Published: March 2017
Future of Humanity Institute staff reviewed this page prior to publication.
The Open Philanthropy Project awarded a grant of £1,620,452 ($1,995,425 at time of conversion) in general support to the Future of Humanity Institute (FHI). FHI plans to use this grant primarily to increase its reserves and to make a number of new hires; this grant also conceptually encompasses an earlier grant we made to support FHI in hiring Dr. Piers Millett for research on biosecurity and pandemic preparedness, and this page describes the case for both this grant and that earlier grant.
The Future of Humanity Institute (FHI) is a research center based at the University of Oxford that focuses on strategic analysis of existential risks, especially potential existential risks from advanced artificial intelligence (AI). This grant fits into our potential risks from advanced AI focus area and within the category of global catastrophic risks in general, and is also relevant to our interest in growing and empowering the effective altruism community.
About the grant
Budget and room for more funding
FHI plans to use £1 million of this grant to increase its unrestricted reserves from about £327,000 to £1.327 million. It plans to use the remainder to support new junior staff members.
FHI’s annual revenues and expenses are currently around £1 million per year, and the bulk of its funding is from academic grants (which are lumpy and hard to predict). It seems to us that having a little over a year of unrestricted reserves will help with FHI’s planning. For example, it might allow FHI to make potential staff members offers that are not conditional on whether FHI receives a grant it has applied for, or to promise staff currently funded by other grants that there will be funding available to support their work when one source of grant funding has ended and another has not yet begun. Much less of this would be possible at £327,000 in unrestricted funds.
FHI’s other current funders include:
- The Future of Life Institute
- The European Research Council
- The Leverhulme Trust
- Several individual donors
Case for the grant
Nick Beckstead, who investigated this grant, (“Nick” throughout this page) sees FHI as having a number of strengths as an organization:
- Nick believes that Professor Nick Bostrom, Director of FHI, is a particularly original and insightful thinker on the topics of AI, global catastrophic risks, and technology strategy in general. Nick views Professor Bostrom’s book, Superintelligence, as FHI’s most significant output so far and the best strategic analysis of potential risks from advanced AI to date. Nick’s impression is that Superintelligence has helped to raise awareness of potential risks from advanced AI. Nick considers the possibility that this grant will boost Professor Bostrom’s output to be a key potential benefit.
- Nick finds FHI’s researchers in general to have impressive breadth, values aligned with effective altruism, and philosophical sophistication.
- FHI collaborates with Google DeepMind on technical work focused on addressing potential risks from advanced AI. We believe this is valuable because DeepMind is generally considered (including by us) to be one of the leading organizations in AI development.
- FHI constitutes a shovel-ready opportunity to support work on potential risks from advanced AI.
We believe that hiring Dr. Millett will address what we view as a key staffing need for FHI (expertise in biosecurity and pandemic preparedness). We have also been pleased to see that some of FHI’s more recent hires have backgrounds in machine learning.
We are not aware of other groups with a comparable focus on, and track record of, exploring strategic questions related to potential risks from advanced AI. We think FHI is one of the most generally impressive (in terms of staff and track record) organizations with a strong focus on effective altruism.
Risks and reservations
We have some concerns about FHI as an organization:
- Our impression is that FHI’s junior researchers operate without substantial attention from a “principal investigator”-type figure. While giving researchers independence in this way may have benefits, we also believe that researchers might select better projects or execute them more effectively with additional guidance.
- It seems to us that a substantial fraction of FHI’s most impactful work is due to Professor Nick Bostrom. Since Professor Bostrom’s own work is already funded and since he offers relatively limited guidance to junior research staff, the impact of additional funding may scale strongly sublinearly. (Our understanding is that Professor Bostrom’s allocation of attention is a deliberate choice, and does not necessarily seem unreasonable to us.)
- FHI has relatively limited experience with policy analysis and advocacy.
- Our impression is that FHI’s technical proficiency in machine learning and biotechnology is somewhat limited, which we believe may reduce its credibility when writing about these topics and/or cause it to overlook important points in these areas. We are optimistic that recent and forthcoming hires, discussed above, will be helpful on this front.
Plans for learning and follow-up
Key questions for follow-up
- What new staff has FHI hired, and what have they produced?
- What has been the most important research output from FHI’s staff?
- Have FHI’s additional reserves been useful, and if so, how?
- How are FHI’s collaborations with industrial AI labs going?
- Has FHI had success applying for other grants?
- What has Dr. Millett produced during his time at FHI?
Nick Beckstead and Daniel Dewey (our Program Officer for Potential Risks from Advanced Artificial Intelligence) previously worked as Research Fellows at FHI. Nick is currently an FHI Research Associate (though this role does not include any financial compensation). Nick is also a trustee of the Centre for Effective Altruism, along with Toby Ord, who is a Research Fellow at FHI. Niel Bowerman, FHI’s Assistant Director, also used to work for the Centre for Effective Altruism.
Carl Shulman is a Research Associate at FHI (paid, unlike Nick) and also consults for the Open Philanthropy Project. Carl is friends with Nick, and previously worked with Open Philanthropy Project Research Analyst Luke Muehlhauser at the Machine Intelligence Research Institute. Carl and Luke were not involved in our decision to make this grant.