Introducing the LoRes AI Policy, version 1.0

LoRes has a particular theory of change that says that if we take technology, and bring it into grass-roots hands, in our local communities, which change our relationship to it. We want libraries and community centres and local food co-ops, not data centres and clouds and billionaires.

This same theory of change has informed our first stab at an “AI Policy” for contributions to this project.

If you’re interested in our approach to this, head over there and have a read, then check out some of our reasoning below.

What are our goals with this policy?

Our group has emerged out of many local people with varied political views but a shared commitment to a Solarpunk ethos. While Solarpunk is more an aesthetic and an inspiration than a rigorous social/technical analysis, it does point to the idea of building a more Just and Sustainable world through Grass-roots Making and Hacking.

As such, we have decided to push back only on big AI ecological sustainability and hyper-capitalist power concentration, and leave room for local makers and hackers to experiment with models they physically run on renewables.

Is it a good ideato allow even local, rewewably powered AI?

A world where we simply replace the AI data-centre boom with local GPUs in every building, running hot to apply LLM agents to all human problems, is still not a good world. It would in fact be less efficient than the data centres.

However, the same can be said of the LoRes project’s goal of moving cloud web hosting (currently at least 2% of the world’s carbon emissions) to local servers in community facilities. Our theory of change is that by moving the compute physically into community hands, we can have better and more democratic conversations about what use is truly important.

This theory of change has not been proven, and we need to remain critical in our analysis of how it is going.

What about the energy use, IP theft, obnoxious scraping and general big tech fuckery used to build even open-weight models?

An argument can be made that because many AI models were produced unethically, it is not ethical to use them. This policy states that it’s not ethical to financially support the makers of those models, but it does allow their use in limited conditions.

If models were a scarce commodity with terrible supply-chain ethics, like “blood diamonds” or ivory, it would be obviously that even using them would increase the market for their production.

However, since open-weight models can be copied onto many machines, a better comparison would be to scientific knowledge that was arrived at unethically, such as rocketry or medical research based on contributions by nazis during WW2. Western nations and capitalist oligarchs have certainly not been slow to use such tools.

The use of massive amounts of power to produce foundational AI models during a climate crisis was, and continues to be an unconscionable waste of resources.

The use of creative works without permission to produce these models is a complex subject. Copyright itself is often used to enrich large corporations rather than individuals, and the “free-use” right to remix them and study them is an important safety net. However, it’s absolutely true that big AI has actively set out to destroy the jobs of creative workers and our capitalist system has no replacement in mind for how art and creation will lead to fair and just livelihoods.

Given these historical harms, using existing models (without paying for them) in some situations is a difficult choice to make. This policy is influenced by the idea that when the AI bubble pops, these models might be amongst the only useful residue left to society in return for all this waste. The same can be said for many pieces of physical infrastructure around the world produced through injustice and polution.

A just future might see factories, dams and roads re-purposed and re-thought with people and planet in mind. But it is unlikely, at this stage in the polycrisis, to result in all the material produced by capitalism and colonialism torn down so we can start again. In some cases we may choose to do so, in other cases we may choose to find ways to remember the historical harms while using the products.

This project aims to join that conversation. We don’t claim to have the answers, but at this stage we are choosing to allow re-use of these already-created models, which necessarily means navigating their extractive legacy.