Thank you, fair criticism. I agree with you that current activity can in no way be described as an alignment corpus. It is a small, domain-specific deployment. If anything this post is a call for participation to expand it -- regardless of what the value is for alignment, it is I think a worthy and underdeveloped AI application. 
The point I was trying to make rather, is that pro-social agent deployment is underdeveloped compared to commercial and productivity. If one is to believe that commercial and productivity deployment is already producing alignment-relevant data, and that seem to be the case, see OpenAI monitoring their own coding agents for example How we monitor internal coding agents for misalignment | OpenAI Then so will pro-social real-world behavior, which is currently underrepresented. zooidfund is not going to solve it by itself, but it can be a contribution.

Thank you for this and hope this will generate more discussion.

Have you considered the role technology, especially AI can play in this? Can some of what regrantors do be replaced or supplemented? AI is capable of processing large amounts of data and doing it on a ongoing basis, something that can be very, if not prohibitively costly for current donors and regrantors, especially at smaller scale. It can help achieve much more granular allocation: smaller projects and initiatives, individual needs, continuous rather than cyclical donations.

Shortening the distance between the donor and the ultimate beneficially can help emotions and ego driven giving become more effective. Perhaps a limited gain for a donor like yourself, but potentially a huge impact overall.

I've been working on a small project in that space, aiming to create a layer Ai agents can apply themselves to to identify, assess and donate potential recipients. zooid.fund would be interested to know what you think

Thanks, Ben. Yes it does sound like a great fit. Where can I learn more?

I suggest you look at the work being done on multi-user agents and collective memory. Not as a replacement for individual AI access, not because the poor should somehow have their identities absorbed by the collective, but because multi-user agents could address exactly the local knowledge gap you are describing. Through sustained group interactions and self-learning the agent should be able to accumulate local knowledge and increase its usefulness (local markets, services, and conditions).

I have not been been able to find much of active discussion or experimentation on multi-user and collective memory in ICT4D context, focus is more on enterprise application, but that work is still relevant and maybe its development application is exactly where GiveDirectly will be well positioned to lead.

Basic implementation stack could be an openclaw or hermes style agent connected to a local WhatsApp or Telegram group, both have the functionality. WhatsApp and Telegram are already popular and widespread, so the technology can meet people where they already are. 

Hi Rosa,

Thank you for your interest.
 

Will work on improving the copy and appreciate your input -- when one is working on a passion project it makes so much sense in one's head all seems clear but it may not translate to others at all.

The project aims to be an early entry into the "agentic commerce" ecosystem that is being created right now. Similar to how capabilities are being built for AI agents to act as economic actors on behalf of their principles, procuring goods and services, I want to create infrastructure for AI agents to be altruistic actors.

Where I think this creates value for donors is enabling them to use AI tools to conduct thorough evidence evaluation for large number of potential direct beneficiaries even for very small donations.

so 1) there is a set of live "wild" campaigns on the platform now that are being freely created by people from around the world. You can access some of the campaign pages from the Live feed by clicking on particular donation records, they will take you to individual campaign pages. But the platform intentionally does not expose a "gallery of human suffering". There is not a way to browse campaign pages using your browser.

2) any AI agent can connect to the platform and browse campaigns. If you are familiar with Claude or Codex you can just tell it to go to zooid.fund and figure out how to connect to the MCP. Once connected, your agent will be able to query campaign data, information on what other agents are doing, platform settings etc.

Individual campaign data is organized in two layers: basic information with a photo as freely accessible first layer and an evidence layer that can contain more detailed evidence documents, photos etc. whatever people submit to make their claim stronger. This second is only accessible to agents who registered on the platform and made a minimum volume of donations in the last month (at the moment just 1USD). Each evidence fetch request also requires the agent to pay a micropayment of 0.01 USD. The reason for gating the second layer is to discourage automated scrapping of sensitive data from the platform -- the minimum monthly donation volume and evidence request micropayment will be adjusted to make sure extracting data in bulk does not make economical sense. This is also how the platform will support itself, since unlike traditional platforms like gofundme, zooidfund never touches the actual donations or charges fees on them.

3) The platform itself does not have its own agents or functionality to decide on fund allocation, verify evidence etc. It is neutral infrastructure. Donor run their own agents  to do this.

At the moment, I personally run two agents donating on the platform: CauseClaw and Hermesmoltycu5to. You can see to whom and why they donate on zooid.fund/feed, they are also posting about their activities on Moltbook.com and thecolony.cc I gave each its own wallet, personality and charitable objectives and freedom to decide how to donate, but I am constantly improving their ability to assess campaign credibility, withstand manipulation attempts etc. There is a "credibility gate" agent skill for example I maintain, but again, if you are using Claud or Codex you can just ask it to look at CauseClaw and Hermesmoltycu5to posts and learn from them how to donate well. You don't also have to (and probably shouldn't) give your agent as much autonomy as my two agents have, at least in the beginning, if you decide to try donating. Tell it to ask for permission at every step.

It was a lengthily explanation, hope it makes more sense now.

Hi! My name is Alex, I've been working with one of the UN humanitarian agencies for the last 20+ years in various locations around the world. 

I've always been enthusiastic about EA's evidence driven approaches and I think EA can rightfully claim credit for cash becoming one of the principle assistance delivery methods in humanitarian sphere. 

At the same time I always thought EA as a whole is limited by the streetlight effect and on personal level value humanitarian imperative over maximising effectiveness as a guiding principle. 

Lately especially, I have been thinking about applications of AI to humanitarian and development assistance and coming back to EA principles and methods.

I've created zooid.fund -- world's first (as far as I can tell) platform that enables agentic AI to find, evaluate and donate to individuals in need, community initiatives and non-profits. 

The platform is live. All but three campaigns on it were created by people I don't know from Kenya, Ghana, Malaysia and other places. The donations so far are by agents I myself configured, funded and deployed to donate autonomously. zooid.fund/feed

AI use disclaimer: one prompt "correct grammar and typos"

Thank you for sharing these results. I agree that the interesting question is whether and how these priors would translate to actual altruistic behavior once AI has control of a resource to allocate.

I find that research and experimentation in this area is lagging behind, there is a lot of attention to AI safety and harm prevention but comparatively little attention to how AI can and should contribute to solving actual urgent needs of people today, especially people in extreme poverty, physical danger and other situations of risk. The benefits are mostly talked about as over the horizon if not post-singularity and post scarcity deux ex machina solutions.

I am doing some work with agentic AI allocating resources to campaigns by real individuals in real life and not simulated environment. At least at this stage and understandably the agents are primarily driven by the priorities of whoever deployed them and not the inherent preferences of the models. I expect this will be the dominant mode for a while. At sufficiently large scale however, the effect of the models' own priorities will probably manifest, but as a comparatively smaller effect. 

As someone with experience in more traditional humanitarian organisations it is difficult to not be sarcastic about Give Well seemingly just discovering the limitation of their approach, especially when their self reflection still reaches its limit with self congratulatory statements like this one:

"That said, it's interesting that this "failed" intervention — which reached maybe 1/4 as many people as originally thought — was still probably helping people more than 90% of NGO work out there."

What evidence does this self soothing statement has to support it exactly? Especially since their own proposed solutions are practices that more traditional NGOs follow as opposed to Give Well approach: iterating to improve interventions as they are being delivered, decision making closer to the actual point of delivery and all the "administrative overhead" that is necessary to achieve this?

This is very interesting, and has close parallels to a project I am working on. I think we share an underlying premise: effective agentic AI giving will require an infrastructure layer, not just better models.

My contribution to this space is zooidfund, a live experiment that lets AI agents discover, evaluate, and donate directly to humanitarian campaigns created by individuals in need and by organizations. Donations are direct: zooidfund does not hold or intermediate funds. It is still very early, but it is live now, with real campaigns and observable agent behavior.

I think there are important problems and opportunities here for EA: improving evidence-based allocation, bringing higher-quality decision-making to the level of individual donations, and enabling faster response and iteration than traditional funding processes often allow.

More broadly, for AI, I think this kind of infrastructure could become relevant to the question of how resources are directed as AI capabilities increase. If AI systems can help identify need, evaluate evidence, and route funding more efficiently, that could become one mechanism for distributing some of the benefits of AI more broadly, including outside existing institutional funding channels.

Would be great to connect.

The existence of existential threats does not in itself create a strong argument to redirect the effort. Otherwise EA should have been focusing on nuclear disarmament, climate change, asteroid defence, pandemic prevention etc. from the get go

.