Weekly Dwell #13
What does "AI alignment" even mean for nonprofits?
In a follow-up to one of my prior posts, I want to dive a bit further into alignment for AI, with a particular vantage point of how it might affect nonprofits. Overall, most folks in the space agree that some type of alignment is necessary, but exactly what to align to and why is where the complication kicks in. There are different camps of alignment that focus on different core aspects, just as there are increasingly different research areas that focus on specific populations, or research methods, or desired outcomes. It all depends on the vantage point of the person writing or speaking - which is essentially always the case. The person who is doing the proclaiming will bring their own bias to the conversation and proclaim it to be the best or most worthwhile. This is human nature and therefore in no way a new concept -- this is just a new medium in which its being highlighted within.
So what does this mean for nonprofits? Essentially, alignment is on the way to becoming more specialized, tailored to the needs of the organization and their communities. While alignment is not the same as model training, it is a component of the training, which can in turn be used as a way to ensure the alignment of the model connects to the needs and values of the organization. This is increasingly powerful as the AI itself will be better fit to serve those more directly in need for its services. Right now, the prominent AI products - ChatGPT, Bard, Gemini - are large language models (LLMs) that are aiming for a wide audience and trying to capture the entire market of AI-for-the-masses. What will start happening is these products will offer more opportunities to build custom models, with smaller, more specific skill sets. This could mean creating a chatbot that connects your community members to different resources, or sends out information when your programs change or have open applications. We've already seen these custom model builders rolled out by OpenAI as their enterprise model, and likely soon they will begin offering options for nonprofits and NGOs.
There are a huge host of potential options for these models within nonprofit organizations. So what can you be doing now? The best way to get started is by playing around with them yourself and seeing what their capabilities are. Start with free ones such as ChatGPT and Microsoft's CoPilot (through Bing) and see what you can accomplish with their assistance. If you like using the model for particular tasks, odds are your community members will like those functionalities as well. If there are certain aspects of your programs or outreach engagements that could be automated, that could be the work of a smaller language model.
There are still issues to be worked out (see the recent issue with NYC's AI), but a lot of potential on the horizon, especially as we align more with our values and specific topics, rather than attempting to make one single AI work with every possible situation.
Are you interested in learning more about small language models and how they could improve your nonprofit? Send me a message at kalie.mayberry@gmail.com and we can start the conversation about how to bring AI to your nonprofit.