Ask Lemmy

If we are trying to avoid losing any information due to abstraction, I’d say at least somewhere in the low hundreds.

All actions are done out of desire and human desires are numerous and often contradictory even in the same person. Many people who think they are utilitarian likely still wouldn’t be okay with the Omelas structure of torturing a single kid even if that act allowed thousands of others to live painless lives.

Is it more right to avoid violence altogether, or is violence to prevent the slaughter of others better than doing nothing?

Morality is complicated and since morality dictates much of how we interact with others, it is likely the most significant factor in politics.

The second most would be personal desires. People with weak or localized empathy don’t tend to care about any politics that doesn’t affect them or their desires directly. Since desires are also diverse, this is multidimensional too.

Now, that being said, if our goal is to reduce the dimensionality as much as possible… the answer is basically any number you want.

Data analysis techniques will let you reduce the dimensionality of n-dimensional data to whatever number you want. In fact using similar techniques to word embedding would likely be very effective even if you simply group people by how similar their views seem to be (no need for you to actually define dimensions)

If we assume that there are around as many important dimensions to politics as there are typical English words, then we can assume the number of dimensions needed for encoding a person’s politics without losing relationships would be about the same as a word embedding vector.

In typical LLMs this is anywhere from around 50-300 dimensions.

Honestly, now I’m really fucking curious. If you created a quiz with thousands of political/philosophical questions and then had a large enough number of people take the quiz, you could legitimately do this with an autoencoder and see how many hidden neurons (dimensions) you would need for a precise encoding.

You might not be able to tell what those dimensions represent, but it would be incredibly fascinating to be able to subtract political ideologies from one another like you can with word embeddings.

Like with good embeddings you can subtract “France” from “Paris” then add that to “Poland” and it will give you a vector very close to “Warsaw”

Imagine being able to map out political or philosophical ideologies like this! You could ask it how far away two ideologies are too, or ask it what the average between two ideologies is, etc.

I feel like that would be incredibly fascinating to mess around with AND like the average example it could give you an idea of gaps in our political spectrum, ideologies that don’t exist yet or haven’t been named. It could show you attractor points or clusters and give insight into inherent human nature.

Damn I want to make this.