I started a new blog! Come check it out and consider subscribing, the content will be similar to what I post here, but hopefully will be more polished.
https://splittinginfinity.substack.com/
My ARIA Application
Below is a part of my application for ARIA’s program director role. Though I was not selected, I hope that it may give other program directors some food-for-thought.
If you could direct £50m to drive a bold and focused scientific/technical step-change in human progress, what would you do?
1. Pilot demonstration generating carbon-neutral fuel from seawater.
CO2 is far more concentrated in seawater than in air, facilitating its capture. This CO2 can be converted into carbon-neutral fuels using green hydrogen. Making this process practical would provide Britain with an independent fuel source, remove the need for aircraft carriers to refuel, and reduce CO2 emissions.
£500K would be spent on techno-economic analysis and design, £3M would be spent on a pilot demonstration.
2. Alternative cattle feed
Alternative cattle feeds can reduce methane emissions while using less energy and land. Seaweed can be grown without land or freshwater and has been shown to reduce methane emissions when added to feed. Methane-eating bacteria are already used in feed and could be grown on carbon-neutral fuel.
I would consult welfare veterinarians and livestock scientists to establish standard protocols for humane research on methane emissions. £5M would be spent on pilot studies in UK sheep and cow farms.
3. Implosion fabrication for semiconductor manufacturing
Implosion fabrication creates nanostructures by isotropically shrinking a gel containing fabricated shapes. This may enable semiconductor manufacturing with cheaper equipment designed for older process nodes. I would spend £500K on proof-of-concept experiments shrinking a photomask using implosion fabrication.
4. Sleep Reduction and Narcolepsy Therapeutics.
Some individuals sleep less than 5 hours a night with no ill effects. Allowing people to safely sleep less would dramatically increase the size of the economy. Intranasal orexin is a promising approach for narcolepsy therapeutics and sleep reduction. I would scale up to a £10M pilot study using intranasal orexin to combat mild sleep deprivation in healthy individuals.
5. Enabling Protein Nanotechnology
Proteins can be used as medications (e.g. insulin, monoclonal antibodies), can manufacture nanoscale materials, and are extraordinary catalysts. However, precision production at scale is extremely expensive, and new production processes should be explored.
Purification is a large cost driver, and I would study the application of self-aggregating proteins based on blood clots to purify specific proteins from a complex mixture. Likely £100K for initial experiments and £2M for replications and larger tests.
Yeast and E. Coli are common platforms for protein production, but other possibilities should be explored. Microarray synthesis, direct synthesis using ribosomes, and bioreactors based on cancer cells (e.g. hybridoma technology) are all possibilities. Each method would need roughly £1M for testing and analysis followed by £10M for scaling the most promising method as a company.
New technologies enabled by extreme scale protein synthesis should also be studied. New paradigms for drug design, synthetic food, and nanoparticle synthesis may become feasible, necessitating follow-on research.
6. Finally, I would interview founders and consider the benefits of ARIA support for companies like Chemify (UK, programming language for chemistry and robotic synthesis), Quaise Energy (US, enhanced geothermal), and Reaction Engines (UK, Mach 5+ engines for travel and space launch).
Total: £34.1M
This budget was chosen to take many promising bets, with extra savings to scale successes and provide the flexibility to start new projects.
3 experts I would want to discuss this vision with further:
Professor Ying-Hui Fu, UCSF. Expert on the genetics of familial natural short sleep.
Dr. Christina Agapakis, Creative Director of Ginkgo Bioworks. Expert on engineering organisms to produce desired proteins and products.
Professor Ermias Kebreab, UC Davis. Expert on seaweed additives in cow feed for methane reduction.
Links #38
The world’s peak population may be smaller than expected
Tim LeBon on how altruistic perfectionism is self-defeating
You Don’t Always Have to be Rational
EDIT: a nice thread on companies working to automate IVF.
Venus Aerospace is working on a Mach 9 passenger plane based on rotating detonation engines. This technology would be pretty valuable for interfacing with space tethers as well.
Merged magnetic resonance and light sheet microscopy of the whole mouse brain
I love Andrew Huang’s approach to secure hardware (see here for a good overview), focusing on simplicity and end-user verification. Huang is working on betrusted which builds upon secure hardware and a secure OS to make a highly secure and verifiable messaging device. Huang is also working on a new method to non-destructively image chips.
While going down rabbit hole (ha) I came across some other projects that are interesting from a secure device standpoint:
Links #37
Alternative Approaches to Monetary Policy
LLM Modularity: The Separability of Capabilities in Large Language Models
Role Architectures: Applying LLMs to consequential tasks
CAIS-inspired approach towards safer and more interpretable AGIs
A Proposed Approach for AI Safety Movement Building
I’ve been enjoying the Model Thinking blog.
Supercharging Innovation. A nice intro to patent buyouts and some new policy proposals.
CLARITY for mapping the nervous system. Clearing system for helping image the brain. Could this be used in conjunction with expansion microscopy?
Thread on thermal batteries, which seem very valuable for making industrial processes green.
Flemtec is another company making green ammonia
Irradiant Technologies is using implosion fabrication to make optical nanostructures.
The ALOHA telerobotics system is impressive:
Learning Fine-Grained Bimanual Manipulation with Low-Cost Hardware
Nice video on rotating detonation engines. 3d printed rocket company relativity space seems well positioned to develop these commercially, if they decide to pursue it.
Church of Reality: Barbara McClintock on Scientific Mysticism and Plant Consciousness
Links #36
Gwern’s Project Ideas
Trellis is a ChatGPT-powered custom tutor and textbook companion
Here’s What It Would Take To Slow or Stop AI
Perfectly Secure Steganography Using Minimum Entropy Coupling
Palm cooling is neat. A similar system might be used to better adapt people to hot and cold climates.
Scientists create mice with two fathers after making eggs from male cells
The emerging landscape of single-molecule protein sequencing technologies
Large-Scale, Quantitative Protein Assays on a High-Throughput DNA Sequencing Chip
Picogram-Scale Interstellar Probes via Bioinspired Engineering
Iron II Selenide superconductors are pretty cool. Unlike other superconductors, the material is pretty simple and when prepared properly, can superconduct above liquid nitrogen’s boiling point, making it easier to study. Seems like a good system to iterate quickly on to improve performance or find practical applications.
Some recent reviews on using peptides and proteins to synthesize inorganic materials:
Biological and synthetic template-directed syntheses of mineralized hybrid and inorganic materials
Biomimetic mineralization based on self-assembling peptides
Designing Sequence-Defined Peptoids for Biomimetic Control over Inorganic Crystallization
Links #35
Attestation by Sarah Constantin has lots of food for thought about speech, cryptographic signatures, language models, and more. I wrote some comments here.
Chapman’s Better without AI
Metauni is a research seminar held in the metaverse. Cool!
Liz Stein has a nice review on hypersonic flight
Yellowstone Caldera Volcanic Power Generation Facility
Cooling Externality of Large-Scale Irrigation. If irrigation can cool local climates, then it might fix a lot of the extreme heat associated with climate change. See also: Jungles Everywhere.
You Are an Impossible Machine. Biology is nanotechnology, this isn’t appreciated enough.
Design of coiled-coil protein-origami cages that self-assemble in vitro and in vivo
Antifreeze proteins are neat. Can we add these to our blood to achieve cryosleep?
Your brain could be controlling how sick you get — and how you recover. I speculated about this a bit at the end of this post. If the brain has some control of the immune system and vice versa, we may be able to use immunotherapy for psychiatric conditions and psychotherapy (and other neuromodulation techniques) for chronic illnesses.
Mazindol is an off-patent stimulant that might be a good nootropic
Other points about RPGF
Here, I wanted to address some points I didn’t get to in my last post about retroactive funding
Verification
Retroactive funding has to verify that charities aren’t scams just like traditional funding does. It might be easier if you can look at past impacts and work done, but it doesn’t entirely remove the need to check.
Fortunately, the retro funder can punt on a lot of tricky philosophical issues and decide on them in the future. Just publishing a statement like “if you do something we think is really bad we won’t give you money” should be enough to move the prices of impact certificates.
Most retro funders will “want” to do something like this; nobody wants to give their money to a retro funder that is accidentally funding risky projects.
Quantifiable Impact
One criticism of traditional funding is that it produces incentives to look good, rather than actually be good. All you need to do is convince someone to give you money, not actually spend it well. A focus on verifiable, quantifiable good can address this problem, and retroactive funding is well suited for this.
However, there’s a risk of going too far; focusing too much on measurable things can mean missing big opportunities that are harder to quantify. Striking the right balance depends on the funders themselves. Retroactive funders aren’t obligated to focus on quantifiable impacts. They can and should incorporate subjective assessments, “warm-fuzzies”, and common sense into their decisions.
Preventing Bad Behavior
We can enumerate approaches to prevent net-negative projects by considering a bad-actors expected value calculation. The incentive to start a risky project can be reduced by lowering the size of the prize, increasing the cost of the project, and lowering the probability of success. Each suggests several ways to disincentivize bad projects.
Lowering the expected prize:
- Awarding fewer funds.
- Longer deliberation times (which reduces the net-present value).
- Dock the winnings of risky projects or dock winnings of managers who produced risky projects in other areas.
Increasing the cost of a project:
- Require projects to post collateral which can be taken away.
- After-the-fact fines for bad behavior.
- Reduce future funding for individuals involved in risky projects.
Lowering the probability of success:
- Award fewer total prizes (especially in risky areas).
- Set a higher bar for a project to be considered positive impact.
- Actively search for bad behavior (e.g. exposé markets).
Note that all methods to combat bad behavior have the side effect of reducing the incentive to do good. Funders must balance these forces carefully.
Other Interesting Research Areas
As a follow-up to my previous post on underrated research areas, here is a list of other ideas I find interesting:
Hash-functions Based on NP Problems: It would be nice to have a hash function known to be secure. It might be straightforward to do reverse 3-SAT problems where one constructs a 3-SAT problem from a given answer. If this is the case, it would be hard to invert the function (solve the 3-SAT problem) but easy to verify that the solution is correct (though I’m not sure if there would be lots of collisions). This would give proof-of-work systems a useful task to complete.
One-time Programs: A cryptographic primitive that allows a user to run a program once, this would unlock a number of interesting applications. One limitation is the requirement for a small amount of secure hardware. But there is some interesting work on making this possible with existing services.
Certifiable Random Numbers: It’s possible to generate truly random numbers using quantum entanglement and a small random seed, assuming that nothing travels faster than the speed of light. Can small groups of individuals generate private keys using a small amount of quantum computation?
Computational Mechanics: Epsilon machine reconstruction seems useful for understanding neural networks, interpreting them, and identifying agents in physical systems.
Dyson’s Eternal Intelligence: What conditions of our cosmology would allow us to complete an infinite amount of computation?
Post-horizon Computing: If you are willing to make a one-way trip into a black hole, you may be able to observe solutions to computationally hard problems and engage in faster-than-light communication (though that communication can never leave). What are the limits of computation once you pass behind an event horizon?
Reversible Computing: This would dramatically lower the energy costs of computation. This might be a key enabling technology for brain emulations.
Entanglement-based Energy Transfer: It’s possible to use entangled bits and a heat bath to transfer useful work purely via communication. Can we do this in practice? Could we transfer work into a black hole and reduce it’s entropy?
Physical Unclonable Functions: Useful for identity systems and secure hardware.
MM Wave Mining: Quaise is using gyrotrons to vaporize rock and drill deeper geothermal wells. Could the gasses they generate be collected to cheaply obtain useful minerals or build tunnels?
Particle Beams for Asteroid Prospecting and Redirection: Laser coupled particle beams can maintain small spot sizes for extraordinary distances. This makes me think: could we use these beams to ablate asteroids and observe their elemental composition from afar? If we can ablate one side of an asteroid, could we steer it out of orbit? This would make asteroid mining and defense far easier.
Terraforming Venus: Mars and the moon get all the attention, but Venus is another potential destination for life. The existing proposals to inhabit or terraform Venus are unsatisfying to me. Instead of settling for airships or bombarding the surface with asteroids, starlifting hydrogen from the sun and reacting Venus’ atmosphere into organic compounds seems better.
Solar Effects on Albedo: Solar panels absorb a lot of light and generate heat, if we build enough of them, will this significantly increase the Earth’s albedo? Will it have effects on local climates? Is it feasible to put infrared reflectors on solar panels to reduce heating?
Cryptographically-provable Geolocation: Can someone offer cryptographic proof of their location? This has applications for location-based authentication, criminal law, e-commerce, and integrating sensor data (see also: FOAM Maps).
Dynamic Social Choice with Changing Population: Can we extend Harsanyi’s aggregation theorem to situations where members of the group are changing (via birth, death, or immigration)? Can this social choice be dynamically consistent?
Decentralized NGDP targeting: Can non-fiat currencies implement NGDP targeting without a centralized authority? The extra currency could be considered Seigniorage that could subsidize public goods funding mechanisms.
Optimal Subsidy of Monopolies: Monopolies under-produce the goods they provide. The natural solution is to subsidize their sales. But what is the optimal subsidy? How can policymakers determine the optimal subsidy when the monopolist is incentivized to misreport marginal costs? What problems arise from difficulties in identifying monopolies?
Incentive-Compatible Public Debate: Debate styles like Oxford-style debate don’t incentivize people to honestly report their opinions before or after the debate, making it hard to know who won. Are there ways to incentivize honest revelation peoples opinions?
Social Epistemology: What’s the right way to update on evidence obtained through your social network? You probably shouldn’t update “all the way” given that the evidence might be filtered. Given knowledge of your source’s motivations, can you determine how true their statements are?
Language Model Preferences: Do language models exhibit consistent preferences? This is an important thing to know when considering the moral standing of neural networks.
Aligning Recommender Systems: A simple step towards AI alignment, can we build recommender systems that learn to satisfy our preferences without manipulating them?
Interconverting Neural Networks and Programs: Interpreting neural networks and verifying their properties would be a lot easier if we could convert them into readable code. Alternatively, compiling code into a neural network might speed up training or make it easier to run.
Neuromorphic Computing and Memristors: Memristors are a hypothesized 4th circuit component that complements inductors, capacitors, and resistors. There have been some implementations, but none are ready for deployment. What new circuits can we build? Could memristors enable true neuromorphic computing?
Solar Sail Materials: Cheap, lightweight, and robust solar sail reflectors are a key component of interstellar travel. What materials are optimal for this task? Could we design metamaterials that are lighter and more effective than a solid material?
Solar Sail Bussard Ramjet: the Bussard Ramjet is a neat idea, but the interstellar medium is likely too dilute for them to work in practice. Could providing additional acceleration with a solar sail allow them to overcome these limitations?
Programmable Droplets: Moving small droplets around on a chip electrostatic-ally is a more flexible route towards lab on a chip applications. It might also enable small-scale chemistry experiments.
Calcium-based Nitrogen Fixation: Elemental calcium reacts with nitrogen in air to produce calcium nitride, which reacts with water to form ammonia. Could this chemical looping synthesis be a cheaper way to produce fertilizer?
Silicon Dioxide Electrolysis: Can we electrolyze molten sand to produce high purity silicon for solar cells and semiconductors?
Methane Chlorination: We have an abundant supply of methane, but it is pretty chemically stable, making it difficult to react into complex products. Much research is invested into methane activation. One interesting approach is to use chlorine (something we can obtain from seawater) to produce chloromethane, which is more reactive. Can we perform this process cheaply and react chloromethane into complex products?
Plastic to Oil: Similar to trash gassification highlighted in the last post, converting plastics back into oil is one way to recycle atoms. This would make landfills an important resource.
Universal Atomic Force Field: Researchers use force fields to approximate the true motions of atoms for materials simulation and drug design. People have applied machine learning to this problem, but the scope remains limited. I want an AlphaFold-style effort to produce a force field that is fast, parallelizable, universal, differentiable, and accurate.
Inverse Materials Design Using Electron Density: Electron density is in theory sufficient to describe all properties of a material. Can we design materials with desired properties via gradient descent on the electron density?
Sonochemistry: Careful application of soundwaves can create transient, extremely high pressures and temperatures. Could we use this to perform high-pressure synthesis without reaction vessels designed for extreme conditions?
Lithium-air batteries: These could have roughly the energy density of gasoline, unlocking new applications in robotics, electric planes, drones, and shipping.
Superlenses: Metamaterials can be used to perform sub-wavelength imaging. Could we use this to more precisely pattern materials and make even better superlenses?
Implosion Fabrication and Expansion Microscopy: These are clever ideas involving the controlled swelling and shrinking of hydrogels to create densely patterned materials and image biological samples. Can we design polymers that more consistently change shape to a larger degree? Can we iteratively apply these techniques to obtain larger volume changes?
Fragment Screening and Dynamic Combinatorial Chemistry: By screening molecular fragments for binding on a target, we can estimate which combinations likely bind a target well. Dynamic combinatorial chemistry allows us to directly combine these fragments to produce new molecules, evolving strong binding molecules directly. Can we compliment this with in silico approaches?
Diatoms for Microstructured Metal Oxides: Diatoms can create tiny, intricate structures out of silica. Is it possible for them to deposit other metal oxides? Copying the process might give us close to nanoscale precision over certain types of matter.
Small-molecule vaccines: A Fentanyl vaccine has been developed in rats. Using the immune system to counteract small molecules creates some interesting functionalities. Could use this to clear environmental toxins, counteract drugs, modify hormone levels, and target non-protein regions of pathogens.
Cholesterol Vaccines: Using the immune system to clear atherosclerosis or reduce cholesterol levels seems like a promising way to prevent cardiovascular disease, the #1 cause of death in the developed world.
Anti-evolution Drugs: Co-administering antibiotics with a drug that inhibits pathogen evolution could significantly slow or stop the development of antibiotic resistance.
Reverse Ribosomes: The central dogma of biology is that RNA gets converted into proteins, never the other way around. But most processes in biology are reversible, and there are many examples of proteins with inverted functionality. For example, transcriptase converts DNA into RNA, and reverse transcriptase converts RNA into DNA. Ribosomes carry out the process of translating mRNA strands into peptides, so is it possible to do the reverse? Reverse ribosomes would be a spectacular feat of engineering (and easily worth a Nobel prize). They would allow us to sequence proteins using high-throughput genomics, create self-replicating proteins, and employ directed evolution on proteins.
DNA Data Processing: New sequencing modalities like single cell sequencing and ticker tape DNA promise to generate massive amounts of data. But rather than read all the DNA into computer memory and process it, can we use the latent information processing capabilities of DNA directly? Simple DNA computing in-situ might report the most common sequences in a sample, summarize trends, or identify novel sequences.
Self-aggregating Proteins for Precision Fermentation: One problem with precision fermentation is the cost of separating out your desired protein from the growth media. Could we attach clotting proteins from the blood to our desired protein, aggregate them into a large blob, and cleave them out later?
Extremely Cheap Antibodies: Antibodies can be used to seek-and-destroy pathogens, immunoprecipitation of desired molecules, and medical tests. They act as a general search function for biotechnology. But all of these applications are limited by the high cost to produce monoclonal antibodies.
Recombinant Albumin: Making human albumin would allow more people to rejuvenate their blood (and slow aging) while artificial chicken albumin is a step towards artificial eggs.
C4 Rice: Rice and other important crops use a less efficient type of photosynthesis. Can we modify crops to be more efficient?
Blood Metagenomics: Can we sequence all of the DNA and RNA in a sample of blood to identify infections? See signs of cancer? Learn about our microbiome? Sequencing the nucleic acids in the blood could give researchers real-time detailed data on a persons health. At the population level, pathogen movement and evolution can be tracked by sampling people who volunteer as honeypots.
Anticipating Pathogen Evolution: Pathogens change surface proteins to improve binding or escape immune defenses. Can we predict and prepare for the kinds of changes a pathogen might make?
Rapid Tests for Pathogenicity: A pathogen-agnostic test for how bad an infection makes you feel can be used to domesticate diseases. By quarantining individuals with particularly virulent strains, the disease should evolve to be more harmless.
Iterated Meiotic Selection: This would allow for significantly faster iterated embryo selection.
Neural Augmentation: Human brain organoids can be transplanted into rats and participate in cognition. Could we perform a similar procedure in humans to increase intelligence?
Stent-electrode Recording Arrays: Putting electrodes on a stent is a less-invasive way to record and stimulate the brain. What are the limits of this technology? Could multiple stents be used to record different parts of the brain? Could techniques like fNIRS or fUS be performed inside the brain without interference from the skull?
Free-running Circadian Rhythm: Some people have a “free-running” circadian rhythm where they can choose to sleep at any time. Can we help other people escape the circadian prison?
Love Drugs: Can we improve peoples relationships with a daily pill? Should we? What about physical attraction? Should people pursue chemical friendships?
Multiplexed Drug Testing: Can we accelerate drug trials by trying several substances on each subject simultaneously? Changing the drug cocktail over time and using sophisticated data analysis might allow us to determine the effects of many drugs all at once.
Ultrasound Therapy and Diagnostics: Ultrasound can be used to activate brain regions, allow drugs across the blood-brain barrier, and visualize organs. Miniaturizing the technology and making it cheap offers a new level of control over our biology.
Personal Cooling Devices: Extreme heat is one side-effect of climate change which disproportionately affects developing countries. Extreme heat increases mortality, lowers productivity, and hurts test scores. Portable, battery-powered personal cooling devices might be cheaper and more efficient fix than A/C.
Health Effects of Saunas: There are some interesting correlations between sauna use and mortality. Could they help by inducing a temporary fever that kills pathogens? Could it trigger heat-shock proteins which perform protein clearing? How do they lower cardiovascular disease?
Depression, Chronic Illness, and Sleep: Sleep deprivation reduces depression symptoms and suppresses the immune system. Depression symptoms are similar to being chronically ill. Short sleepers have unusually low rates of depression. What are the relationships between sleep, the immune system, and depression? Could immuno-supression fix depression? Or sleep-reducing drugs?
Chronic Illness and Psychotherapy: Is there a relationship between immune system and brain that could explain chronic lyme, chronic fatigue, long COVID and other chronic illnesses? Mechanical back pain can be effectively treated with Sarno’s method, a form of psychotherapy. Given a connection between the nervous system and the immune system, could psychotheraputic methods treat these chronic illnesses? Finding a connection between the nervous system and cytokines might lend itself to effective pharmaceuticals as well.
Social Contagion of Mental Illness: To what degree does discussing mental illness lead to its spread? Could social media be causing an increase in mental illness?
Moral Bioenhancement: Could we modify ourselves to be more honest, open-minded, altruistic, and virtuous?
Edit 2/15: I added a few more research areas
Links #34
Experiment is a crowdfunding platform for research
The Collective Intelligence Project
Notes on “Managing to Change the World”. Good general management advice.
Thread on proposal to expand Manhattan and a list of unbuilt structures.
Logic Through the Lens of Neural Networks
Looped Transformers as Programmable Computers
Black holes as tools for quantum computing by advanced extraterrestrial civilizations
I wrote a thread on the importance of automated food preparation and a thread of ideas for making flights cheaper.
Captura is pulling CO2 out of the ocean using seawater electrolysis.
Lonnie Johnson, the inventor of the Super Soaker, has founded a solid-state battery company and heat pump company
Dynamic combinatorial chemistry is a pretty neat idea.
Protein- and Peptide-Directed Syntheses of Inorganic Materials
A Catalog of Big Visions for Biology
What Should We Do With The New Fentanyl Vaccine? Paper here. The ability to produce a vaccine against a specific small molecule allows us to neutralize specific molecules in our body without the use of monoclonal antibodies.
The Prospect of Extracting Brain-Region-Specific Exosomes in the Human Bloodstream
Whole-mouse clearing and imaging at the cellular level with vDISCO. Could this be useful for cryonics?
A list of potential genetic enhancements
Gwern’s links on short sleepers
In vitro ribosome synthesis and evolution through ribosome display
I went down a rabbit hole with wikipedia’s lists of emerging and hypothetical technologies and came across some interesting papers:
Inhibiting the Evolution of Antibiotic Resistance. Could these anti-evolution drugs be used to slow mutation of pandemic diseases?
Addressing the Risks of Retroactive Funding
Summary: Retroactive funding has some advantages over traditional funding, but can potentially lead to bad behaviors such as supporting net-negative projects. I list some straightforward steps that can be taken to mitigate these risks. With these precautions in place, Retro funding seems like a valuable way to support altruistic endeavors.
Thanks to Dony Christie and Dawn Dresher for feedback on this post.
Introduction
Retroactive funding is an interesting way to finance impactful projects. Essentially, funders reward impactful projects after-the-fact, once the consequences of the project are known (similar to prizes or social impact bonds).
One important feature is to allow projects to sell shares of their future rewards (called “impact certificates”) in order to raise money for the project. By supporting a market in impact certificates, projects can receive funding proportional to their expected impact.
Retro funding has some advantages over traditional (or proactive) funding. First, funders get to reward projects they know had a big impact. It is often easier to identify how valuable an endeavor was after-the-fact, making it easier to support good causes. Second, it’s easier to find good causes when founders can point to the good deeds they’ve done. This reduces search costs and allows funders to consider a wider scope of projects.
This is an exciting possibility, but before we try a new funding approach, we should consider possible downsides and how to address them. Here, I want to review some of the potential risks of retroactive funding and identify steps funders can take to mitigate them.
Let’s define some terms:
Project: Any plausibly altruistic endeavor. These could be short, discrete events or ongoing programs.
Proactive or Traditional funding: This is the normal funding process; the project receives money roughly before or during the time an objective is completed. For ongoing projects, new funding generally comes in before a goal is achieved. For example, Givewell sends AMF funding before AMF buys and distributes bednets.
Retroactive funding: This is the process of rewarding projects after-the-fact for some successful altruistic endeavor. Funding typically comes after a project is completed, and could potentially come long after a project is finished. Prizes are one type of retroactive funding, where a project is awarded for achieving some goal after its completion.
Impact certificate: These are “shares” in a project’s future rewards. Projects can sell these shares to impact investors in order to raise funding and complete their goals. For example, a project might sell the rights to receive 1% of all future prize money in exchange for startup funding.
Impact market: The market for buying impact certificates. These certificates and shares are exchanged on a large, fluid market. Ideally, a well-supported market will “price in” the future value of a project, and finance impactful projects.
Where each funder participates on a project’s timeline.
Potential problems and their solutions
Risky projects
One of the benefits of retroactive funding is the fact that it encourages risk-taking. Impact certs can support projects that few people are interested in which nonetheless turn out to be successful. Because a project might be impactful, it’s often valuable to give it a try.
This has an unfortunate side effect: it also encourages people to take on risky projects with large downsides [1].
For example, a researcher might try to develop new bio-safety measures by doing gain-of-function research on a deadly pathogen. Even if this research is net-negative overall, this kind of work could be profitable in an impact market! This is because, if the research works out, the researcher can get a substantial prize from retroactive funders. On the other hand, if the research causes an outbreak, the researcher suffers no loss, since the least they can receive from retroactive funders is $0. This no-downside of risky projects is what gets people to try new things in the first place, but it can also lead to clearly-net-negative projects receiving funding.
While troubling, there are several ways to reduce this risk. The simplest approach is for the funder to avoid risky areas like biosafety. A blanket ban on a certain type of research is unambiguous and easy to enforce.
Alternatively, the retroactive funder can simply blacklist specific projects it thinks are net-negative. Even if these projects turn out well, the funder refuses to buy their impact certificates, which should lower their value dramatically and prevent the project from moving forward. This tactic would require funders to actively observe new projects in order to stop them early, which may be impractical.
A more sophisticated approach is to require projects to post some sort of collateral. In the case that the project ends up being harmful, the project forfeits their collateral. This takes away some of the “no downside” effect from retroactive funding and discourages risky projects.
Incentivizing Bad Behavior
There are several ways retroactive funding can accidentally encourage harmful behavior.
- To the degree that projects are competing for funding, there may be circumstances where projects try to sabotage each other.
- Alternatively, since projects are rewarded for fixing problems, there may be incentives to create new problems and then receive funding to solve them.
- Projects also have incentives to inflate the value of their work or manipulate impact metrics.
Note that these problems are not original to retroactive funding; traditional funding faces similar potential issues. Fortunately, there are few precedents for this kind of behavior in traditional funding, and I don’t expect it to occur much in retroactive funding [2].
That being said, since retroactive funding is so new, it seems valuable for funders to devote some of their resources towards rooting out bad behavior. In general, retroactive funders should pay attention to project activities, investigate accusations of bad behavior, and avoid areas where these kinds of problems are common. One solution is to use Exposé certificates to reward whistleblowers who reveal harmful activity. These are flexible enough to reveal many types of malicious activity and make the information public to all funders.
Unscrupulous Funders
Just like in traditional funding, less-careful funders may use retroactive funding to support ineffective or harmful projects [3]. For example, a retroactive funder may start rewarding risky-but-ultimately-successful projects in biosafety. This is unfortunate, and seems difficult to stop.
Hopefully, like in traditional funding, people who donate to retroactive funders will take these factors into account and prefer more careful funders. However, as the set of retroactive funders grows, the community should take steps to ensure that they collectively avoid funding risky projects or reward bad behavior [4].
One neat idea is to use prediction markets to estimate project risk ex ante [5]. Funders can run a prediction market on how impactful a project will be, with the market closing before the project has completed. This gives funders a snapshot of how valuable the project looked at the beginning, so retro funders can avoid projects that appear net-negative at their inception.
Are impact certs a good idea?
Despite these concerns, I still think impact certs are a good idea. The theoretical problems have straightforward solutions, and I think it’s unlikely that they will become major issues. In the worst case, it seems feasible to stop retroactive funding entirely if it turns out to be harmful.
It’s important to consider the right counterfactual when thinking about retroactive funding. Proactive funding is the current best alternative and has several issues of its own. Because retroactive funding provides the benefit of hindsight, redirecting some funding towards impact certs seems like an improvement on the status quo.
However, retroactive funding can’t solve all of the coordination problems inherent to charitable giving; this sets the bar too high. Instead, we need to determine whether retroactive funding delivers more impact than traditional funding in practice. I’m not certain whether it can deliver on this promise, but I think it’s worth a try.
Notes
- In reality, all participants are risk-averse, reducing the value of taking on risky projects.
- It could be argued that many charities use marketing tactics to draw more donors to ineffective causes. This doesn’t really involve inflating impact metrics as much as it involves convincing donors, and doesn’t seem like the kind of problem retroactive funding was set up to solve.
- One small consolation is that these funders are only rewarding projects that actually had a positive impact, though their actions may still cause net-negative projects to receive support.
- Better coordination can also address the “Retrofunder’s Dilemma”, where projects with widespread approval receive little funding because each funder expects someone else to foot the bill.
- H/T to Dawn Drescher for this neat idea.
Harsimony 