Ontology talk:9k/RD/Q46,99/critique

comment 1

I put a summary of my "violet Marxism" framework into the AI, and... the results were interesting. [1] I think I might not have given it enough information to actually figure out what I was talking about. Some things it said were fair, like that trying to create a level of meta-analysis above Marxist parties might not be dialectical materialism and might be a separate thing; between "violet Marxism", "violet Materialism", and "meta-Marxism" I don't entirely know what's the best name to call this thing any more. I guess I'll lean toward "meta-Marxism" for now.

Anyway. I think Socialism AI just accidentally tried to "disprove" itself. I asked it a question that boiled down to how two Marxist parties that do not have an equally good grasp on the class struggle but both believe that they're on the side of the proletariat (and that to some extent that makes their analyses against each other inherently correct) could possibly know which party is correct. And it gave me an answer where it tried to say that having a framework that attempts to organize information better than it otherwise would be in order to make better decisions is not as important as practice and observing real-world events. (I immediately had a lot of questions as to what it was trying to imply with that. It sounded way too similar to the way Stalinists argue and try to suggest that because early Trotskyism wasn't strong enough to defeat Stalinist parties and actually start killing bureaucrats that Trotskyists deserved to die unless they could build stronger parties.) And that left me thinking. The answer it gave me was dubious, and I gave it 1 star, but if that answer was even broadly right, it might have just told me that it was pointless.
You see, meta-Marxism has an interesting history. I first started developing it around the same time that I was trying to develop an alternative kind of formal logic that was more for organizing thoughts into categories than for trying to actively determine whether statements are true or false (given that you need information from material reality to really know if statements are true, and every other assignment of true/false is only a prediction). I had seen a large language model used to create a simple "game" where you would add two phrases to create another word or phrase that would logically follow. (Example: fire ∘ water = steam.) And I thought, huh, if you made tokens explicitly represent concepts rather than words and assigned the results of combinations based on actual materialist reasoning or real-world relationships between objects instead of just word frequencies, you could almost make that into a sort of very crude physics engine. And it spawned its own particular kind of formal logic that could organize concepts or propositions into syllogisms. (Example: labor commodity ∘ minimum wage = surplus value.) So, I had a logic process for mapping ideas, but that process was also basically me learning to reason the way an LLM does in a more power-efficient way as a human. I didn't consider it "foolproof" or a substitute for real-life tests, I just thought, maybe this could organize known true statements limited to their historical period and one day be more solid than Wikipedia. And then the LLM implies that if I go around using information collected from reality years ago but accurate to its specific situation to judge other information, that must be the wrong way, because you can't possibly get a better view of reality by just becoming better at organizing the information you already have and gaining new perspectives on it, you have to collect new real-world information. In the case of a controlled science experiment, I can see that; you can't just decide to run new tests on the same set of results in the same time period. But as for understanding history, that "stale data" problem is unavoidable because you can't run past events again. I feel like I've unintentionally created a perfect test case for whether this AI is worth anything. Meta-Marxism has a nasty problem of being made of a bunch of old historical data it only reinterprets to try to come back with theory, in a way Trotskyism-proper does not. LLMs have a nasty problem of being made of a bunch of days- or weeks- or months-old data that they only reinterpret to try to come back with theory. So, it seems like either my framework is wrong and the concept of Socialism AI is also wrong or my framework is okay and the concept of Socialism AI is okay. And I would not be offended if my framework is wrong, even despite the fact it and I have been developing side by side for the past 6 years; if Trotskyism is more impressive than the thing I created in order to supposedly help improve it, if climb up to where it actually is I don't have much to lose. But after the ICFI has said so much about 'augmented intelligence' and it's sounded more coherent than any argument against LLMs the thought that that could all just be wrong anyway has blown my mind. [2] [3]

I don't know if I can articulate this properly but I'll try. If Trotskyism has gotten to where it is based on practice, and the knowledge it gains daily through each event on the ground, and the active consciousness of organizations inside themselves of what is currently happening and will happen imminently or at least somewhat soon, but has not gotten its edge over other apparent forms of Marxism from some kind of bigger-picture view where it tries to emulate seeing it and its surroundings from "outside itself" [a concept in meta-Marxism], if everything is about continuous practical tests and anticipating them (though not about total spontaneity), then isn't every LLM inherently Idealist by that standard? While a human can apply the same basic reasoning process as an LLM — by applying old information to new situations only when it is relevant, etc. — the act of a human reasoning to create or apply theory will never be the same as an LLM doing it because an LLM can never be defined by practice in the way a party is. This may sound like it's splitting hairs a bit, because "humans" often get new real-world information from news articles, and LLMs can do that too. But there is a very particular kind of difference between humans and LLMs that applies only to organizations and not to individual people. An LLM can't actually do real-world tests on whether it is collecting more and more local parties into a larger world party, while parties themselves can do real-world tests of whether they are increasingly ordering humans into a larger organization. So, there is a very particular way in which the ICFI's claims that 'augmented intelligence' collecting all the world's information and propositions together will be a foundation for great improvements in industry, education, and class consciousness begins to seem dubious. It seems to me like the great achievement of Marxism doesn't involve collecting information or inferences together, but is fundamentally about the process of collecting great numbers of people together who are all interconnected material objects. Like putting so much emphasis on the wasted potential versus untapped capabilities of LLMs actually puts ideas over material objects and downplays the power of the Fourth International itself. Like even the processes of industry are somehow better explained by the process of organizing physical people into the Fourth International and how the creation of a world workers' state out of a lot of interconnected people and groups of people than they are by the "accumulation of knowledge".
That is just my attempt for the day to try to move myself closer to "actual Marxism" after the LLM called out my attempts to expand the Marxist method as potential 'petty-bourgeois sophistry'.

comment 1B

Okay, let me see if I can put this more clearly.

I (R.D.) have been loosely affiliated with an Australian organization called Socialist Alliance but not really part of it. Socialist Alliance is something of a Western-Marxist organization (so, revisionist). And I'm in the United States (!). Both the U.S. and Australia have really terrible organizations; they both have highly visible "Trotskyist" parties that are not actually Trotskyist and the United States has a bunch of other even-worse organizations like CPUSA that is Abolitionist (?) and the DSA that is just center-Liberal. It is very hard to actually go join an organization in the United States because most of the time all of them that crop up are worse than having nothing. And I am also disabled, somehow managing to get by but going through life wishing I could work and wishing I could be part of a workers' state despite being basically useless to anyone. (Not to mention how all the bigger businesses seem to be clustered in a few cities while what must be over half the United States and Australia is low-density cities whose population is half or a third petty bourgeoisie.) So, circa 2020, these conditions made me convinced that the only way to get a functioning Marxist party was a thing called "meta-Marxism" that could be applied by individuals and things smaller than Marxist parties to generate Marxist parties almost "out of nothing", because it felt like there was no other way for parties to become accountable than for every party to be analyzed from the outside by other parties and un-organized people. I know anything I say to justify meta-Marxism could simply be wrong, so I won't go further. But, a lot of its purpose was to try to be something that could stand between workers and Western-Marxism or postmodernist philosophies so people wouldn't fall for those any more. (Yes, I have yet to fix Socialist Alliance.)

So. This is the problem facing both AI and meta-Marxism: theory, organizations, and consciousness are different things. Theory can describe organizations but if it's describing them from the sidelines it arguably isn't Marxism. When theory is embodied inside an organization it becomes consciousness. A party containing theory that is tested against its surroundings is the consciousness of that particular object (group consciousness), or any interconnected group of workers (like the IWA-RFC could create) may be the consciousness of a class population. But LLMs cannot have embodied consciousness that connects individuals into a group around that consciousness, because they cannot have relationships to humans as-such. LLMs can only have a consciousness of parties that comes from outside all the parties, inasmuch as LLMs are made of ideas or statements (connected material ideas containing physical logic, to be fair) and cannot test anything practically in the sense a party does. LLMs are not necessarily worse than any existing Marxist text but they are reasoning entirely within the sphere of books and media, which it can be argued is a form of Idealism.

So, by that argument (which I admit to be based on meta-Marxism more than on a living Trotskyist party) it is impossible for Socialism AI to do Trotskyist analysis, and it can only create crimson Materialism similar to modern Stalinism or Bordigism. (In this case a very strange version of Bordigism that is actually able to evolve yet still contains static models inside.)

comment 1C

So, for instance, whenever the WSWS posts an article that was generated by Socialism AI,
(and I understand those are to demonstrate how the thing works so people know whether to use it or not)
there is a terrible possibility that every one of those articles is technically a Stalinist article, due to the methods used to write it. Or whatever negative label is most appropriate here.