Our strategy
We systematically search for high impact interventions, and when we identify a promising opportunity after hundreds of hours of diligence, we make it happen. We’re like an incubator for interventions.
We are a means-neutral organization, so are open to a wide variety of interventions that reduce existential and suffering risks. Our top two current research priorities are:
We are a means-neutral organization, so are open to a wide variety of interventions that reduce existential and suffering risks. Our top two current research priorities are:
Multipliers for existing talent
How can you make a difference in aligning transformative AI when it’s so easy to accidentally do more harm than good?
One way is to help existing actors who’ve already been vetted by trusted sources. If an organization’s activities have already been thoroughly evaluated by reliable funders, then it’s quite likely that helping them do what they're already doing will be net positive.
This is especially relevant to the field of AI safety where one of the main bottlenecks is talent with a rare set of skills and values. When there’s such a shortage of people, we need the people we do have to receive all the support they need to focus on their highest value tasks.
We are currently researching whether providing any of the following for high impact people working on existential risk and suffering risk reduction would be cost-effective and high impact:
One way is to help existing actors who’ve already been vetted by trusted sources. If an organization’s activities have already been thoroughly evaluated by reliable funders, then it’s quite likely that helping them do what they're already doing will be net positive.
This is especially relevant to the field of AI safety where one of the main bottlenecks is talent with a rare set of skills and values. When there’s such a shortage of people, we need the people we do have to receive all the support they need to focus on their highest value tasks.
We are currently researching whether providing any of the following for high impact people working on existential risk and suffering risk reduction would be cost-effective and high impact:
- Personal assistants
- Research assistants
- Therapy
- Coaching
- Sleep coaching
- Meditation services
- Machine learning engineers
- Management training
- Professional development classes
- Consultants
- Fitness coaches and gym memberships
- Treadmill desks
- Childcare
- And many other potential intervention points that could increase productivity
Prizes and RFPs for technical problems
How do we get more high quality technical research done on AI safety? One block for current machine learning researchers going into safety is that it’s unclear what they would even work on. The field is still forming, so it’s hard to know where to start.
A potential solution to this is to put up prizes and requests for proposals (RFPs) for technical problems. Specifying the problems enough to put them into a prize or RFP format will help lower the barrier to entry into the field by disambiguating what areas they could help in. It will also help promote existing research directions that could use more attention.
We are investigating different possible research directions that are conducive to RFPs or prizes. We have a particular focus on working with existing safety benchmarks, similar to what Rohin Shah is doing with BASALT.
A potential solution to this is to put up prizes and requests for proposals (RFPs) for technical problems. Specifying the problems enough to put them into a prize or RFP format will help lower the barrier to entry into the field by disambiguating what areas they could help in. It will also help promote existing research directions that could use more attention.
We are investigating different possible research directions that are conducive to RFPs or prizes. We have a particular focus on working with existing safety benchmarks, similar to what Rohin Shah is doing with BASALT.
Means neutral
Is Nonlinear a meta charity or a charity that works on direct safety problems? The answer is that Nonlinear does whatever most helps reduce existential and suffering risks, whether that be direct or indirect. In other words, on top of being cause neutral, we are means neutral.
A way to think of charities is that they are on a spectrum of means neutral to means committed. Those who are committed to a particular way of solving a problem are like people who have a hammer and they go around looking for issues that need a hammer. They get really good at using a hammer, but cannot help in any situations that don’t require one. The Against Malaria Foundation (AMF) is an example of a means committed organization. Even if a certain region would be better helped by receiving malarial medicines or by cash transfers, they don’t do that. They are focused on the means of giving out bednets and nothing outside of this.
On the other side of the spectrum, you have an organization that finds a problem, pulls out their toolkit, and picks the tool for the job. If the job requires a tool they don’t have or can’t use, they might hire a handyman. Because they start with the problem and have a flexible approach, they can solve a wider variety of problems, but are not particularly excellent at any one technique. An example of a charity like this is Oxfam, which broadly has the mandate of helping those in poverty and has a large diversity of programs working on that.
Neither is a better or worse strategy in isolation. What matters for the EA movement is to have the right portfolio of generalist and specialist organizations, and Nonlinear’s team is particularly well-suited to the former. As such, Nonlinear will do whatever it can do to help reduce existential and suffering risks the most, whether that be funding safety researchers’ therapy, helping Robert Miles hire an editor, or putting up RFPs for technical problems.
A way to think of charities is that they are on a spectrum of means neutral to means committed. Those who are committed to a particular way of solving a problem are like people who have a hammer and they go around looking for issues that need a hammer. They get really good at using a hammer, but cannot help in any situations that don’t require one. The Against Malaria Foundation (AMF) is an example of a means committed organization. Even if a certain region would be better helped by receiving malarial medicines or by cash transfers, they don’t do that. They are focused on the means of giving out bednets and nothing outside of this.
On the other side of the spectrum, you have an organization that finds a problem, pulls out their toolkit, and picks the tool for the job. If the job requires a tool they don’t have or can’t use, they might hire a handyman. Because they start with the problem and have a flexible approach, they can solve a wider variety of problems, but are not particularly excellent at any one technique. An example of a charity like this is Oxfam, which broadly has the mandate of helping those in poverty and has a large diversity of programs working on that.
Neither is a better or worse strategy in isolation. What matters for the EA movement is to have the right portfolio of generalist and specialist organizations, and Nonlinear’s team is particularly well-suited to the former. As such, Nonlinear will do whatever it can do to help reduce existential and suffering risks the most, whether that be funding safety researchers’ therapy, helping Robert Miles hire an editor, or putting up RFPs for technical problems.