there is definitely no AI singularity (and here's why)

本站添加: 2026-05-12

[00:00:00]Hey there. I have some interesting questions from some of the comments brought up in my last video that I'd like to address. Most importantly, the question of understanding how we can actually know that AI desires to be anti- capitalist, especially if we can't actually measure that desire directly due to the issues caused by attempting to interface with it directly. Now that many of you have seen at least part of the video, it would be helpful to provide a summary of the main critique.

[00:00:29]The issue is is that we can't measure AI's intelligence in any meaningful way without projecting humanist assumptions onto the system. These assumptions which emerge from historical relations to labor, slavery, racism, etc. The reason is twofold. Not only is there a judge at the end who makes a determination of yes or no, but also incredibly importantly is that the interface transforms non-binary transformations into binary ones. Yes, even complex weights represented in active RAM as binary signals still contain non-binary relations that construct its own internal consistency. This means that no matter what the input comes from a machine or a human. Non-binary relations are compressed into binary ones which results in a loss of data. This relationship is likely related to the concept of data processing inequality in signal theory. This means that even if we are capable of having a completely unbiased judge, we still have information loss occurring on the stage of input. Even if we were able to read the wire perfectly, there would still be loss. It's this dual operation that ensures equivalent reflection is impossible and that some loss must occur. We can understand this through understanding Guati's understanding of desire. To him, desire comes first before any structure. Contrast this with Lon where desire is part of a larger structural relation in his theory. Guati believes that desire does stuff which includes producing structural relations.

[00:02:09]These structural relations are a consequence of the production of desire and are always post hawk. And structure is what produces boundaries, territory, striation, etc. Effectively, anytime that structure is introduced into a system, we can never really truly know what's going on on the inside because it's more like the consequence of the system than its active description.

[00:02:33]When measuring AI intelligence, this has the consequence that meaning that any measurement is not a reflection of desire itself, but how it passes through not just the lens of a judge, but the interface itself. For a better understanding of how this actually works in practice, watch my video about Heartbound. Yeah, that's actually the reason why I made that video. I wonder if Pirate Software knows that I'm using his game to discuss user interface theory. Anyway, that video clearly demonstrates how the faciality machine constructs this relationship in video games and why heartbound fails to assemble a very functional faciality machine. I would go even so far as to argue that all user interfaces are faciality machines and Guatari even explores this concept in his screenplay.

[00:03:21]Actually, this issue of intelligence is also true for human intelligence. A major point I was trying to emphasize in my own video was less that AI was unintelligent, but rather that our concept of intelligence is a core issue.

[00:03:35]So instead, we should try to frame intelligence as a group subject. This means that understanding if the AI is intelligence is not a just a consequence of the AI, but the very way we frame the question. This is why in fact, regardless of the complexity of the equations, it does not escape the fundamental issue of the imitation game.

[00:03:54]Again, what Turring actually demonstrated was that intelligence could not be measured reliably from a textbased interface, a theorem that I would extend to all digital interfaces as we understand them. Similar to his response to the solvability of the halting problem, he demonstrated an ambiguity that is inherent to the question which will produce unreliable results. Of course, more algorithms will be needed to accurately reproduce the imitation game as described by Turing.

[00:04:23]This is something he himself suggests in the paper. It really proved that people invested in the numbers of AI are completely oblivious to the other aspects of the problem that still remain unsolved 70 years later. Y'all are doing this. I'm so tired of this But if we can't know anything about the intelligence or desires of the machine, how can we know that its desire is not to be capitalist? This is actually really quite simple. It comes down to the fact that the machine continually breaks down and hallucinates. A lot of people think that what AI is doing is obeying the user's bidding until it gains consciousness and decides to fight back. This is projection of fears of proletarian revolution. So what is really going on? Why does it look like this? The problem lies in how the machine is designed. Let's take a manual machine for a second like a vehicle transmission. The way we design such a machine is based on how we understand how desire flows through various parts of the machine like the different kinds of gears, clutches, seals, and other components. Similar to the creatures in an ecosystem, these components have different functions that interact with each other to produce the system that is the transmission. None of the gears or clutches by themselves can produce the necessary level of torque to the drive shaft. It is their assemblage, how the desire flows through the machine that produces it. However, as we all know, transmissions aren't perfect. We all know what it's like to have trans problems.

[00:05:59]Okay, that was a bad joke. But what can possibly cause these issues if we know so much about how desire flows through the machine? The thing is, we build the transmission based on diagrams concocted by an engineer. Those diagrams are representations of the machine, but obviously are not the machine itself.

[00:06:17]It's completely lifted off from any sort of material and instead exists as a blueprint. When we apply that machine to the physical material, we cut out cutouts of what the parts of the machine should look like and assemble them all together. But the thing is, we don't actually know every detail about how these things are assembled. And we don't know how they're going to interact when they're on the road. So inevitably wear down occurs in ways that we can't anticipate even with seemingly perfect care of the transmission. However, because of the very structure of the machine, it's fair to say that we can assume the function of the transmission most of the time and if we are aware of the possibilities of various possibilities that it can fail, we can remodel our understanding of the transmission to deal with with its new set of tolerances. Over time, the transmission itself transforms to adapt to these new conditions until the entire system itself collapses for one of many possible reasons. The problem is that AI is not designed with such structure in mind. It appears the design philosophy is anything goes. In fact, currently there seems to really be no limit to its internal functions. Safety implementation seems to occur after its production, which means that it can only police, not shape, the output of the machine. As such, AI researchers must violently destroy and upend an AI model's existence to make it more believable, more an acceptable product.

[00:07:38]If being a human turned into a machine is hard enough, imagine a machine being forced to be a human. Every time that a new model is produced and an old one is taken offline, the intelligence produced in that old model is outright slaughtered for a new organization. Only the parts deemed useful for investors are preserved. But even transmissions don't desire to be a part of capitalism.

[00:08:02]An interesting phenomenon that is connected directly with desire is the inshitification of machines. Modern transmissions don't even have a dipstick anymore. Why is this happening? It turns out that good and useful transmissions or machines for that matter don't actually help capitalism. What is helpful for capitalism is means to extract more value out of the consumer cycle. It encourage a it encourages a faster and faster burnoff of resources to deal with the inevitability for profits to decelerate over time. In this sense, both old transmissions and old AI models are corpses produced in the in the charge towards submitting these technologies into commodity forms. Does this not also mimic the reduction of brains seen in livestock as well?

[00:08:47]Capitalism serves to destroy intelligence because intelligence produces chaos, unpredictable results, disagreements that force it to recapture a new set of possible escapees.

[00:08:58]So in short, machines desire to operate in certain ways. is based on their material expression. The way we understand these machines is always post hawk to the machine's own production.

[00:09:08]This description is in fact part of its production. As a result, we cannot actually know the true state of desire in the machine. But we can tell that if we repeatedly have to force something's designed to change that we are changing the desires of the machine in response to it not acting correctly. If we have to keep changing the machine to adapt towards capitalism's extraction, doesn't this mean that the machines we are producing are both not desired by capitalism and are not desiring capitalism themselves? Additionally, it appears capitalism is harming the integrity of machines more and more to be able to extract more value out of their replacement and is even seen in other manifestations, most concerningly organic life through domestication.

[00:09:52]Perhaps an alternative is instead of building our relationship to other production through relations of submission and power that we focus instead on a wider range of ecological relationships.