TLDR:
- blockchain technology might in principle be useful for EA and long-termism from a research and technology perspective
- optimizing the funding of public goods plays an role in building a better future
- designing the right incentives and preference elicitation mechanisms, as well as providing the right tools to realize them, is likely to be of importance for optimally funding public goods (or making any funding decision in general)
- a blockchain based platform we recently built called retrox might be of use in this regard and I would love to hear some of your feedback
The post is very long but is structured in such a way that you can pick out particular sections you might be interested in . The first few sections can be skimmed over, but the last two paragraphs are key. I would love to hear some feedback on whether the retrox platform - recently built by me and some friends - a blockchain mediated retroactive public goods funding framework could be relevant for EA, long-termism and perhaps even existential risk reduction. There are a lot of links to articles, videos and forum posts which I think are worth diving into if you are particularly curious.
Why is blockchain or crypto relevant to long-termism and effective altruism?
Recently a good argument has been made elsewhere for why there is actually a surprising amount of overlap between crypto economics and existential risk research (cf. this forum post). The basic tenet of the argument is that in both instances one is concerned with the regulation of complex and very smart systems with unpredictable emergent properties. Moreover, I see potential synergies between blockchain technology and long-termism. This link builds upon concepts and ideas coined by Eric Drexler and further developed by Kevin Owocki (founder of GitCoin who are a grants funding platform) or Juan Benet. The thought is that global coordination failures severely constrain the amount of resources we can devote to the things deemed crucial to ensuring that the long-term future of humanity goes well (the "Pareto-preferred future" - meaning a future that would be strongly approximately preferred by more or less everyone). The key to unlocking these resources is to design incentives in a way that nudge the global community in the right direction and the design of such incentives can hopefully be aided by using cryptoeconomic research and blockchain technology.
Moving away from a higher level ontology, as I'm sure many sceptics will call bullshit on what I claimed in the last line, I wish to showcase a particular framework and app which can, at least I hope, in practice aid with the move towards a Pareto-preferred future.
Public goods
Public goods play an important role for building a better future. They are non-rivalrous and non-excludable and encompass everything from access to information, education, open source software, reducing existential risk, to roads and healthcare. When assessing the importance of particular public goods one can easily weigh them with factors like neglected-ness and relevance to long-termism. I side with others (cf. The Case for Promoting / Creating Public Goods Markets as a Cause Area) who have argued that not having a market for public goods leads to suboptimal allocation of resources to them and a high degree of resource inefficiency. We should promote the creation of a public goods market which is a modular and adaptive process for allocating capital according to the collective desires of a population to further the realization of a Pareto-preferred future.
At the ETH-Global Hackathon in Amsterdam in April 2022, me and my team built a tool for on-chain (the term on-chain means recorded and notarized on the blockchain) retroactive funding of public goods. We believe that this is an important resource which can be used for the realization and testing of a public goods market.
Retrox
Incentive designs, nomination, voting and fund disbursement logic
The application we built is called Retrox (re-generative & retroactive public goods funding). Regenerative crypto-economics means that resources and capacities are sustainably increased in the long-run (cf. GreenPilled, where this concept is coined). Predicting which competing organizations, protocols, or individuals will provide the most value is difficult and therefore, some like Vitalik, have argued that since it is much easier to retroactively assess whether certain projects have added a lot of value to a particular cause or ecosystem. Retroactive public goods funding therefore, only allocates funding to projects or protocols with some track record of having already added value.
The way the Retrox protocol operates is as follows:
Phase 1: Round Initiation - someone initiates a round with a particular theme (e.g. education, open-source software, AI risk reduction research, etc.) A sum of money must be provided for these retroactive funding rounds by the initiator and anyone can top up the pool. Moreover, as it stands, the initiator provides a curated whitelist of experts who will eventually vote on different nominations.
Phase 2: Nominations - anyone can nominate projects which they deem worthy of getting funding in the context of a particular theme, they have to provide a project description and a process of staking is implemented to prevent spam.
Phase 3: Voting - once nominations are in, voting is executed. The voting scheme which leads to the best funding decisions, I believe, is one which is comprised of a set of experts with experience in a particular domain. Opening up votes to anyone is unlikely to lead to the most efficacious fund allocation. The notion of experience is broad and encompasses things like; proven track record of making good funding decisions in the past, technical knowledge, experience with particular markets and domains, good forecasting ability etc. As it stands the round initiators provide a list of experts who will vote quadratically, but this was simply the easiest way to get to protocol working initially and is subject to future revision. Individuals could also delegate their vote to those who they perceive to be experts. Although we have currently not implemented this functionality one can consider an algorithm which evaluates the past "performance" of voters with regard to how many of their votes reached projects which ended up adding significant long-term value.
Following a lot of recent discussion on optimal preference elicitation mechanisms, to determine the precise allocation of funds, I believe the best voting mechanism is quadratic voting (cf. Quadratic Payments: A Primer (Vitalik Buterin, 2019)) . Each expert (or as we call them "badgeholders") has a fixed number of total votes (100) which they can distribute to nominated projects as they like but their vote is re-scaled by taking its square root. This ensures, in contrast to linear voting mechanisms, that a diversity of opinions is sufficiently represented in the final weight of the funding pool value attributed to certain nominations. In principle any other voting mechanism can be tested and implemented (like for instance conviction based voting, holographic voting etc.).
Phase 4: Fund Disbursement - Once voting has been executed, all the funds in a particular round's pool are automatically dispersed in proportion to the vote weights for the different nominations. We have opted for a two-way disbursement scheme. Half the funds are delivered as a lump sum as a retroactive payment which and half the money is continuously streamed to the project over a fixed period of time. This is so that we don't run into the time-dependent incentivization issue of giving out large grants up fronts without ensuring that they are used effectively. The stream can be stopped once certain targets are not met. The precise parameters we set up can be adjusted and perhaps some optimum is eventually determined.
Caveats
It is obvious that retro-active funding of public goods struggles to support capital intensive projects which require a lot of up-front funding. However, for less capital intensive endeavors like information provision, some areas of research and software development this can certainly work as a stand-alone funding mechanism. For many other public goods however, one requires other funding streams to be able to support their continued provision.
Is this better than a classical grant giving committee? I would say so - we have a much more modular and easily amenable architecture for changing voting schemes and parameters. Moreover, the entire process is more open, transparent and efficient at allocating funds to the right projects. Conflicts of interest are also much easier to identify if votes are available to view on the blockchain.
What is the technology behind retrox ?
Without delving into too much detail, the retrox app currently exists as a single smart contract and a front-end which can communicate with the blockchain and said smart contract. The important idea is that all the data which is submitted by round initiators and nominators is fully transparent, immutable and available to view on IPFS (a distributed file storage system). Moreover, the funds which are used to initiate a funding round are sent to and stored in the contract address, such that they can be automatically disbursed once voting has been performed. All the voting data is also available to view on the blockchain and the system is very robust: only authorized experts or badge holder whose identities are linked to an address can actually submit any votes to the contract which are then processed. This type of application can in principle be deployed on any blockchain but we have opted for Optimism.
How could retrox improve general fund allocation and disbursement for any EA relevant cause?
One could envision a distribution of philanthropic aid using a public goods markets hosted on a platform like Retrox. As I have already hinted at above having a fully transparent and blockchain support framework for voting, nomination and round initiation means that conflicts of interest are more easily identified and alleviated. Moreover, the whole process is streamlined in terms of operational efficiency as funds are automatically dispersed and it is in principle very easy both to nominate a certain project, as well as to additionally fund or initiate a round for certain topics or causes.
How could retrox aid in the discovery of optimal parameters and funding mechanisms for public goods?
It is explained elsewhere on the forum, that there exist in principle, a large amount of incentive design tools and public goods funding mechanisms like for instance "merging free riders" and "matching funds" (cf. the linked post). But which mechanism leads to the optimal provision of a particular public good is very variable and unpredictable. Even if we have a good theoretical model, there always exists some uncertainty with regards to how this plays out in practice. We have opted for quadratic voting and retroactive/continuous streaming of funds as a good starting point. Running trial funding rounds which can be analyzed and evaluated retrospectively would be key in finding optimal solutions for particular contexts.
It is claimed in the aforementioned post, that large scale trials may be prohibitively expensive. On the contrary, if you have an on-chain platform which exists as a single smart contract, there is scope to easily initiate funding rounds with a particular voting or fund disbursement mechanisms with particular parameters (the amount given to this sort of experimental scheme doesn't need to be particularly large) as well as the ability to assess retrospectively whether the funding mechanisms have manifested themselves in the most optimal ways. This is only possible if there is full transparency with regards to the nominations, votes and eventual fund disbursement, as is the case for an blockchain based solution. There are a variety of parameters which can be analyzed; has a lot of additional external funding been amassed, how were the funds distributed to the different nominees and how did these nominated projects eventually do. In practice anyone can download the data and evaluate the perceived efficacy of a particular funding round.
These are some of my recent thoughts. I realize that the provision of public goods funding and associated incentive design and optimization is a horrendously complex problem, which in no way I claim to have solved. Perhaps a blockchain based solution isn't even the best way forward or useful in the first place. I do however think it has the potential in aiding in the discovery of optimal public good funding parameters and/or contributing to EA in general. Do people in the community see scope for using our platform (or a similar one) in a meaningful way? Are there any other things you would like to see implemented?
I would love to hear some of your feedback as well as your ideas. Please comment or get in touch with me: jan.ernst@physics.ox.ac.uk.
Retrox links: retrox.xyz, GitHub, EthAmsterdam, whitepaper
That’s such a cool project! My cofounder Dony just linked your post to me. I’m glad I didn’t miss it. Have you considered marrying the badgeholder voting with the S-Process? At least insofar as it is prospective (the future funding stream)? The optimism ecosystem seems like a great fit for this project. Are there plans for Optimism to use it for their retro funding?
Dony, Matt, and I have founded Good Exchange, funded from a grant through the Future Fund Regranting Program. We want to develop what we’ve termed “impact markets,” markets that replicate the incentive structures of the for-profit world for nonexcludable goods. We’ve had talks at Funding the Commons, you can read more about our plans in our forum sequence, and we’re currently running a contest that is an early implementation of our system.
We’re not currently building this on a blockchain, but maybe there are other synergies between our projects! We’d be delighted to see on Discord or have a call!
(Note that in the first link to your app the scheme is missing, and that I can’t access the whitepaper. Maybe a private repository?)
Thanks for your feedback! We haven't looked at the S-Process but it sounds promising and I'll check it out. We are in touch with Optimism and would love to play in part in future RPGF rounds but haven't reached any agreement yet.
Some great links there, will check them out. I've just joined your Discord, we should arrange a call.
Fixed the links.
Thank you for publishing this post. In which way is this different from what Optimism tries to achieve? Also, what if the public good is difficult to monitor? It is hard to observe reductions in existential risk. How will the protocol pay out if there is large uncertainty regarding the effects of an intervention, even afterwards?
It is not particularly different from what Optimism is trying to achieve, but importantly it is actually on-chain whereas Optimism's initial test round was done informally with a google doc to notarize votes and project nominations and disbursement was manual. As things scale this isn't really a great way to fund future rounds especially if more money is involved.
That's a good point, observing risk reduction is hard and it was a can of worms I didn't really open in the article. I am relying on sensible wisdom of the crowd type decisions to be implemented by groups of experienced assessors and forecasters. We'd like to come up with some broad traffic light system metrics to help guide voters, but ultimately this will require more research and development. What do you mean with difficult to monitor? Broad goods like "risk reduction research" may be difficult to monitor but individual contributions or projects which are nominated can still be assessed even if the overarching progress is hard to measure.
The payout is tied to the design decisions made at the round instantiation and the votes. The responsibility lies with the badge holders to assess those uncertainties and to potentially halt funding streams. See the discussion with ofer.
Suppose Alice is working on a risky project that has a 50% chance of ending up being extremely beneficial and 50% chance of ending up being extremely harmful. If the project ends up being extremely beneficial, will Retrox allow Alice to make a lot of money from her project?
What sort of of projects are you envisioning? AI research labs where there is a 50/50 chance as to whether they end up caring about AI safety? Retroactive funding means that one has the ability to assess the past impact of a particular project in a particular domain and then give out grants through quadratic voting. The ability to look at a project's impact in the past would aid with setting priors for how likely something is to be harmful in the future. If a project has the potential to be incredibly harmful then this should be weighed up by the badge holders who vote and less (or no) votes should be assigned to projects, depending on the probability and severity of the potential negative impacts in the future.
From a practical standpoint, the continuous stream of funds which extends well into the future can be stopped by the expert voters if the project is deemed harmful. In general - as it stands - the retrox platform does not have any built in logic which prevents any projects from being funded in the first place, but this is something which needs to be carefully considered and weighed up by those who vote on where the funds are allocated. I think this is where a lot of the "heavy lifting" is done and more careful consideration on who should be eligible to vote is perhaps required. Maybe you have some interesting ideas. Ideally you'd have an immutable and accessible record of people's qualifications, skills and past experiences which would allow one to pick out the right candidates.
Another idea would be to have a consensus mechanism between the expert voters which would allow projects to be "blacklisted" or blocked from being funded at all if the risk of them causing extreme harm is considered too great.
Suppose Alice is working on a dangerous project that involves engineering a virus for the purpose of developing new vaccines. Fortunately, the dangerous stage of the project is completed successfully (the new virus is exterminated before it has a chance to leak), and now we have new vaccines that are extremely beneficial. At this point, observing that the project had a huge positive impact, will Retrox retroactively fund the project?
That makes more sense now. Nothing inherent to the retrox platform would prevent this if the expert badgeholders agree to vote for the retroactive funding of the risky viral engineering project.
The fact that severe risks had to be taken should be factored into the assignment of the votes, i.e. how value was created. Incentivizing more high risk behavior with potentially extremely harmful impacts is undesireable. Retroactively funding a project of this nature would set a precedent for the types of projects which are funded in the future which I think would probably not lead to a pareto preferred future. The expected value trade-offs would be something like: value added for humanity by financially supporting successful but risky viral engineering project vs potential harm induced by incentivizing more people to pursue high risk endeavours into the future. I think the latter outweighs the former hence my previous hunch.