckily, I only got around to telling my parents before they talked me out of it. I feel bad about that particular lapse in decision-making: my poor parents were concerned that I’d gotten myself into a cult, and it really just was a confluence of a bunch of psychological buttons that had me that vulnerable in the first place. To be clear, I don’t think it would have necessarily been a bad thing if I exited the PhD at any particular point, or for the reason described to me (I do strongly believe in this reasoning), but doing it like this wouldn’t have been the way to go. Specifically, with no plan, a deep sense of guilt, and a sort of existential despair. (Why must I always do the math whyyyyyyyyyy I’ll never been accepted unless I do math, why do people always want me to do the thing that’s so HARD, why can I never do the things that I want, why must I always with the shoulds.) I’d tried to quit my lab twice before in college for different reasons, and my parents had talked me through one of those times as well. (I just sent off a text message to my parents to thank them for this particular PhD-quitting episode; I’d forgotten their involvement in this.)
I ended up wrestling with this issue for the rest of my PhD. The Effective Altruism community, Rationalist community (apologies for the pretentiousness of the name), CHAI, and CFAR all turned out to be related, and this is the place where I’ve found my home. And since becoming acquainted with their ideas (I’m not going to unduly describe them here, but check out the links), I’ve felt a pull towards working on AI safety, and more broadly on questions relevant to existential risk. (Whether or not you believe AI has the possibility of leading to humans becoming extinct, the broader thing I want to work on is topics that have the potential to cause humans to become extinct.) I became frustrated that my PhD research didn’t seem to be helping people, frustrated with my lack of interest in working on math + CS, and also, eventually, very tired of the internal pressure from not resolving this.
(I also opted into some social pressure about this, because I actively choose my role models and communities, and this reasoning seemed meaningful enough to me that I wanted in. “Choose environments with motivating social pressure” is the mechanism via which I got myself to learn a lot of math and CS, actually– seeking out environments where I’d be expected to know hard content because all the cool kids knew hard content, after which I’d feel highly motivated to learn the content myself. It’s a bit convoluted as a strategy, and I eventually stopped doing this and instead just went straight for what I wanted without routing through social pressure, but I only figured out how to do that around year 5.)
You’ll notice the “but should I work on existential risk, specifically AI safety?” thread throughout the rest of this narrative. It ended up as the background to the rest of my PhD, and I anticipate the rest of my life. It has its ups and downs, but despite the initial instability and continued confusion, I eventually claimed these communities as my own. I’ve formed myself into a person that is shaped around these beliefs and goals, that takes the core and makes it work for me. I’ve never felt as good a fit as I grew to have within these frameworks and with these people, and this is why I consider myself particularly lucky to have landed in Berkeley, not knowing any of this was ahead.
Ha– it’s always hard, when I’m talking to new incoming PhD students, to explain why I like Berkeley so much. There’s obvious things, like my advisor being astonishingly great, and the amazingness of the program and people and environment. But you know, there was also that period where I had a giant shift in worldview and slowly remade myself over four years into someone who could fit into the shoes I now wanted to fill.
…And. Well, uh. CFAR wasn’t actually just epistemically destabilizing, it also led to really large social changes for me?
Just after CFAR, a few things happened. I met some boys through CFAR I thought were cool, and ended up in a long-distance, open relationship with one of them for about four months. There were some amusing experiences in that one, and that plus another later experience started me down some major “introspective projects”, which is my term for when I decide I want a different approach to how I’m interacting with the world. These projects have historically been aimed at improving my experience of social interactions, though occasionally I’ll mount “projects” for improving my experience of something work-related.
(“Make myself into someone who can do AI safety research”, for example, is a work-related introspective project. “Enjoy reading papers” would be a subgoal within that. “Go on a date with a girl” is a social project, as is “Be less anxious, subgoal be less anxious before parties”, or “stop attracting this type of person”. These projects usually last on the order of weeks to months, but the higher-level ones can span years, especially if I’m not actively working on them. “Projects” are an important part of how I operate and internally push myself, and “I’m working on the x project” will be a useful explanation for what I’m doing in my free time if you ask.)
In this case, I mounted a project in the beginning of year 2 which was like: “omg, stop doing this weird pattern where you seek out people who want the best for you but will push you really hard, so that in some instances they’re pushing too hard, and then you feel overwhelmed and start crying and then fail to know what to do at all”. It’s a weirdly specific pattern, and I think I had to have it happen with five people (a couple in short succession) before I was like: haha, no, I’m getting a lot of benefit out of these people but also how about I skip the part where I’m getting pushed where I don’t want to be pushed. I think I basically stopped seeing people outside of work meetings for a month in November and just brooded about this for a while. I was thinking about what the patterns were, why it was happening, what I was getting out of it, what I was losing, what my beliefs were. Somewhat surprisingly, this worked. I made major internal changes after that point, and proceeded to stop having major instances of that pattern, though “overly dependent on others’ judgment” remained a trend.
(People’s mileage on introspective projects tends to vary a lot. I’ve been doing projects explicitly for a while, usually subprojects within some larger project, so I’ve got a decent sense of how hard something will be to change and how long it’ll take. I seem to have more success in doing this sort of thing than average– in that I’ll set a goal, and then have some shift in how I think about things, which will manifest in some sort of enduring, observable behavior change. I’m not super accurate about my predictions though, and part of these projects is also knowing when to give up: the “acceptance” route is just as much of an option and a good outcome from these projects as “change”, if I can’t get something to go how I want.)
This was a major period for me for going on dates, also. I got asked out by a LOT of people immediately after CFAR, because this was my first major introduction to the larger Rationalist / Effective Altruism community. There’s a substantial gender imbalance in the Rationalist community (also reflected in tech spheres), so it was clear why this was happening, but I’d never actually had this happen before. My crushes in high school were mostly unattainable, and I had felt far too confused by the whole ordeal to push for them. I went to an all-women’s college next. There were plenty of bisexual and lesbian women there, but the dance is a little different for female-female approach, and I was pretty socially oblivious and very busy, so I expect I missed some passes. (I’d known I was bisexual since about 8th grade, though I wasn’t classifying it as such for a bit. Regardless, I had crushes on women in high school, and had a crush on a woman my first year of college. When those feelings stopped, I abruptly stopped having crushes, except for that one time for like ten minutes my junior year. I had another crush during my Master’s year, but low-intensity. Low-intensity has been the name of the game since then, and I think it’s gotten even lower.) In terms of people asking me out, I’d had the odd person here and there in high school, college, and my Master’s program, but nothing on the scale that happened right after CFAR. I found this flattering and somewhat overwhelming. I met with a lot of people, since I was still feeling out my place in this community.
The “going on dates” period had a few relevant consequences. One is that the summer after CFAR, I also went on a series of dating adventures: the Tinder Adventures, Part II, where I was aiming to meet girls, since I was experiencing a lot of male attention. I’d met six cool guys the last time I did this in Cambridge, and I met another… 3? 5? women this time too. I considered my goals to have been met during this round! (Tinder Adventures III two years later was unsuccessful, but I did learn some things.)
Another consequence of that “going on dates” period was that suddenly I was forcibly being made aware of my gender. Being a woman was pretty much fine up to this point. In fact, I’d thought surprisingly little about gender when I was at an all-women’s college, because besides the “women are strong, we will overcome” mantra, which seemed cool to me, I never had to worry about gender because almost everyone was the same one. I only was really concerned about gender when I was in male-dominated environments in my MIT classes, but even then no one treated me differently, it was just something I noticed. In my Master’s program, I lived in an all-female dorm and was on an all-female sports team, and my labmates were mixed gender but cool and mostly partnered-off, so it didn’t matter much. My experience in grad school was that I was in male-dominated environments again, but like in most of my academia experiences (especially the research-focused ones) people didn’t really care that much or make me aware of it. This wasn’t always true– at least once, someone in class came up to me and was trying to get to know me kind of aggressively, without responding to how I was reacting to him, and my first experience of stalking was from a postdoc in a shared class. But the vast majority of time, in classes I noticed gender imbalances, but there weren’t obvious effects. However, in this community, dating was clearly on the table, and it was a whole new ball game. All of a sudden, “pretty woman” was the first thing people thought about me, and they hovered around me, tried to talk to me, asked me out, were interested in this particular aspect of me. That threw me off a lot.
In academia, the norm is that people care about how good your research is, and I was used to performing on that metric. “Research ability” is a bit contextual, but people generally seemed to evaluate me on competency axes: ability to do work well, ability to assist other people, ability to be broadly charming / friendly / nice to people, ability to move a boat or swim or run fast or lift things. I’m also very happy to be evaluated based on “interestingness”, which is a function of how I think and behave. But in the new world, I didn’t really need to do anything, didn’t need to prove my capability, wasn’t pressed on how I thought, nothing– people just kind of flocked. I was used to people being nice to me and approaching me because of my appearance, but I was also used to working a lot harder after that initial connection. Usually after we’d met, we had to feel each other out, figure out what we were interested in, meet as equals, see if we wanted to be friends or if we wanted to work together or whatever. I liked the attention of the new world, but it felt both like I had too much power in these interactions, and not enough. I wasn’t being challenged or evaluated based on the things I cared about. I didn’t work for beauty, and while I did work hard for charm, I’m not, like, great at it– I don’t consider these traits themselves as the most interesting things about me, though they’re nice as side dishes. In conclusion, I felt unhappy and uncomfortable with having my primary identity be “pretty woman”, and continued being confused about what to do about this for a while. (I lay this out as foreshadowing, because I continued to have feelings about gender, but a lot of my unhappiness with gender started here.)
(A happy note from the future: as soon as I stopped attending very large, more public Rationalist parties, and started making friends and attending smaller more tight-knit gatherings, I basically never had this problem again. I currently very rarely get asked out, and when I do, it’s after we’ve done the whole back-and-forth feeling-each-other-out thing for a while. My experience with the above problem went down to my normal base rates, which was: happens occasionally, in classes, in public (my second mild stalker came from this population), and very rarely at parties. These days I can’t even remember the last time this happened, and it would be really weird for it to happen among the subcommunities I’m a part of.)
(There’s also the observation that I seem to have been shielded from gender roles for an abnormally long time– grew up with two sisters, really career-focused and supportive family, almost all female friends all the way up to high school, female friends in college, good experiences in academia, was pretty oblivious on the whole for a while, and wasn’t interested in dating or partying so was basically never in situations where it made sense to approach me. When I did enter the dating scene, I was like… 23, and was being really systematic and self-driven about it. So these experiences were unusual for me, but I expect other people encountered these experiences a lot earlier. I was never super thrilled about gender (having breasts and periods, for example, always seemed kind of annoying) but I don’t know how my feelings would have been different if I’d been younger when I started diving in.)
Another consequence of this “going on dates” period, and of dating that long-distance person: I figured out I was pretty asexual, and also in the range of grey-aromantic. (Though determining the latter is hard, because I have been rather dissatisfied at most people’s ability to define what romance means, even when I stare at them and demand they do so.) I feel kind of amused, kind of oh-man-sorry-guys, and kind of indignant that I didn’t figure this out earlier. It certainly seemed stressful for many of the guys I went on dates with when my responses were “yep, this pretty much doesn’t feel like anything, but carry on”. However, I’m mostly aware of this inflicted stress in retrospect, because people aren’t very forthcoming with “uh you know you’re reacting to this really weirdly, like everyone else I’ve dated starts feeling something about now”. That would have been really nice to know, but, okay, understandable.
This was also around the time I started floating the idea that I was a bit autistic, and accepting that as part of my identity. That had been told to me before, by at least three people, but in a “this is shameful” or “I’m insulting you” kind of way. Insults historically kind of bounced off of me if they weren’t a specific type– “you’re not trying hard enough / you’re not performing to standards / you should have known to perform this better”. So I was mostly pretty confused by people trying to insult me by telling me I was kind of autistic (people did all sorts of weird and confusing things all the time, also I didn’t like when people tried to hurt me), and didn’t pay much attention to whether I was or not.
So this was the golden age of me declaring neurodivergence. Grey-ace, grey-aromatic, sort of autistic. Pansexual when I was being sexual. (And nonbinary later, whoops.) It would have been more clear if I’d been fully aromantic or fully asexual, but no, I’m just low-intensity on both. I also characterized a trait which is something like “self-interested, doesn’t care as much about other people compared to average”, which can be described negatively as “fundamentally selfish, cold” and more positively as “strongly independent, focused”. This results in socially-negative traits like “will bail from hard times, not a true friend” and correlates with socially-positive traits like “self-knowledgeable, interested in people in a vaguely off-putting but interesting way, great at self-regulating and very stable”. This self-focused trait isn’t immediately obvious to people, because I care about things like cooperation and being nice and not hurting people, and am upfront about it when I think it’ll come up. But it can result in some behaviors that seem quite foreign to people, and is a little weird, as far as traits go.
Sometimes I think back on this period and try to remember what this sort of confusion (e.g. around whether I was asexual) felt like. I think it was like a lot of social things in my childhood, where the thought process went something like the following. “I feel x, which makes me want to do y. I’m sure they must also feel like x. But they’re doing action z and not y? Why? Don’t know, mysterious. Am I supposed to do z? I guess I’ll try to do z, especially if people tell me to, and figure out how to restructure my internal experience so that z feels fine to do. Okay, new goal.” My guess is that I didn’t properly understand that people could be feeling different things than I was, and often people’s explanations of why they were doing things were incomplete. Friends were very bad at explaining things due to defensiveness, impatience, and not understanding why I was asking; my mother was better. (I was asking bluntly and tactlessly about pretty sensitive topics, especially as a teenager, e.g. “Why are you wearing your bra outside your shirt?” There was a cami-over-shirt trend back in the day.) Moreover, pretty much all of the expectations I was meeting were about behavior and not feelings, so it made sense to focus on concrete behaviors, which people would more readily concretize for me.
Finally, the last consequence of the “going on dates” period: I got to meet and interact with the Rationalist and Effective Altruism communities! I really liked these communities; I had never encountered a group of people (I affiliate with both EAs and Rationalists) who felt so similar to me. These people felt so similar to me that I was confused about where my place was, which triggered a minor identity crisis (am I interesting enough? smart enough?). But I resolved this eventually once a cool person told me to stop being insecure. I made many friends, and spent a lot of time making in-roads with the cliques I wanted to be a part of. (These communities are fairly large, in the hundreds / thousands of people, so there are distinct subcommunities.) I also went to parties in San Francisco, went to some Burning Man spinoffs (“regional burn”), tried many weird woo activities, and moved into one of the group houses. I made an important work introspective goal: to be motivated to work not from guilt and anxiety but from something more positive. I got helped by many people in that quest, and eventually succeeded (I think that one took me years). I made introspective goals like not being so motivated by externally imposed and internally imposed “shoulds”, and that one also took me years. Lots of growth in that time period, lots of new friends, lots of events and parties and costumes, lots of newness and challenge.
The new normal
I explored a lot in years 1 and 2, and tried many new things. At some point I stopped being quite so adventurous, and settled into the community in the subgroups that worked out. One output from all of this exploring is that I’d gotten very into introspective stuff, and also into the idea of helping people with their introspective issues. By the end of the second year summer I’d set up a website to do coaching on the side, and was generally paying a lot of attention to mental health as a topic. This felt to me like an extension of my interest in other people’s minds, social inference, the like– but the applied version, the helping-people version. Academia is usually incremental, small improvements on the frontiers of knowledge, but this is actually less my vibe than accumulating existing knowledge and making it useful. I was also spending time in a very altruism-oriented community, so I was thinking about how I could be helpful post-PhD or elsewise, and thinking mental health might be a good bet. (I continue to have a coaching website up, and still see people occasionally.)
Throughout all of these social shenanigans, I was actively pursuing a PhD. I didn’t have a great way to deal with the post-PhD career stuff, wanted time to build additional psychological scaffolding, thought working towards a PhD seemed good, and thought having a job that I liked and was good and stable while I figured everything out seemed like an excellent idea. So I basically kept doing normal activities and research, with a bias towards AI safety research projects when the opportunity came around. Having a default path which was “get a PhD” is pretty amusing when I think about it– but I did put in the work to get there, and I also did plenty of work in it!
Thus: academia. In my first year, I finished doing my academic rotations with three excellent professors, was accepted into Tom’s lab, and was trucking along on two research projects. In August (the beginning of my second year) I got the chance to attend the Marine Biological Laboratory’s Center for Brains, Minds, and Machines Summer Course at Woods Hole, MA, and I have wonderful memories from that trip. I met a lot of great scientists, got to go back to the East Coast, got to see a lot of MIT professors that I’d known in college, and got to work with some great collaborators and a fun project, in a beautiful environment surrounded by lots of water. I completed the requisite classes in year 1 and year 2, and solidified my academic friendships: I really loved hanging out with my neuroscience cohortmates in year 1, and became even closer with my labmates in year 2. I went to the big conference for our lab, CogSci, in London and Madison, WI, and had great times at both. I had projects, Tom was slowly getting me to be less deferential to him, and had settled into the groove of work. (One is supposed to get good at developing/proposing research projects over the course of their PhD… but I just got good at finding people who would give me projects which I thought seemed good and that I’d like. This was a surprising outcome of grad school for me, but also, man, many professors and other great thinkers are just overflowing with research ideas they’re delighted to take labor for.)
I also got to teach my second semester of year 2! It was a class I’d taken with Tom my first semester (Tom is an amazing teacher and I loved all of his classes), though another professor was teaching it for the first time this round. This was a difficult class to teach, because the content was really hard and there was a lot of behind-the-scenes technical backend, and I also wasn’t really prepared for teaching at all, I was just told to go lecture. (I was supposed to concurrently take a teaching course, but somehow I missed that memo and rediscovered the missing requirement quite a bit later.) However… I always thought I’d really like teaching– and though I thought I was probably being naive, it turns out I like teaching pretty much exactly as much as I expected. So I was busy and stressed, because I was teaching a class, preparing for qualifying exams (which happened in spring of year 2 for us), and also trying to do research and social– but I enjoyed each of the activities therein. Even studying for my qualifying exams has great memories: all of my neuroscience cohort studied together, and I passed my quals on April 25, 2018.
My first year, second semester, I’d started a research project that I’m still working on the extension of today (…why does research take so LONG), affiliated with CHAI, focused on social inference in human-robot interaction. It’s a cool project– my coolest on paper– and helped settle in my fit in terms of what I was going to focus on in my PhD. “Computational models of social interaction” was the general topic, and that’s pretty much the title of my PhD now.
My second year, second semester, we got a new collaborator on that project: a guy with the cutest ever smile who was actively good at computers and helping me make progress on research. I eventually got him to date me, which was a lot of fun and also educational, of course. With him, I really started working on my aversion to being touched, and made a ridiculous amount of progress on that in less than a year. (Previously, I didn’t like being touched at all. Then I would do hugs as a social greeting formality only– I reached that stage in high school, and was still at that stage around year 1. By the end of year 2, I actively enjoyed touch.) He broke up with me (I’m at about even odds at the moment?) a couple of months in, but at least for all three people I’d tried dating up to this point, it had been obvious when the end was coming. I learned a lot from all of them– especially 2, from whom I learned a lot of communication skills– and was pretty “eh, no more dating” in between each of them, but kept being excited by the new people.
In fact, by the end of that summer, I’d picked up… three more people. This feels represented to me by one of my happiest, brightest memories of a pool party with friends. Settling into the polyamorous life, I’ve actually continued dating three people, to some definition of dating, through the end of my PhD. I’d also started playing Dungeons and Dragons that September with an awesome group, which I’m still a part of even as 2-3 of the members have switched out over the years. That group also kind of feels like dating, and is a particularly bright spot of my time here. Happy Vael in Vael’s community.
Buuuuuut, one last upheaval
Something Tom had told me about, when I interviewed with him, was that he was maybe going to leave Berkeley. This is actually why he was trying to not expand his lab when I joined. I flagged this as a future-Vael problem, where the immediate Vael problem was that I really wanted to get into this person’s lab. But by the end of my first year, it was pretty clear that everyone in the lab needed to be making decisions about whether they wanted to follow Tom to Princeton or not. And by that time, I definitely did not want to leave Berkeley; I was very happy here. (I look at all the above text and I’m like: how did all of that happen? What even was year 1 and year 2 of my PhD?)
I spent a lot of time thinking about this– it shows up over and over in my emails. Tom wouldn’t be able to fund me my fifth year, because he would lose Berkeley affiliation two years after he left, and it’s also generally pretty bad for a student’s career not to follow their PI. There were lots of considerations that went into this, but in the end, I decided to stay, along with one other of Tom’s main-advised students, and two of his co-advised students, while the rest of the lab (10-20?) followed Tom. I feel really grateful that Tom let me make the decision to stay, and has continued to be an excellent advisor from afar; it wouldn’t have worked for everyone.
Second upheaval: during my second year summer (after CFAR and my parents talking me down from quitting, after I’d decided to stay in Berkeley, after I’d finished my quals, after doing ~2 years of PhD work), one of my cohortmates was thinking of leaving with a Master’s, which you could do after two years. He in fact did leave, and ended up quite happy with how the whole thing went. I, in the meantime, was awash in feelings of “I want to do good” and “I don’t think my research is useful” and “I have a direction now: mental health”, and was seriously considering leaving as well. This time, I percolated on it for months, and talked it over with many people. Finally, at a wedding in Greece, in which two of my labmates from my Master’s year were getting married, I called up Tom long-distance and told him about it.
Tom was a dream. He told me that this was completely normal around this time, that it would be fine for me to take a leave of absence next spring, and that in the meantime we should talk about how to make the PhD better meet my goals, and what kind of research we could pursue to make it better for me. I mentioned mental health; he mentioned computational psychiatry, and maybe collaborating with a faculty member at Princeton who I’d met previously. I was sitting there, alone in a wire-framed chair in the evening in a restaurant in Greece, phone held to my mouth and headphones in, and just felt incredibly, incredibly relieved. It was all going to be okay. I could take a leave of absence; Tom wasn’t mad. I still had to figure out what the heck I was going to do, now that I’d really discussed it with Tom, but I would find my way.
Year 3, PhD
Year 3 gets more… depressed. Not greatly so, but I was mildly depressed from the fall until around June 2019. It was almost entirely career-based unhappiness, where I’d wake up every morning with feelings like “my work is pointless”, / nihilism / “ugh what is my life, the future feels overwhelming and like I don’t know what to look forward to”. Everything else was going well in my life, including that I started seriously dating one of my partners (we lived in the same group house, so this was easily accomplished). But career’s always been very important to me– it has historically been my most important life goal, and currently still is– so not knowing where I was going next was hard for me. I call this period “mild depression” because it was maybe only a 7/40 on the Beck Depression inventory, but was way off from my usual mood, and it lasted a long time. However, it didn’t really affect my activities, and it wasn’t evident to people from the outside (though I certainly talked about it). I actually decided to escalate with a partner partly because I thought I needed more support in figuring everything out, which was also my reason for restarting therapy in the spring. Both surprisingly and unsurprisingly, once I’d figured out a good direction to go next, the depression ended and I was good to go. This is somewhat unusual because sometimes depression just happens to brains, but mine was just very clearly cognitive and was a response to my anticipated future.
A major conflict with my decision-making was that I really wanted my career to be about AI safety. Remember back in first year when I wanted to quit my PhD to do something useful, namely AI safety, but also I didn’t want to do math all day? Exact same problem. I still had literally the same problem two years later, had been trying to figure out this problem for two years, and had made very little progress. This was very frustrating. I’d initially started out wanting to do AI safety because other people around me told me it was the right thing to do, but over the years I’d decided that it was something I personally really wanted to do, so I didn’t see why I couldn’t just do it. I was angry about my interests: couldn’t I just be interested in technical AI work? But I felt so unhappy at the possibility of doing technical AI research for the rest of my life.
Because I didn’t know how to resolve this, I decided to explore my more innate interests. While I very much wanted to do AI safety, I was also at my lowest mood in ages, and didn’t want to live like this the rest of my life. So I started looking for some sort of alternative route, to see if I could find some kind of compromise that was not AI safety, but at least felt altruistic and helpful to me.
I was very active this period, which you maybe wouldn’t expect, but it was a direct response to “oh god I need to figure out where I’m going next”. In my second year, I’d done a lot of informational interviews with professionals who worked on topics of possible interest to me: ethnography, science writing, mental health startup, data science, coaching, etc. (Those interviews are available at vaelgates.com/writing, by the way, if you too are curious what those jobs are like.) From those interviews, I was circling closer to the interests of “mental health + data science”, with maybe science writing in the mix, so I made exploring that the focus of year 3, especially the spring. I went and interviewed a bunch of clinical psychologists to see what that was like. I was interested in going into science writing, so I wrote a couple of articles about psychotherapy research to see how I liked that. I applied to do a clinical psychology program the next year (and got in! That was very exciting). I was interested in mental health startups, so looked into a bunch of those and applied widely to internships (was completely unsuccessful in that), and investigated startups via talking to people and attending YC120. I started doing clinical research with a new collaborator, and I learned a lot within her care and the generous opportunities she offered me. (That work also got me in the trenches with linear and logistic regressions, which was additionally appreciated.) Tom was very supportive throughout all of it, and Tom and I thought about various potential collaborative projects with the clinical psych faculty in addition to my continued projects.
I was actually expecting to take a leave of absence in the spring to do all of this exploration, and had it all set up up until the day that I was needed to confirm. That morning, I sat down with a housemate, because I thought: actually, I think it might be best for me to stay in the PhD program while I do this; I’d rather have income and work and stability while I figure this out, and it actually might be better to stay in academia. Having decided this for myself, I then cancelled my leave of absence and did all of the above while still doing research. My decision-making feels kind of odd sometimes: I’m obviously a planner, but some of my largest decisions have been made on “gut”, where I have duly informed my gut with a bunch of information over time, but I’ll occasionally reverse major decisions very rapidly at decision points. (For example: my choice of college, my choice of grad school, this decision. These decisions have historically felt like “lunging”, like “I know what I’m supposed to do, but actually I really want this thing, whatever man I’m going!” I haven’t felt this sense of “lunging” recently, I expect because I’m more in touch with where my “supposed to”s come from, but I do sometimes make rapid decisions on a dime that I don’t regret afterward.)
As an aside, an amusing thing about my PhD path is that I look very standard on paper. No leave of absence, consistent research, everything on time and mostly under the same research heading. You would have thought that when I came in saying “I want to be a professor”, I just completely stuck to that and remained unchanged through the whole thing. In fact, it feels like I was throwing myself out of the PhD over and over, and then coming back to it over and over: I feel like I had one of the most churn-filled PhDs you could have while still looking almost entirely straight-lined.
In any case, the end result of that period was that I’d resigned myself (there was a strong mood of “must do acceptance”) to a mental health + data science career, had decided I didn’t have to do a startup (I thought I should, since motivated and leader-type people did startups), and that I was unhappy about this (because AI safety! It calls!) but it was okay because I was stuck with the interests that I had. This didn’t seem like the most exciting future to me, but I didn’t feel like everything was pointless, so I adopted that as my plan near the end of third year.
As best as I can remember, most of my introspective goals third year were around “how to not be unhappy”, “what am I doing in the future”, “I think I should be a leader and internally motivated by my work: how??”, and “I should learn to think for myself”, with a lot of content exploration about mental health. And that was most of my third year in a nutshell!
Sidequest: the gender thing
You know, ideally for narrative flow, I’d only have one thing going on at a time. This whole main-thoughts / academia / dating / miscellaneous threading is not making this document a clean read. But here we are, since we must accept non-ideal organization flow: I also had some pretty intense gender stuff happen my third year!
During this period and now, gender has always felt like an annoying side project to me– which is definitely not everyone’s experience, and I’m only trying to speak for myself here. However, I never quite figured out what to do about my unhappiness about gender, and several talks with nonbinary people eventually convinced me that identifying as nonbinary was the route I wanted to go down. It’s really not a perfect fit, and I feel apologetic and a few other things about that, but I just kept being bothered by it and introspective solutions weren’t working.
In my third year, I changed my name on Facebook and with friends, got the process started for legally changing my name, and got the process started for top surgery (removing breasts). I was pretty unhappy about decision-making around various of these things for a long time. (How do I feel about gender? How do other people feel about gender? Do I change my name? What to? Is this okay? How should I dress? Do I want surgery? What kind of surgery? Do I want testosterone? How much testosterone? How do I want to identify? What does this do to taxes / publication record / etc? Do I tell my parents? What friends? The public? What pronouns? Do I enforce pronoun usage, and how? Do I enforce name usage, and how? Is it worth making others do cognitive work for me? How will others see me? How will I see me? Will I still be attractive? To who? Will I make my life harder? What behaviors should I adopt? How do I feel? Do I feel enough gender dysphoria? Do I even care about this? Why do I care about this?) I talked to a lot of people, and thought a lot about it, and had various periods of confusion and upsetness that came and went. In January, I had a conversation with a friend which made me feel a lot better about it, and it was nice to get through all of the items on my checklist eventually, after lots of administrative work and a major surgery. I continue to feel kind of resentful about having feelings about gender (man, nonbinary in addition to all of the other things?), but all’s well that doesn’t bug me that much, and given that I have those feelings and it doesn’t interface great with society, I’ve enjoyed many of the individual changes I’ve made.
One last year 3 hurrah
The last major arc that happened during year 3 was probably being in an intense relationship for the first time. My other two partners were stable, but I saw them infrequently, something like a few hours once a week. (“Dating” is pretty broadly defined in my circles, especially with polyamory, and my grey-asexuality and grey-aromanticness don’t help clarify that.) However, I moved into a new group house early in my third year (in a somewhat messy house transition: come to think of it, pretty much all of my living arrangements in my PhD were great throughout and then had kind of terrible end transitions). Then the two of us happened to be living together, and coworking together since Tom had left for Princeton, and spending a whole lot of time with each other. I learned the joys of coworking, and from then on sought that out frequently, spent a lot of time decorating my room with his encouragement, and also did something like “falling in love”, veeeeeeery slowly. Also mildly, and at low intensity, because that’s how I roll. Of course, I was very concerned about this potential falling-in-love thing, just as I had been very concerned about dating and handholding and kissing and every single other thing, and I expressed my worry frequently and to a variety of kindly friends. I was particularly worried that I’d damage myself irreparably, acquiring a trait which seems weirdly common among the population, which is the desire to never not date again… And as I feared, that does seem to have happened. I haven’t tested it though, and with polyamory it is harder to not be single, so I’m not sure when I next will.
Year 4-5, PhD
Gosh this is long. Thankfully, much less happened in the last two years of my PhD. I was worried about funding for a while, since Tom had to stop funding me my fifth year. However, we got it worked out through a combination of delaying teaching to my fifth year, applying for fellowships (which would have worked if I’d gotten them), and the neuroscience program generously paying for a semester. I changed my name officially, got top surgery (eventually, it got messy but it happened), started dating a new person, escalated with that person, stopped dating one of my previous people, deescalated with the other person. Continued playing DnD, bought a 404-pound elliptical, developed hand and knee repetitive stress injuries (from typing and biking respectively). My friends and partner made an amazing birthday gift for me in 2020; I moved units in my house.
There was a global pandemic starting in year 4 spring, through when I’ll graduate. That was a really big deal.
However, it was actually a very small deal for me personally. Around January of my fourth year, I’d become kind of restlessly dissatisfied with large-group social events, in a way that felt like coming home to my college self much more than my grad school habits. I’d limited my party-type events to the excellent ones, and was basically trying to figure out how to withdraw from the social scene and stay home and work. I was working at home, in a great setup where I was coworking with two partners regularly. I was focused on getting the PhD done. I was also doing clinical work, but I’d finished most of my clinical work by the time the pandemic started. My community was extremely on top of the pandemic, and I was informed that the Wuhan virus might turn into a pandemic in late January-early February, so I’d been planning for a while. We started playing D&D outside, and I was living with two close partners. So I basically settled into pandemic life with extremely little trouble, and enjoyed it a lot more than almost anyone else I knew. The only two things that were severely disrupted for me were my top surgery, which was originally scheduled for March and I ended up chancing in June, and the housing situation at the beginning of the pandemic, which went poorly but worked out in the end. I thus had a very productive (…because I stopped doing most things), quite happy year, due to circumstance and time of life and personality. I feel very lucky. The pandemic literally couldn’t have happened at any other point in my life and been less disruptive for me personally.
Switching topics, something that I really enjoyed during my fourth year was starting as a part-time PhD student in Berkeley’s clinical psych program. Everyone was so nice (different culture from computer science, heh, though my experience is that academics are generally very nice) and kind to me, and I even got to see a patient and administer a set of psychological tests in the spring. I learned a lot about psychotherapy research, was challenged, and thoroughly enjoyed the experience. It was supposed to be a two-year program, but I actually asked to leave my second year. Not because I didn’t love it (I thought it was great) but because I felt I was running out of time to do my PhD and job applications, and I was worried about the time commitment. It cost the clinical psych PhD program money to take me on, for no benefit to them (especially since I left early), so I’m particularly grateful to the program and all of the wonderful people who helped and taught me there.
(I’ve been writing my Acknowledgements section for my dissertation, and I have so many people to thank. My mentors and peers are amazing and plentiful, and I’m especially grateful to Tom, who has been so continually supportive. I shielded my academic mentors from most of my career flailing, but Tom definitely got hit with some and was absolutely perfect pretty much every step of the way.)
Another experience I liked during my fourth year was starting and mostly-completing a research paper during the late spring and summer. I’d never felt so competent in research before– I was working with a team of great people who were very responsive, and I’d done this type of research enough that we just slammed through the research project. I also felt less insecure about stuff I didn’t know by that point, so that was pleasant. (I’d been doing introspective projects with “acceptance of interests” vibes– while I’m capable of learning things, sometimes I kind of innately don’t want to, and I’m learning to live with that.) I also hadn’t worked quite this closely with a team before, and hadn’t done research with peers in a while, so that was also great. Overall, I felt like a real scientist, which was fun!
My fifth year fall, I got to teach! I continue to really like teaching, though it’s definitely time-consuming. I would do that again if it made sense, opportunity-cost wise.
And now, for real, post-PhD jobs
Somewhere in the midst of fourth and fifth year, I started reconsidering about AI careers again. Not in the way I had in year 3, but more like my informational interviews from year 2– not particularly urgent, since I had a plan already (mental health + data science), but poking around. I stumbled upon AI policy, and thought that maybe that was something I could do. It’s not technical AI work, but it’s important AI work, and it benefits from a technical AI background. I applied for policy internships my fourth year summer, and totally failed to get any (that was true before the pandemic hit, and also true after). But AI policy was on my radar, and I started reading about AI governance at some point.
This was kind of crazy. I’d spent literal years bashing my head against a metaphorical wall, trying to get myself to do useful AI safety work. In the end, I’d concluded that my interests weren’t changeable, there wasn’t really anything I could do about it, and I’d do a mental health + data science job, which sounded innately interesting and helpful to me. I’d given up on the AI safety route. And then here I was, fourth and fifth year, flouncing around and deciding I’d just sneak on by, try another angle, maybe think about this policy thing, let’s just see if we can slide that one through my brain.
Policy didn’t seem like it would be a great fit. (This was not surprising. I don’t want to primarily work with people; I stride in the other direction when someone says I should try to convince or persuade people of things; I don’t want to move to Washington D.C.; I’m leery about disagreement, etc.) Policy research? I suggested to myself. Huh, maybe. Social science research? Look, I know I find social science research interesting, but that job basically doesn’t exist. But… AI safety thoooooough.
So I stared at a bunch of job openings, and got progressively more grumpy as I realized that none of them were quite what I wanted. How hard it is to employ a person like me, I grumbled. Here I was, wanting to work on AI safety, not wanting to do technical work, wanting to do empirical work, wanting to do social science work, wanting to study people, a few other specifications. Come on, man, this shouldn’t be that hard, someone should seriously be able to use me. My preferences weren’t unreasonable, what the hell.
And then, something clicked. Somehow, after all of those years of internal shaping, after storming towards and falling away, after all of the frustration of the exact issue, the “why can’t I just work on AI safety”, after angsting and anger and sadness and guilt for years and years… I thought: This is dumb. My preferences here are not unreasonable, and they’re fairly fixed. I know well enough what I want in a job, and the system is refusing to give it to me. I know I love clear-trod paths, and I wasn’t ready before, but I am ready enough now, and I will make my own.
So my fifth year and final spring, I’ve been applying for a position which doesn’t quite exist. It’s a combination of policy research and social science, focused on existential risk / AI safety. Academia’s where you go when you want to research something weird and it’s not monetizable, so I’ve been applying to postdocs. (Despite my claims for years that I was probably going to go into industry post-PhD when I graduated… well, we know my track record with this kind of thing!) I don’t know if the specific research proposals I’ve been sending out will work, actually. I’m still messing around, trying to get opportunities so that I can figure out if I enjoy them. Trying to figure out how I’m going to be helpful, given the interests and motivations I have. But the greater point, for me, is not that I’ve found my career, but that I now feel ready enough to make it up. I’m ready to do something outside of a ten-year plan, and the idea of making a career up doesn’t fill me with fear. I know myself well enough, and have enough practice at learning and trying things, that I expect to get there.
This feels so momentous to me! I’d been struggling for so long to be the kind of person who could do useful AI work, and be a leader, and think for myself, and care less about social judgment, be confident in research, be able to tolerate disagreement, all of these things, so many introspective projects that I could never quite figure out… and I made it in time for my PhD to finish. I am graduating fairly early, spending 4 ⅔ years total in my PhD, both because I’m out of funding and I want to get out and do the new work. I started out being someone who was mainly motivated by guilt and duty, strongly driven by shoulds and very sensitive to what everyone wanted me to do: I ended as someone for whom it felt actually feasible to make up my own job, maybe lead my own lab, who finds research useful in itself, who has figured out how to problem-solve and introspect and be as ambitious as I want to be. It took so LONG.
That said, all of this was wonderful and good… but, man, I still have to get accepted to a job. So there I was, in January/Feb 2021, excited but not that excited, because I was genuinely unsure how competitive I was as a postdoc. I did research, sure, but I was graduating a bit on the early side, and I messed around too much in my PhD to go for e.g. faculty jobs. I’d failed to get internships my last few summers, though that’s perhaps because I kept on attempting to switch fields (mental health/data science one summer, and I started applying late, and policy the next summer). So I was applying to a bunch of things, mostly postdocs because I had a dream, but also the deadlines are earlier than industry so I had to do postdocs first. I was trudging through that fairly unhappily, because applications aren’t much fun, and not knowing where I stood was making it hard to plan.
And then… I think I may have gotten my dream, doesn’t-really-exist, gonna-give-me-opportunities-to-try-targeted-things job. It’s on the horizon. I don’t quite have it yet, but it’s looking good, and their interest in me has shown me that I can achieve my dreams, I haven’t become uncompetitive as a PhD student, and please take me, I’m actually excited about this, oh man you don’t know how long I’ve been waiting for my finicky little interest stars to align. This isn’t as “omg blow me out of the water please may I work with Tom Griffiths” (will anything top that?) but it’s pretty damn good, and spawned all of the prose above about hopes and culmination. (This potential job is also half in AI and half in a totally unrelated existential risk field, whoops.)
(3/24/21 addendum: I didn’t end up getting offered the half AI / half unrelated existential risk field job, and instead was offered the full-AI version of this job, and took it :).)
In short, that’s where I am now. On the cusp of graduating, totally not done with all my research projects (we all have naive wishes), set with an army of great people ready to help me, with an awesome support network, great partners, great friends. A mind which is more aligned with itself, finally ready to try to meet this challenge I set for myself so long ago. A great mental toolkit, and excitement about the challenge. A maybe-probably job! Turns out I kind of like switching fields all the time– I didn’t know this, I thought I just happened to switch as a function of trying to figure out what I want. But there is something in me that apparently craves not knowing what’s going on at all and having to figure it out, get basic literacy, talk like the right kind of scientist, learn and grow, and this new maybe-probably job will definitely have that. I’m excited to learn about policy, I’m excited to read more about AI, I’m excited to maybe delve into a new field, I’m excited to stay home if I can swing that and cowork with my partners instead of having grand new social adventures. In many ways my base goals and ambitions haven’t changed at all– I’ve just gotten better at achieving them– and in a few ways, I’ve settled into myself and the desires look a little stranger than I imagined.
When I was a senior in college, I told people I wasn’t going to do a senior thesis, because it wasn’t necessary and would take up too much time. My best friend told me I was just the sort of person who would totally write a senior thesis, so I would. She was right: I wrote it. When I was (x years) into my PhD, I told people I wasn’t going to finish it, that I was going to leave. Over and over again, almost accidentally, here I am, pretty much done. Partway through my PhD, I said I’d go into industry after, not academia, I was pretty sure. Here I am, probably about to do a postdoc. My awareness of the options change, and my reasons, but my path just looks so straight on paper. If I end up becoming an academic professor, just like my high school senior self imagined, because I wanted a ten-year plan… gah!
And there’s a story of a PhD, mostly accurate, feels like forever. Hope you enjoyed.
Bonus commentary, from the writing of this essay: