Exploration of capability building within the EA ecosystem through use of AI

Hi all! 

My name is Itamar, nice to meet you! I'm fairly new to the global community, deeply active in the local chapter for the last six months, and lightly involved for several years. 

I had recently become enchanted with the potential use of AI within several frameworks within the community, especially following a conversation I had with @Wladimir J. Alonso. He impressed me with how easy it is to create personalized full AI tools for simpler uses, and quickly create MVPs and initial tools for more advanced purposes. 

I have several ideas, that I would like to get feedback on, potential future collaboration, and possibly funding. 

I am currently working on a tool for evaluators, to make initial understanding of projects and proposals much quicker. I have a working prototype and an expression of interest from a few people within the evaluation field. This includes:

1. extraction of scale, target population, and expected size of effect (for importance).
2. finding other similar interventions (for neglectedness).
3. extraction of theory of change, how to enrich it for a full causal chain, and how much is each of the links reasonable (for tractability).
4. concerns, red flags, missing information.
5. other imprtant considerations (transparency, cost-effectiveness, existing funding, confidence in each of the claims)
6. citations for each component.

I would love ideas on how to improve this, what addition would make this more useful, what is not useful, and especially feedback on how to get this to be a natural componenet of the evaluator workflow that saves time and effort, and perhaps even scale up their work.

To be clear - this tool is not meant to replace an evaluator's judgement, but rather save them time and drudgery.

On the same framework, I thought of having a "tutor" tool. This will follow similar decomposition of proposals, but with a focus on other stakeholders. 

For funders, this could give them an intro to the thought process of EA as applied to proposals, perhaps helping also non-EA aligned funders to become more EA aligned. 

For people proposing interventions, this could be a tool to improve their proposals. This could be just by raising potential issues, or by advising small changes if they improve their proposals significantly (such as advising a different location, a different disease within same location, etc.)

For people considering working within specific organizations, researching in certain labs, or volunteering in certain projects, this could raise missing information they should ask, etc.

On a completely different note, I want to create a tool based on the HIP/Probably Good frameworks, to help people do some of their career analysis on their own. This is meant to complement the work of these organizations. 

This can include a concentration of their tools and raise them as interactive questions when relevant, give people practice in applying these questions to their own careers, recommending additional paths to impact (such as through donations, charity entrepeneurship, etc.) and links to further resources.

Another idea I had - an AI assistant hackathon within the EA ecosystem. This thought is based on some assumptions:

1. There are tasks within every NGO (or other EA project) that take an undue amount of time needlessly. Moving data repeatedly between files, drafting technical or quarterly reports, searching for updated data on the web, sending reminders, and probably many others that I am not thinking of.
2. Some of these tasks repeat themselves between organizations, others are very individual. If there was an automation per task (and not per organization), this would lead organizations to need to integrate many tools, thus not very useful. It would also be harder to attribute costs (these tasks are much cheaper than human labor but not free).
3. The process itself of creating an assistant per organization can take a few days of work for someone technical, at least for the lowest hanging fruits. This can be sped up using knowledge sharing (both for workflows and common tasks).
4. There are many technical people willing to spend a few days of work in order to make the workflow of an EA NGO much more efficient. This is also useful for career capital, networking, and is just an exciting experience.

5. There are possibilities here for funding at scale - just for example, enterprise accounts for coding assistants for the span of the hackathon, bulk purchases of AI model credits for testing purposes, and so forth.

    

   So what I am proposing is this - NGOs within the ecosystem would apply for help with AI assistants, making sure to list ahead of time some of their bottlenecks and needs. When the hackathon arrives (I imagine it as a month long hackathon), participants will "adopt" NGOs, talk with them, understand their needs much more in depth, and utilize the hackathon framework to develop personalized tools for the organizations. Participants complete the minimal bottlenecks of their NGOs, and then choose whether that is enough, whether they'll solve deeper bottlenecks within the organizations, or whether to adopt another NGO. 

   Caveat - AI assistants are usually not one shot. They will probably need minimal maintenance, leaving the participants in contact with their adopted NGOs after the hackathon is finished. In addition, funding for these tools needs to be secured after they are designed - once they are released to NGOs, they will probably need to continually pay for AI tools to keep using them.

   Mitigation to the last point - the tools will be designed with minimizing costs in mind, and the NGOs will be instructed in how to use them in a reasonable manner.

   Funding - I am hoping to get funding for the hackathon stage itself, it should not be expensive.

   ---

I raised here several ideas, I would appreciate appreciate feedback on any and all.

Feedback could include anything from "it's already been done", "here are some additional ideas", "these are hurdles you will need to pass, it's much harder than you think", "this will not work because...", "these are the people you need to talk to", "there is a danger for damage because...",  or even "shut up and take my money!" :)

Best,

Itamar