The Shoggath Factor

Q is the heat loss, from which the building’s carbon footprint can be correlated.

U, the classic “fudge factor” is a function of many things, such as the building material, e.g., brick is a high number and fiberglass low.

A is area (bigger window, bigger heat transfer/loss).

ΔT, the difference between the temperatures outside and inside the building envelope.

§, the Shoggoth factor, the power structure that aims to reproduce itself at all costs. Ironically, this fudge factor is exactly the sort of thing LLMs are really good at figuring out.

dB, database size, a local hard drive approaches zero, the entire Internet approaches infinity.

ΔE, crudely expressed, distance to energy. Backyard, low-impact renewables approach zero, Big Data infinity. Of course power source delivery, and other quantities will factor in as well.

For example, I made a Zotero database of every article in the “AI and the Environment” Database, and the Modules (417 items), aka “~pit”, and linked my ChatGPT account using the Aria plugin. I spent $3.40 on prompts:

Here is one of the prompts.

~pit (417 items) how are server farms harmful?

Building search query (Input: how are server farms harmful?)
Searching Zotero database
Generating the reply (this may take some time...

Server farms, also known as data centers, can be harmful in several ways:

1. Energy Consumption: Server farms require a significant amount of electricity to operate. This high energy demand often leads to increased carbon emissions, especially if the energy is sourced from fossil fuels.
2. Water Usage: Cooling systems in server farms consume large quantities of water, which can strain local water resources, particularly in areas prone to drought.
3. Environmental Impact: The construction and operation of server farms can lead to habitat destruction and increased pollution. The materials used in building these facilities and the electronic waste generated also contribute to environmental degradation.
4. Local Community Effects: The presence of server farms can lead to increased noise and traffic, affecting the quality of life for nearby residents. Additionally, the demand for land and resources can drive up local costs and create socio-economic disparities.
5. Resilience and Risk: As highlighted by Luke Munn, the infrastructure of server farms is often imagined as resilient, but they are vulnerable to failures and disruptions, which can have cascading effects on the services they support.

These factors highlight the need for more sustainable practices in the design and operation of server farms to mitigate their negative impacts on the environment and communities.

References

(1) Gonzalez Monserrate, S. The Staggering Ecological Impacts of Computation and the Cloud. The MIT Press Reader. https://thereader.mitpress.mit.edu/the-staggering-ecological-impacts-of-computation-and-the-cloud/ (accessed 2024-12-22).

(2) Munn, L. Injecting Failure: Data Center Infrastructures and the Imaginaries of Resilience. Information Society 2020, 36 (3), 167–176. https://doi.org/10.1080/01972243.2020.1737607.

---

Returning to the Shoggath Factor, the entire design of this prompt engineering is contained within the 5199 course description, so a low value here.

The database, 417 items, is more than I could hold in my brain at any time. But it is miniscule compared to all of the slop LLMs usually have to go through. So a low value here too.

As for the energy source, although I want to take my family’s house “off the grid,” right now I am completely dependent on the ‘System’ of Automobility., so this may be a high number—I really don’t know where my electricity comes from.

Which returns me to the intelligent green building I began with. How many kilowatts can a smart home save? How many are required to run the AI making the building’s energy use more efficient? What good is lower your oil furnace use if somewhere on some server farm a bunch of GPUs are burning white hot?

Only if the AI for that home was completely dependent on local energy sources. At least then the ΔE could be minimized. A further step would be to run the house network on Raspberry Pis, shrinking the Shoggath Factor.