MATS is hiring for new research managers, so we figured that we would share a bit about what it is like to work as research managers for an AI safety training program. This is sort of a hybrid between "here is what it is like to work as a research manager" and "come work for us."

In the context of MATS, research managers (RMs) function as amplifiers, accelerating the growth and effectiveness of the burgeoning AI security, alignment, and governance fields. Some of us are based in Berkeley, California, and some MATS research managers are based in the UK, in London. While the researchers we support focus on pushing the frontiers of knowledge, we create the conditions for the researchers to thrive. We guide, coordinate, and support these scholars so that promising ideas can be cultivated into impactful research.

Daniel Filan

I'm a senior research manager at MATS, and on the side, I make a podcast about AI x-risk research. In my undergraduate education, I became persuaded that AI x-risk was an important global priority, and that research could reduce the risk level, and so pursued a PhD in AI at UC Berkeley under Stuart Russell. I ended up deciding that I didn't want to pursue a research career myself, but one of the things I most enjoyed about the PhD was being around researchers and talking about research. Figuring out how to turn that into a job led me to the research manager role at MATS. The bread and butter of this role is talking with scholars and their mentors, helping scholars stay productive, talking to scholars about what happens next, and helping scholars and mentors communicate. On top of that, I've been able to help run our mentor selection process, and I've been lucky enough to be included as a co-author on one of my scholars' papers.

One way to think about the impact of research management is as a facilitator to the scaling of the field of AI x-risk research. The field is growing quickly enough that providing high-quality mentorship is a serious bottleneck, meaning that the field can be amplified a lot by alleviating that bottleneck. Another way to think about it is that scholars on average estimate that they get 18% more done during MATS than they otherwise would due to research management. If each RM has something like 6 scholars, and if these numbers can be taken at face value (which is plausible but not obvious in my view), then a research manager is effectively adding the equivalent amount of value as an average MATS scholar, which is pretty good considering what our scholars accomplish.

In general, I've liked MATS as a way to stay connected to the AI x-risk research world and think it succeeds at amplifying the impact of that world.

Jeffrey Heninger

I've been interested in the development of future technologies for a long time, and especially interested in how to use them in a way that helps people and helps society. Prior to getting involved with AI safety, I worked in plasma physics trying to make fusion. I got introduced to AI safety through AI Impacts (co-founded by Katja Grace), looking at the history of technology to better understand how progress typically occurs and looking at better or worse ways people have tried to influence that progress. I don't think that any particular path of technological development is inevitable, nor do I think that it will go well by default - instead, people have the agency to influence what the future will hold. In addition to thinking about technology, I also have long enjoyed being a teacher. I spent several years teaching physics to undergraduate students, and there is something beautiful about people encountering and wrestling with new ideas.

Being a research manager at MATS exists somewhere between being a researcher and a teacher (or tutor). You touch a bunch of different research projects, although you are not the primary person pushing them forward. At the same time, being a research manager is supporting individual people with what they need. For some people, the most helpful thing is diving into the weeds for their particular research topic. For others the most helpful thing is doing project management and scheduling tasks to boost their productivity. There are still others for whom the most helpful thing is being someone they can talk to about their emotions regarding all of the challenges that come up in an intensive research program. Being a research manager is a great role for combining my interests in helping technology improve humanity, while playing a very hands-on role helping people grow and develop. 

Avery Griffin

I used to work as a research manager at MATS. My interest in AI was piqued by the performance of AlphaGo and AlphaZero, and later by Nick Bostrom’s book Superintelligence. I applied to AI Safety Camp in late 2021, which led to a grant that let me quit my job and focus on research. I actually first participated in the MATS program as a researcher, which ultimately led to me working at MATS as a research manager. For me, being a research manager is all about the people. It’s kind of like being a teacher, and being able to see scholars grow from being really anxious at the beginning to publishing great work after the cohort is really fulfilling. With organizations desperately trying to hire more quality AI Researchers, research managers are essential for scaling the field, just like supervisors, mentors, and coaches are essential for developing talent in other fields. The job has its challenges, as we must learn to handle intricate communications, juggle multiple project proposals, and track the progress of multiple scholars and multiple research directions simultaneously. All this is on top of having excellent EQ and a strong ability to model other people's thinking. One rare skill great research managers develop is the ability to quickly surface unspoken needs: in a 30-minute conversation, I need to be able to extract out how this person is doing, including what they’re thinking about that they aren’t necessarily telling you.

Join MATS as a Research Manager?

So is being a research manager for you?

Drawing on our experiences, a great research manager primarily needs to enjoy working with people. You don’t necessarily need to be an extrovert, but you do need to be friendly and caring. A genuine interest in helping others succeed is vital. In the same way a sports coach helps an athlete perform better on a sports field, a research manager helps a researcher perform better in their academic field. Adaptability and comfort with ambiguity are also really helpful traits; the research itself is often exploratory research and discovery research, but many aspects of working with an educational program also involve figuring out novel situations.

But you also need relevant domain knowledge and research experience in order to relate to provide helpful feedback and guidance to the scholars; a research manager can't simply be friendly and supportive without any technical background. Research managers often have advanced degrees in computer science, physics, applied mathematics, or a similarly technical field.

Finally, project management skills (stakeholder management, task decomposition, understanding motivation, clear communication about the current status & goals, being able to follow up on tasks) are very helpful for the core tasks of a research manager. Indeed, outside of the core tasks there are many opportunities at MATS to poke around in different projects, so talents in project management (alongside data analysis, people management, and other related skills) often end up being useful.

We think MATS is a great place to have a positive impact, but we reached out to some former colleagues to get their perspectives as well:

I loved working at MATS! Coming in to work was always fun because my colleagues were passionate about the organization's mission, and I learned so much about AI on the job. There is no better way to become familiar with technical AI safety research than supporting the world's leading researchers in the field, and there is no better place to do so than MATS. - Anonymous (Former MATS Employee)

At MATS, I saw the AI safety and security community truly come together. Researchers from academia, labs, and non-profits intersected in ways that broke the usual silos, making the scope and impact of the work genuinely unique. As an Amplifier, it was especially fulfilling to mentor and support scholars and then watch them grow into researchers who published high-quality work and went on to take on key roles after the program. What I valued most was the team’s decentralized and empowering culture. People were encouraged to take initiative, try new ideas, and really own their projects, which made the work deeply rewarding. - Shujaat Mirza (Former MATS Employee)

Working at MATS was just as educational for me as it was for the scholars  I was supporting. I was constantly learning and growing merely by being around brilliant researchers tackling such an important problem. There’s something truly special about helping researchers find their footing in AI research and eventually going on to publish amazing papers and join exciting organizations. I also got to work alongside a healthy, collaborative team that genuinely cared about the mission and about each other. It’s rare to find work where you genuinely feel like you’re part of building something that matters with a team that cares. - Anonymous (Former MATS Employee)

MATS is currently hiring for new research managers and senior research managers to join our Berkeley and London offices. Senior research managers lead teams of 3-4 research managers, as well as supporting our mentors & scholars. In addition to the research manager roles, MATS is also hiring for roles on the community management team, the program team, and the operations team. We are even hiring a specialized role focused on handing our compute needs. So if you could see yourself helping build the field of AI safety, we'd be happy to receive your application.