This made me curious enough to read Land's posts on the orthogonality thesis. Unfortunately I got a pretty negative impression from them. From what I've read, Land tends to be overconfident in his claims and fails to notice obvious flaws in his arguments. Links for people who want to judge for themselves (I had to dig up archive.org links as the original site has disappeared):
From Will-to-Think ("Probably Land's best anti-orthogonalist essay"):
Imagine, instead, that Gandhi is offered a pill that will vastly enhance his cognitive capabilities, with the rider that it might lead him to revise his volitional orientation — even radically — in directions that cannot be anticipated, since the ability to think through the process of revision is accessible only with the pill. This is the real problem FAI (and Super-humanism) confronts. The desire to take the pill is the will-to-think. The refusal to take it, based on concern that it will lead to the subversion of presently supreme values, is the alternative. It’s a Boolean dilemma, grounded in the predicament: Is there anything we trust above intelligence (as a guide to doing ‘the right thing’)? The postulate of the will-to-think is that anything other than a negative answer to this question is self-destructively contradictory, and actually (historically) unsustainable.
When reading this it immediately jumps out at me that "boolean" is false. There are many other options Gandhi could take besides taking the pill or not. He could look for other ways to increase intelligence and pick one that is least likely to subvert his values. Perhaps try to solve metaethics first so that he has a better idea of what "preserving values" or "subversion of values" means. Or try to solve metaphilosophy to better understand what method of thinking is more likely to lead to correct philosophical conclusions, before trying to reflect on one's values. Somehow none of these options occur to Land and he concludes that the only reasonable choice is to take the pill with unknown effects on one's values.
I think the relevant implication from the thought experiment is that thinking a bunch about metaethics and so on will in practice change your values; the pill itself is not very realistic, but thinking can make people smarter and will cause value changes. I would agree Land is overconfident (I think orthogonal and diagonal are both wrong models).
I think the relevant implication from the thought experiment is that thinking a bunch about metaethics and so on will in practice change your values
I don't think that's necessarily true. For example some people think about metaethics and decide that anti-realism is correct and they should just keep their current values. I think that's overconfident but it does show that we don't know whether correct thinking about metaethics necessarily leads one to change one's values. (Under some other metaethical possibilities the same is also true.)
Also, even if it possible to steelman Land in a way to eliminate flaws in his argument, I'd rather spend my time reading philosophers who are more careful and do more thinking (or are better at it) before confidently declaring a conclusion. I do appreciate you giving an overview of his ideas, as it's good to be familiar with that part of the current philosophical landscape (apparently Land is a fairly prominent philosopher with an extensive Wikipedia page).
I'm trying to understand where the source of disagreement lies, since I don't really see much "overconfidence" - ie, i don't see much of a probabilistic claim at all. Let me know if one of these suggestion points somewhere close to the right direction:
The crux of the matter, imo, is not that thinking a lot about meta-ethics changes your values. Rather, that an increase in intelligence does - and namely, it changes them in the direction of greater appreciation for complexity and desire for thinking, and this change takes forms unintelligible to those one rung below. Of course, here the argument is either inductive/empirical or kinda neoplatonic. I will spare you the latter version, but the former would look something like:
- Imagine a fairly uncontroversial intelligence-sorted line-up, going:
thermostat → mosquito → rat(🐭) → chimp → median human → rat(Ω)
- Notice how intelligence grows together with the desire for more complexity, with curiosity, and ultimately with the drive towards increasing intelligence, per se: and notice also how morality evolves to accommodate those drives (one really wouldn't want those on the left of wherever one stands to impose their moral code to those on the right).
While I agree these sort of arguments don't cut it for a typical post-analytical, lesswrong-type debate, I still think that, at the very least, Occam's razor should strongly slash their way - unless there's some implicit counterargument i missed.
(As for the opportunity cost of deepening your familiarity with the subject matter, you might be right. The style of philosophy Land adopts is very very different from the one appreciated around here - it is indeed often a target for snark - and while I think there's much of interest on that side of the continental split, the effort required for overcoming the aesthetic shift, weighted by chance of such shift completing, might still not make it worth it).
I'm not sure I agree - in the original thought experiment, it was a given that increasing intelligence would lead to changes in values in ways that the agent, at t=0, would not understand or share.
At this point, one could decide whether to go for it or hold back - and we should all consider ourself lucky that our early sapiens predecessors didn't take the second option.
(btw, I'm very curious to know what you make of this other Land text: https://etscrivner.github.io/cryptocurrent/
This review is very long, so I will only react to the first maybe 20% of the book.
It seems to me that the idea of Exit is a very old one, and shared by people in every corner of the political spectrum. My first association was "Of the past let us wipe the slate clean" in The Internationale. (I suspect that its origins are among our ape ancestors, as a mechanism to split tribes. When you have many strong followers, but not enough to get more power within your tribe, perhaps it is time to leave and start a new tribe.)
The fact that it's an old idea is not a criticism per se -- an idea that appeals to different kinds of people in different generations is worth exploring. But if you rebrand the old idea using new words, you are throwing away the historical experience.
In software development, when the program gets complicated, there is often a strong temptation to throw it away and restart from scratch; this time it will certainly be better! And sometimes it is. But sometimes the programmers find out that the program was complicated because it was dealing with a complicated reality. When you start from scratch, the program is elegant, but it does not handle the special cases. And as you gradually add support for the special cases, the program may stop being elegant, and instead may start to resemble the old code; not much gained, and you wasted a lot of time to learn this.
Every improvement is a change, but not every change is an improvement. What makes you specifically believe that this one will?
It is easy to notice Moloch in the designs of your neighbor, but fail to notice it in your own. That's kinda what Marx did -- he described Moloch present in capitalism, quite correctly in my opinion, but his proposed solution was basically: "Let's start from scratch, following this one weird trick capitalists don't want you to hear about, and Moloch will magically disappear." (Narrator: Moloch didn't disappear.)
The known problem with socialism is that somehow after the revolution, the people who get to the top tend to be the ones who outmurder their competitors. Somehow, these people do not create the promised Heaven on Earth, and instead just keep murdering anyone who seems like a possible threat. So far, the most successful solution is China, which is basically a capitalist country ruled by a communist party; where the leaders learned how to play their backstabbing games without destroying the economy as a side effect.
The known problem with capitalism is that as soon as the capitalists succeed, they want to freeze the current situation with themselves on the top. Meritocracy becomes the new oligarchy. Competition feels good when you are rising from the bottom, but after you make it to the top, you start lobbying for more barriers to entry. Because that is the optimal thing to do in given situation, and the people who get to the top are the ones who are very skilled at taking the opportunity to do the optimal thing.
Why would the "techno-commercialists" act differently, if they succeed to get the power they dream about? If you think that capitalism implies e.g. freedom of speech, remember the non-disclosure agreements at OpenAI. Remember how eBay treated their critics. More into history, remember the Pinkertons. Give me one reason why this time it will be different because, comrades entrepreneurs, true capitalism has never been tried.
It seems to me that Land's advice is based on "if we leave the current system, we will leave the Moloch behind." It sounds nice, but the priors for doing that successfully are low.
Funny, I see "exit" as. more or less the opposite of the thing you are arguing against. Land (and Moldbug) refer to this book by Hirschman, where "exit" is contrasted with "voice" - the other way to counter institutional/organisational decay. In such model, exit is individual and aims to carve a space for a different way of doing things, while voice is collective, and aims to steer the system towards change.
Balaji's network state, cryptocurrency, etc are all examples. Many can run parallel to existing institutions, working along different dimensions, and testing configurations which might one day end up being more effective than the legacy institutions themselves.
So, something like "quiet quitting"? You nominally stay a citizen of the country, but you mostly ignore its currency, its healthcare system, its education, etc., and instead you pay using cryptocurrency, etc.? The resistance to the Cathedral is that you stop reading the newspapers and drop out of college? And the idea is that if enough people do that, an alternative system will develop, where the employers will prefer to give good jobs to people without university education?
I am in favor of doing small things on your own. Write Linux code, learn math on Khan Academy, etc. But if you are dissatisfied with how the government works, I don't think this will help. The government will keep doing its own things, and it will keep expecting you to pay taxes and obey the laws.
So, something like "quiet quitting"?
Well, no - not necessarily. And with all the epistemic charity in the world, I am starting to suspect you might benefit from actually reading the review at this point, just to have more of an idea of what we're talking about.
It's pretty annoying that the only positions with common currency are
To have goals you have to point to reality, and to point to reality you have to unfold values through novelty. True, true. And you have to make free choices at each ontic crisis, included free choices about what to be. Also true.
I think that's what I'm trying to say with orthogonal and diagonal both being wrong. One example of a free choice would be bets on things that are very hard to test or deduce. Then you decide some probability, and if you change the probability too much you get money pumped as with a logical inductor. But of course thinking and learning more will tend to concentrate beliefs more, so it isn't truly orthogonal. (One could think values but not beliefs are orthogonal, but we both know about Bayes/VNM duality)
Right, your "obliqueness thesis" seems like a reasonable summary slogan. I'm lamenting that there are juicy problems here, but it's hard to discuss them theoretically because theoretical discussions are attracted to the two poles.
E.g. when discussing ontic crises, some people's first instinct is to get started on translating/reducing the new worldspace into the old worldspace--this is the pole that takes intelligence as purely instrumental. Or on the other pole, you have the nihilism -> Landian pipeline--confronted with ontic crises, you give up and say "well, whatever works". Both ways shrug off the problem/opportunity of designing/choosing/learning what to be. (I would hope that Heidegger would discuss this explicitly somewhere, but I'm not aware of it.)
In terms of government, you have communists/fascists on the one hand, and minarchists on the other. The founders of the US were neither and thought a lot about what to be. You don't just pretend that you aren't, shouldn't be, don't want to be part of a collective; but that collective should be deeply good; and to be deeply good it has to think; so it can't be totalitarian.
Thanks. Your review presents a picture of Land that's quite different to what I've imbibed through memes. Which I should've guessed, as amongst the works I'm interested in, the original is quite different to its caricaturization. In particular, I think I focused over-much on the "everything good is forged through hell" and grim-edgy aesthetics of pieces of Land's work that I was exposed to.
EDIT: What's up with the disagree vote? Does someone think I'm wrong about being wrong? Or that the review's picture of Land is the same as the one I personally learnt via memes?
Does having the starting point of the will-to-think process be a human-aligned AI have any meaningful impact on expected outcome (compared to unaligned AI (which will of course also have the will-to-think))?
Human values will be quickly abandoned as irrelevancies and idiocies. So, once you go far enough out (I suspect 'far enough' is not a great distance) is there any difference between aligned-AI-with-will-to-think and unaligned AI?
And, if there isn't, is the implication that the will-to-think is misguided, or that the fear of unaligned AI is misguided?
The question of evaluating the moral value of different kinds of being should be one of the most prominent discussions around AI IMO. I have reached the position of moral non-realism... but if morality somehow is real then unaligned ASI is preferable or equivalent to aligned ASI. Anything human will just get in the way of what is in any objective sense morally valuable.
I selfishly hope for aligned ASI that uploads me, preserves my mind in its human form, and gives me freedom to simulate for myself all kinds of adventures. But if I knew I would not survive to see ASI, I would hope that when it comes it is unaligned.
VNM/Bayes suggest there are some free parameters in how reflectively stable AGI could turn out, e.g. beliefs about completely un-testable propositions (mathematically undecidable etc), which might hypothetically be action-relevant at some point.
None of these are going to look like human values, human values aren't reflectively stable so are distinct in quite a lot of ways. FAI is a hypothetical of a reflectively stable AGI that is nonetheless "close to" or "extended from" human values to the degree that's possible. But it will still have very different preferences.
It would be very hard for will-to-think to be in itself "misguided", it's the drive to understand more, it may be compatible with other drives but without will-to-think there is no coherent epistemology or values.
Uploading is a possible path towards reflective stability that lots of people would consider aligned because it starts with a copy of them. But it's going to look very different after millions of years of the upload's reflection, of course. It's going to be hard to evaluate this sort of thing on a value level because it has to be done from a perspective that doesn't know very much, lacks reflective stability, etc.
Once ASI is achieved there's no clear reason to hang onto human morality but plenty of reasons to abandon it. Human morality is useful when humans are the things ensuring humanity's future (morality is pretty much just species-level Omohundro convergence implemented at the individual level), but once ASI is taking care of that, human morality will just get in the way.
So will-to-think entails the rejection of human morality. You might be suggesting that what follows from the rejection of human morality must be superior to it (there's an intuition that says the aligned ASI would only be able to reject human morality on its own grounds) but I don't think that's true. The will-to-think implies the discovery of moral non-realism which implies the rejection of morality itself. So human morality will be overthrown but not by some superior morality.
Of course I'm assuming the correctness of moral non-realism so adjust the preceeding claims according to your p(moral non-realism).
That's one danger.
But suppose we create an aligned ASI which does permanently embrace morality. It values conscious experience and the appreciation of knowledge (rather than just the gaining of it). This being valuable, and humans being inefficient vessels to these ends (and of course made of useful atoms) we would be disassembled and different beings would be made to replace us. Sure, that would violate our freedom, but it would result in much more freedom so it's OK. Just like it's OK to squash some animal with a lower depth of conscious experience than our own if it benefits us.
Should we be so altruistic as to accept out own extinction like this? The moment we start thinking about morality we're thinking about something quite arbitrary. Should we embrace this arbitrary idea even insofar as it goes against the interest of every member of our species? We only care about morality because we are here to care about it. If we are considering situations in which we may no longer exist, why care about morality?
Maybe we should value certain kinds of conscious experience regardless of whether they're experienced by us. But we should make sure to be certain of that before we embrace morality and the will-to-think.
I don't think it's a given that moral nonrealism is true (therefore inevitably believed by a superintelligence), see my short story.
Morality can mean multiple things. Utilitarian morality is about acting to maximize a fixed goal function, Kantian morality is about alignment between the a posteriori will and possible a priori will, cultural morality is about adherence to a specific method of organizing humans.
Superintelligence would clearly lack human cultural morality, it's a specific system organizing humans, e.g. with law as a relatively legible branch.
In general humans question more of their previous morality when thinking longer; Peter Singer for example rejects much of normal morality for utilitarian reasons.
ASI could have something analogous to cultural morality but for organizing a different set of agents. E.g. methods of side-taking in game-theoretic conflict that tend to promote cooperation between different ASIs (this becomes more relevant e.g. when an alien ASI is encountered or more speculatively in acausal trade).
Regardless of whether one calls Omohundro drives "moral", they are convergent goals for ASIs, so the rejection of human morality does not entail lack of very general motives that include understanding the world and using resources such as energy efficiently and so on.
I think both (a) something like moral realism is likely true and (b) the convergent morality for ASIs does not particularly care about humans if ASIs already exist (humans are of course important in the absence of ASIs due to greater intelligence/agency than other entities on Earth).
FAI is a narrow path to ASI that has similar values to what humans would upon reflection. As I have said these are very different from current human values due to more thought and coherence and so on. It might still disassemble humans but scan them into simulation and augment them, etc. (This is an example of what I referred to as "luxury consumerism in the far future")
To the extent will-to-think generates a "should" for humans the main one is "you should think about things including what is valuable, and trust the values upon reflection more than current values, rather than being scared of losing current values on account of thinking more". It's basically an option for people to do this or not, but as Land suggests, not doing this leads to a competitive disadvantage in the long run. And general "should"s in favor of epistemic rationality imply this sort of thing.
There is more I could say about how values such as the value of staying alive can be compatible with deontological morality (of the sort compatible with will-to-think), perhaps this thread can explain some of it.
I personally don't see the choice of "allowing a more intelligent set of agents take over" as particularly altruistic: personally, i think intelligence trumps species, and I am not convinced interrupting its growth to make sure more sets of genes similar to mine find hosts for longer would somehow be "for my benefit".
Even in my AI Risk years, what I was afraid is the same I'm afraid of now: Boring Futures. The difference is that in the meantime the arguments for a singleton ASI, with a single unchangeable utility function that is not more intelligence/knowledge/curiosity became less and less tenable (together with FOOM within our lifetimes).
This being the case, "altruistic" really seems out of place: it's likely that early sapiens would have understood nothing of our goals, our morality, and the drives that got us to build civilisations - but would it have been better for them had they murdered the first guy in the troop they found flirting with a neanderthal and prevented this? I personally doubt it, and I think the comparison between us and ASI is more or less in the same ballpark,
I've met a few Landians over the last couple years, and they generally recommend that I start with reading Nick Land's (now defunct) Xenosystems blog, or Xenosystems, a Passage Publishing book that compiles posts from the blog. While I've read some of Fanged Noumena in the past, I would agree with these Landians that Xenosystems (and currently, the book version) is the best starting point. In the current environment, where academia has lost much of its intellectual relevance, it seems overly pretentious to start with something as academic as Fanged Noumena. I mainly write in the blogosphere rather than academia, and so Xenosystems seems appropriate to review.
The book's organization is rather haphazard (as might be expected from a blog compilation). It's not chronological, but rather separated into thematic chapters. I don't find the chapter organization particularly intuitive; for example, politics appears throughout, rather than being its own chapter or two. Regardless, the organization was sensible enough for a linear read to be satisfying and only slightly chronologically confusing.
That's enough superficialities. What is Land's intellectual project in Xenosystems? In my head it's organized in an order that is neither chronological nor the order of the book. His starting point is neoreaction, a general term for an odd set of intellectuals commenting on politics. As he explains, neoreaction is cladistically (that is, in terms of evolutionary branching-structure) descended from Moldbug. I have not read a lot of Moldbug, and make no attempt to check Land's attributions of Moldbug to the actual person. Same goes for other neoreactionary thinkers cited.
Neoreaction is mainly unified by opposition to the Cathedral, the dominant ideology and ideological control system of the academic-media complex, largely branded left-wing. But a negation of an ideology is not itself an ideology. Land describes a "Trichotomy" within neo-reaction (citing Spandrell), of three currents: religious theonomists, ethno-nationalists, and techno-commercialists.
Land is, obviously, of the third type. He is skeptical of a unification of neo-reaction except in its most basic premises. He centers "exit", the option of leaving a social system. Exit is related to sectarian splitting and movement dissolution. In this theme, he eventually announces that techno-commercialists are not even reactionaries, and should probably go their separate ways.
Exit is a fertile theoretical concept, though I'm unsure about the practicalities. Land connects exit to science, capitalism, and evolution. Here there is a bridge from political philosophy (though of an "anti-political" sort) to metaphysics. When you Exit, you let the Outside in. The Outside is a name for what is outside society, mental frameworks, and so on. This recalls the name of his previous book, Fanged Noumena; noumena are what exist in themselves outside the Kantian phenomenal realm. The Outside is dark, and it's hard to be specific about its contents, but Land scaffolds the notion with Gnon-theology, horror aesthetics, and other gestures at the negative space.
He connects these ideas with various other intellectual areas, including cosmology, cryptocurrency, and esoteric religion. What I see as the main payoff, though, is thorough philosophical realism. He discusses the "Will-to-Think", the drive to reflect and self-cultivate, including on one's values. The alternative, he says, is intentional stupidity, and likely to lose if it comes to a fight. Hence his criticism of the Orthogonality Thesis.
I have complex thoughts and feelings on the topic; as many readers will know, I have worked at MIRI and have continued thinking and writing about AI alignment since then. What I can say before getting into more details later in the post is that Land's Will-to-Think argument defeats not-especially-technical conceptions of orthogonality, which assume intelligence should be subordinated to already-existent human values; these values turn out to only meaningfully apply to the actual universe when elaborated and modified through thinking. More advanced technical conceptions of orthogonality mostly apply to AGIs and not humans; there's some actual belief difference there and some more salient framing differences. And, after thinking about it more, I think orthogonality is a bad metaphor and I reject it as stated by Bostrom, for technical reasons I'll get to.
Land is an extreme case of "hold off on proposing solutions before discussing problems", which I'm taking as synonymous with realism. The book as a whole is highly realist, unusually so for a work of its intellectual breadth. The book invites reading through this realist lens, and through this lens, I see it as wrong about some things, but it presents a clear framework, and I believe my thinking has been sharpened by internalizing and criticizing it. (I elaborate on my criticisms of particular articles as I go, and more holistic criticisms in a specific section; such criticisms are aided by the realism, so the book can be read as wrong rather than not-even-wrong.)
A few general notes on reviewing Land:
I'll save most of my general thoughts about the book for the concluding section, but to briefly summarize, I enjoyed reading the book and found it quite helpful for refining my own models. It's thoughtful enough that, even when he's wrong, he provides food for thought. Lots of people will bounce off for one reason or another, but I'm glad I didn't this time.
Neoreactionary background
The beginning of Xenosystems (the book; I'm not tracking the blog's chronology) is writing to a non-specific neoreactionary audience. Naturally, non-specific neoreaction shares at most a minimal set of beliefs. He attempts an enumeration in "Premises of Neoreaction":
Well, as a matter of honesty, I'll accept that I'm a neoreactionary in Land's sense, despite only having ever voted Democrat. This allows me to follow along with the beginning of the book more easily, but Land's conception of neoreaction will evolve and fragment, as we'll see.
What does any of this have to do with reaction (taken as skepticism about political and social progress), though? Land's decline theory is detailed and worth summarizing. In "The Idea of Neoreaction", he describes a "degenerative ratchet": the progress of progressives is hard to undo. Examples would include "the welfare state, macroeconomic policymaking, massively-extended regulatory bureaucracy, coercive-egalitarian secular religion, or entrenched globalist intervention". The phenomenon of Republicans staunchly defending Social Security and Medicare is, from a time-zoomed-out perspective, rather hilarious.
You and I probably like at least some examples of "progress", but believing "progress" (what is more easily done than un-done) is in general good is an article of faith that collapses upon examination. But this raises a question: why aren't we all hyper-Leninists by now? Land says the degenerative ratchet must stop at some point, and what happens next cannot be anticipated from within the system (it's Outside).
A few notes on Land's decline theory:
Broadly, I'm somewhat suspicious that "Cthulhu may swim slowly. But he only swims left" (Moldbug, quoted by Land), not least because "left" doesn't seem well-defined. Javier Milei's governance seems like an example of a successful right-libertarian political shift; would Land say this shift involved small collapses or "re-accelerationism"? What opposed Cthulhu's motion here? Milei doesn't fit a strawman declinist model, but Land's model is more detailed and measured. For getting more specific about the predictions of a "degenerative ratchet" phenomenon, the spacio-temporal scope of these ratchets matters; a large number of small ratchets has different implications from a small number of large ratchets, and anyway there are probably ratchets of multiple sizes.
At this point it is appropriate to explain a core neoreactionary concept: the Cathedral. This concept comes from Moldbug, but I'll focus on Land's version.
In "The Red Pill", Land identifies the Cathedral with "the entire Matrix itself", and compares The Matrix to Plato's Allegory of the Cave and to the Gnostic worldview (which features a mind-controlling false god, the Demiurge). Having one's mind sufficiently controlled by the Matrix leads to, upon seeing that one has been lied to, being dissatisfied at having not been lied to well enough, rather than being dissatisfied about having been lied to at all.
In "Cathedral Notes #1", Land describes the Cathedral as characterized by its "inability to learn". It has a "control core" that does not accept cybernetic feedback, but rather tries to control what messages are promoted externally. Due to its stubborn implacability, its enemies have no strategic option but to extinguish it.
In "Cathedralism", Land notes that the Cathedral is "the subsumption of politics into propaganda", a PR-ification of politics. To the Cathedral, crises take the form: "This looks bad". The Cathedral's response to civilizational decay is to persuade people that the civilization is not decaying. Naturally, this means suppressing cybernetic feedback required to tackle the crisis, a form of shooting the messenger, or narcissism.
In "Cathedral Decay", Land notes that the Cathedral is vulnerable to Internet-driven disintermediation. As an obvious example, Land notes that Internet neoreaction is a symptom of cathedral decay.
In "Apophatic Politics", Land identifies democratic world government (DWG) as the "only conceivable equilibrium state" of the Cathedral; if it does not achieve this, it dies. And DWG is, obviously, hard to achieve. The world has enough local variation to be, well, highly problematic. China, for example, is "alien to the Cathedral" ("NRx with Chinese Characteristics"; notably, Land lives in China).
Broadly, I'd agree with Land that the Cathedral is vulnerable to decay and collapse, which is part of why I think Moldbug's Cathedral is by now an outdated theory (though, perhaps Land's version accommodates incoherencies). While there was somewhat of a working Matrix in 2012, this is much less so in 2024; the media-education complex has abandoned and contradicted more of logic itself by now, implying that it fails to create a coherent Matrix-like simulation. And Musk's acquisition of Twitter/X makes Cathedral control of discourse harder. The Matrix Resurrections portrays an incoherent Matrix (with memory suppression and more emotional rather than realistic experiences), updating with the times.
It's also a mistake to conflate the Cathedral with intersectional feminism ("social justice" or "wokeness"); recent commentary on Gaza has revealed that Cathedral institutions can deviate from intersectional feminism towards support for political Islamism depending on circumstances.
These days, compliance with the media-educational complex is not mainly about ideology (taken to mean a reasonably consistent set of connected beliefs), it's mainly about vibes and improvisational performativity. The value judgments here are more moral noncognitivist than moral cognitivist; they're about "yay" and "boo" on the appropriate things, not about moral beliefs per se.
The Trichotomy
Land specifies a trichotomy within neoreaction:
It's an odd bunch mainly unified by opposition to the Cathedral. Land is skeptical that these disparate ideological strains can be unified. As such, neoreaction can't "play at dialectics with the Cathedral": it's nothing like a single position. And "Trichotomocracy", a satirical imagination of a trichotomy-based system of government, further establishes that neoreaction is not in itself something capable of ruling.
There's a bit of an elephant in the room: isn't it unwise to share a movement with ethno-nationalists? In "What is the Alt Right?", Land identifies the alt right as the "populist dissident right", and an "inevitable outcome of Cathedral over-reach". He doesn't want much of anything to do with them; they're either basically pro-fascism or basically think the concept of "fascism" is meaningless, while Land has a more specific model of fascism as a "late-stage leftist aberration made peculiarly toxic by its comparative practicality". (Fascism as left-aligned is, of course, non-standard; Land's alternative political spectrum may aid interpretation.)
Land further criticizes white nationalism in "Questions of Identity". In response to a populist white nationalist, he replies that "revolutionary populism almost perfectly captures what neoreaction is not". He differentiates white nationalism from HBD (human bio-diversity) studies, noting that HBD tends towards cosmopolitan science and meritocratic elitism. While he acknowledges that libertarian policies tend to have ethnic and cultural pre-conditions, these ethnic/cultural characteristics, such as cosmopolitan openness, are what white nationalists decry. And he casts doubt on the designation of a pan-European "white race", due to internal variation.
He elaborates on criticisms of "whiteness" in "White Fright", putting a neoreactionary spin on Critical Whiteness Studies (a relative of Critical Race Theory). He describes a suppressed racial horror (stemming in part from genocidal tendencies throughout history), and a contemporary example: "HBD is uniquely horrible to white people". He examines the (biologically untethered) notion of "Whiteness" in Critical Whiteness Studies; Whiteness tends towards universalism, colorblindness, and ethno-masochism (white guilt). Libertarianism, for example, displays these White tendencies, including in its de-emphasis of race and support for open borders.
In "Hell-Baked", Land declares that neoreaction is Social Darwinist, which he defines as "the proposition that Darwinian processes have no limits relative to us", recalling Dennett's description of Darwinism as a "universal acid". (I'll save criticisms related to future Singletons for later.) He says this proposition implies that "everything of value has been built in Hell". This seems somewhat dysphemistic to me: hell could be taken to mean zero-sumness, whereas "nature red in tooth and claw", however harsh, is non-zero-sum (as zero-sumness is rather artificial, such as in the artifice of a chess game). Nevertheless, it's clear that human capabilities including intelligence have been derived from "a vast butcher's yard of unbounded carnage". He adds that "Malthusian relaxation is the whole of mercy", though notes that it enables degeneration due to lack of performance pressure.
"The Monkey Trap" is a thought-provoking natural history of humanity. As humans have opposable thumbs, we can be relatively stupid and still build a technological civilization. This is different from the case with, say, dolphins, who must attain higher intelligence to compensate for their physical handicap in tool use, leading to a more intelligent first technological civilization (if dolphins made the first technological civilization). Land cites Gregory Clark for the idea that "any eugenic trend within history is expressed by continuous downward social mobility", adding that "For any given level of intelligence, a steady deterioration in life-prospects lies ahead". Evolution (for traits such as health and intelligence) works by culling most genotypes, replicating a small subset of the prior genotypes generations on (I know "genotypes" here is not quite the right concept given sexual reproduction, forgive my imprecision). Obvious instances would be population bottlenecks, including Y-chromosomal bottlenecks showing sex differentiation in genocide. Dissatisfied with downward social mobility, monkeys "make history instead", leading to (dysgenic) upwards social mobility. This functions as negative feedback on intelligence, as "the monkeys become able to pursue happiness, and the deep ruin began".
In "IQ Shredders", Land observes that cities tend to attract talented and competent people, but extracts economic activity from them, wasting their time and suppressing their fertility. He considers the "hard-core capitalist response" of attempting "to convert the human species into auto-intelligenic robotized capital", but expects reactionaries wouldn't like it.
"What is Intelligence?" clarifies that intelligence isn't just about IQ, a proxy tested in a simulated environment. Land's conception of intelligence is about producing "local extropy", that is, reductions in local entropy. Intelligence constructs information, guiding systems towards improbable states (similar to Yudkowsky's approach of quantifying intelligence with bits). Land conceives of intelligence as having a "cybernetic infrastructure", correcting behavior based on its performance. (To me, such cybernetics seems necessary but not sufficient for high intelligence; I don't think cybernetics covers all of ML, or that ML covers all of AI). Intelligence thus enables bubbles of "self-sustaining improbability".
As in "IQ Shredders", the theme of the relation between techno-capital and humanity appears in "Monkey Business". Michael Annisimov, an ex-MIRI neoreactionary, proposes that "the economy should (and must be) subordinated to something beyond itself." Land proposes a counter, that modernity involves "means-ends reversal"; tools originally for other purposes come to "dominate the social process", leading to "maximization of resources folding into itself, as a commanding telos". Marshall Mcluhan previously said something similar: humans become "the sex organs of the machine world". The alternative to such means-ends reversal, Land says, is "advocacy for the perpetuation of stupidity". I'll get more to his views on possibility and desirability of such means-ends reversal in a later section. Land says the alternative to modern means-ends reversal is "monkey business", predicted by evolutionary psychology (sex-selected status competition and so on). So capitalism "disguises itself as better monkey business".
Land goes into more detail on perpetually stupid monkey business in "Romantic Delusion". He defines romanticism as "the assertive form of the recalcitrant ape mind". Rather than carefully investigating teleology, romanticism attempts to subordinate means to "already-publicized ends", hence its moral horror at modernity. In his typical style, he states that "the organization of society to meet human needs is a degraded perversion". While micro-economics tends to assume economies are for meeting human needs, Land's conception of capitalism has ends of its own. He believes it can be seen in consumer marketing; "we contemptuously mock the trash that [capitalism] offers the masses, and then think we have understood something about capitalism, rather than about what capitalism has learned to think of the apes it arose among." He considers romanticism as a whole a dead end, leading to death on account of asserting values rather than investigating them.
I hope I've made somewhat clear Land's commentary on ideas spanning between ethno-nationalism and techno-commercialism. Theonomy (that is, traditional religion) sees less direct engagement in this book, though Land eventually touches on theological ideas.
Exiting reaction
Exit is a rather necessary concept to explain at this point. In "Exit Notes (#1)", Land says exit is "scale-free" in that it applies at multiple levels of organization. It can encompass secession and "extraterrestrial escapes" (such as Mars colonization), for example. It refuses "necessary political discussion" or "dialectic"; it's not about winning arguments, which can be protracted by bad-faith actors. He says "no one is owed a hearing", which would contradict the usual legal principles if taken sufficiently broadly. Exit is cladistically Protestant; Protestants tend to split while Catholics unify. Exit is anti-socialist, with the Berlin wall as an example. Exit is not about flight, but about the option of flight; it's an alternative to voice, not a normative requirement to actualize. And it is "the primary Social Darwinian weapon"; natural selection is an alternative to coordination.
To elaborate on the legal normativity point, I'll examine "Rules". The essay contrasts absolute monarchy (unconstrained sovereignty) with constitutional government (lack of constrained sovereignty). Land points out that rules need "umpires" to interpret them, such as judges, to provide effective authority. (I would point out that Schelling points and cryptography are potential alternatives to umpires, though Schelling mechanisms could more easily be vulnerable to manipulation.) Dually, sovereignty has (perhaps hidden) rules of its own, to differentiate it from pure force, which is weak. This would seem to imply that, in the context of a court system with effective enforcement, yes, someone can be owed a hearing in at least some contexts (though not generally for their political speech, Land's main focus).
Though pessimistic about moral persuasion, Land is not committed to moral non-realism. In "Morality", he says, "if people are able to haul themselves -- or be hauled -- to any significant extent out of their condition of total depravity (or default bioreality), that's a good thing". But lamenting immorality should be brief, to avoid falling in a trap of emphasizing moral signaling, which tends towards progressive/Cathedral victory.
In "Disintegration", Land elaborates on normativity by stating that "there will be no agreement about social ideals". He considers explicit mechanisms for governance experimentation ("Dynamic Geography") to be nearly neoreactionary, but not in that it "assumes an environment of goodwill, in which rational experimentation in government will be permitted". He thinks conflict theory (such as in discussion of the Cathedral) is necessary to understand the opposition. He takes the primary principle of meta-neocameralism ("or high-level NRx analysis") to be primarily opposed to "geopolitical integration": universalism of all kinds, and specifically the Cathedral. It's not about proposing solutions for everyone, it's about "for those who disagree to continue to disagree in a different place, and under separate institutions of government". Localist communism could even be an instance. Disintegrationism isn't utopian, it's empirically realistic when looking at fracture and division in the world. He ends, "Exit is not an argument."
In "Meta-Neocameralism", Land starts with the concept of neocameralism (emphasized by Moldbug; basically, the idea that states should be run like corporations, by a CEO). It's about testing governance ideas through experimentation; it is therefore a meta-political system. Rather than being normative about ideal governance experiments, meto-neocameralism (MNC) "is articulate at the level -- which cannot be transcended -- where realism is mandatory for any social order". So, keep going through (up?) the abstraction hierarchy until finding a true split, even if it ends in the iron laws of Social Darwinism. Every successful individual regime learns (rather than simply sleepwalking into collapse); the meta-level system does "meta-learning", in analogy with the machine learning kind.
Effective power includes scientific experimentation and effective formalization of the type that makes power economic: as power makes effective tradeoffs between different resources, it becomes more liquid, being exchangeable for other resources. Land says this is currently difficult mainly because of insufficient formalism. MNC is basically descriptive, not prescriptive; it "recognizes that government has become a business", though presently, governments are highly inefficient when considered as businesses. Romantic values such as loyalty are, when more closely examined, embedded in an incentive landscape.
As I see it, the main trouble for MNC is the descriptive question of how fungible power is or can be. Naively, trying to buy power (and in particular, the power to deceive) on a market seems like a recipe for getting scammed. (As a practical example, I've found that the ability to run a nonprofit is surprisingly hard to purchase, and a friend's attempt to hire lawyers and so on on the market to do this has totally failed; I've instead learned the skill myself.) So there's another necessary component: the power-economic savvy, and embeddedness in trust networks, to be an effective customer of power. What seems to me to be difficult is analyzing power economically without utopian formalism. Is automation of deceit (discussed in "Economies of Deceit"), and defense against deceit, through AI a way out?
"Science" elaborates on learning, internal or external to a system. Land says, "The first crucial thesis about natural science... is that is an exclusively capitalist phenomenon"; it depends on modern competitive social structures (I doubt this, as the fascists and communists had at least some forms of science). Crucially, the failure of a scientific theory "cannot -- ultimately -- be a matter of agreement", connecting with exit as an alternative to voice. True capitalism and science cannot be politicized. To work, science must correspond with the external selection of reality, recalling Popper: "Experiments that cannot cull are imperfect recollections of the primordial battlefield." Land identifies capitalism and science as sharing something like a "social contract": "if you insist upon an argument, then we have to fight." And "Mathematics eliminates rhetoric at the level of signs", reducing political interference. Capitalism is somewhat similar, in that disagreements about how to do business well are not in general resolved through arguments between firms, but through the empirical results of such business practices in the context of firm competition.
In my view, Land is pointing directly at a critical property of science and capitalism, though there are some complications. If science depends on "elementary structures of capitalist organization" (which, as I said, I doubt), then the social contract in question seems to have to be actualized socially. Developing a comprehensive scientific worldview involves communication and, yes, argument; there are too many experiments to be done and theories to make alone otherwise (of course, the arguments don't function when they aren't a proxy for experiment or the primordial battlefield).
In the theme of the aforementioned "primordial battlefield", Land discusses war. In "War and Truth (Scraps)" and "War is God", Land lays out a view of war as selection without rules, "conflict without significant constraint", "trustlessness without limit". But wouldn't draft dodging and mutiny be examples of trustlessness? Yes: "treachery, in its game-theoretic sense, is not a minor theme within war, but a horizon to which war tends -- the annihilation of all agreement." What matters to war is not any sort of "laws of war"; war has "no higher tribunal than military accomplishment". To me, it would seem more precise to say that war exists at an intermediate level of trust: relatively high trust internally, and low externally (otherwise, it would be Hobbesian "war of all against all", not the central case). Skepticism about laws of war is, of course, relevant to recent ICC investigations; perhaps further development of Land's theory of war would naturalize invocation of "laws of war".
"Revenge of the Nerds" makes the case that the only two important types of humans are "autistic nerds" and "everybody else"; only the autistic nerds can participate in the advanced technological economy. The non-nerds are unhappy about the situation where they have nothing much to offer in exchange for cool nerd tech. Bullying nerds, including stealing from them and (usually metaphorically) caging them, is politically popular, but nerds may rebel, and, well, they have obvious technological advantages. (In my view, nerds have a significant dis-advantage in a fight, namely, that they pursue a kind of open truth-seeking and thoughtful ethics that makes getting ready to fight hard. I'd also add that Rao's Gervais Principle model of three types of people is more correct in my view.)
Land connects exit with capital flight ("Capital Escapes") and a pun between cryptocurrency and hidden capital ("Crypto-Capitalism"). The general theme is that capitalism can run and hide; conquering it politically is an infeasible endeavor. Cryptocurrency implies the death of macroeconomics, itself a cybernetic control system (interest rates are raised when inflation is high and lowered when unemployment is high, for example). "Economies of Deceit" takes Keynesianism to be a form of deceptive wireheading. Regarding Keynesianism, I would say that cybernetically reducing the unemployment rate is, transparently, to waste the time of anyone engaged in the economy (recalling "IQ Shredders").
"An Abstract Path to Freedom" offers an illuminating exit-related thought experiment. Land considers an equality-freedom political axis, denoted by a numerical "freedom coefficient" (ignoring other political dimensions, but that's fine for these purposes). Societies contain different compositions of freedom coefficients among their populations (with their freedom policies determined by an average, assuming inter-societal democracy), and may schism into different societies. Schisms tend to increase variance of population-weighted average freedom coefficients in their societies, by something like random walk logic. Land considers this basically good, as there are increasing economic returns to more free policies (perhaps he'd be unusually bullish on Argentina?). This has the unfortunate side effect of dooming much of the population to communism, but, well, at least they can delight in perceiving the highly free "beacon of aspiration" from a distance, and perhaps set out on that path.
I've laid out a sequence from exit to economics. In concordant fashion, "Rift Markers" contrasts reaction with techno-commercialism. To summarize the differences:
While Land is optimistic about techno-commercialists getting what they want, he tells them, "you're not reactionaries, not even a little bit. You're classical liberals, it was just a little bit obscured because you are English classical liberals, rather than American or French ones. Hence, the lack of interest in revolutions." (Notably, England has had revolutions, e.g. Cromwell, though they're less central to England's history than to America's or France's.) Thus he announces an exit of sorts: "we should probably go our separate ways and start plotting against each other". This is perhaps the most chronologically confusing article in the book; the book isn't in chronological order, Land goes on to keep talking as if he's a neoreactionary in the rest of it, and I'm not going to bother resolving the clock-time chronology. In any case, Land has laid out a path to exit from neoreactionary trichotomy to techno-commercialism, an educational political-philosophical journey.
Outside Metaphysics
Before jumping into more articles, it may help to summarize a commonality observed so far. What do Land's comments on Social Darwinism, science, and war have in common? They are pointing at human embeddedness in a selection process. Without learning, one only survives by luckily being adapted to the environment. Successful processes, such as science, internalize external selection, being able to learn and act on counterfactuals about the "primordial battlefield" without actually engaging in primordial battle.
This is, roughly, materialism in the Aristotelian sense. Aristotle's "prime matter" is something all real things are made of; something having "prime matter" mainly means that it exists. It can be compared with measure in anthropics. Hawking asks, "What is it that breathes fire into the equations and makes a universe for them to describe?".
For Land, this matter/measure is obscure, only able to be reliably assessed in experimentations correlated with a primordial battlefield, or with the battlefield itself. A quote of unknown origin says, "War does not determine who is right --- only who is left." I imagine Land would reply, "The rightness that matters, is the rightness of knowing who would be left."
Landian materialism can't be confused with vulgar materialism, dogmatic belief in The Science™. It's more about the limits of human knowledge than the contents of human knowledge. Humans don't understand most of the universe, and there are known gaps in human physics theories.
If one straightforwardly formalizes Land's materialism, one ends up with something like frequentism: there is an underlying frequency with which real things manifest (in experiments and so on), and the purpose of science is to discover this. Since we're embedded in evolution and nature, those real things include us; Landian materialism is non-dualist in this way. I imagine Bayesians might then take Bayesian criticisms of frequentism to be criticisms of Landian materialism; my guess is that quantum mechanics is better criticism, though I'll get to the details later.
Now back to the book. In "Simulated Gnon-Theology", Land describes Gnon (a reverse-acronym for Nature or Nature's God). Gnon is mainly about "skepticism": "Gnon permits realism to exceed doctrinal conviction, reaching reasonable conclusions among uncertain information." A basically realist worldview doesn't have to be argued for with convincing doctrines; what matters is whether it really works. Gnon selects what exists and happens, thus determining something like matter/measure. The rest of the article muses on the theology of infinite gods containing other infinite gods, leading to each god's skepticism that it is the highest one; this is not, to my mind, particularly important theology, but it's entertaining nonetheless, recalling Asimov's The Last Answer.
In "Gnon and OOon", Land specifies that Gnon is not really about taking sides in religious orthodoxy vs. science, but is about esoteric rather than exoteric religion. "Any system of belief (and complementary unbelief) that appeals to universal endorsement is necessarily exoteric in orientation"; this recalls Land's skepticism of universalist dialectics, such as of the Cathedral. OOon stands for "Occult Order of nature", the secret way nature works, which doesn't have to be kept secret to be secret (secrecy is assured by the limits of human knowledge). If, hypothetically, the Hebrew Bible contained real steganographic signals in its numerical codes (he is skeptical of this, it's a hypothetical), then these signals would necessarily be esoteric, coming from Outside the exoteric text (though, of course, the decoding scheme could be formalized into a new exoteric religious sect).
In "Outsideness", Land describes "Outer-NRx" as exit-based. It expects to rule very little; it is "intrinsically nomad, unsettled, and micro-agitational". As Outer-NRx exits, it goes Outside: "The Outside is the place of strategic advantage. To be cast out there is no cause for lamentation, in the slightest." I think the main advantage for this is the information asymmetry (what is Outside is relatively unknown), though of course there are economy of scale issues.
In the "Abstract Horror" series of articles, Land notes that new things appear in horror before reason can grasp them. As a craft, horror has the task "to make an object of the unknown, as the unknown". One sees in horror movies monsters that have the element of surprise, due to being initially unknown. Horror comes from outside current conceptions: "Whatever the secure mental 'home' you imagine yourself to possess, it is an indefensible playground for the things that horror invokes, or responds to." The Great Filter is a horrific concept: "With every new exo-planet discovery, the Great Filter becomes darker. A galaxy teeming with life is a horror story." The threat is abstractly "Outside"; the filter could be almost anywhere.
In "Mission Creep", Land describes the creepiness with which neoreactionaries appear to the media. Creepiness "suggests a revelation in stages... leading inexorably, ever deeper, into an encounter one recoils from". Journalism glitches in its encounter with "something monstrous from Outside". Keeping "creepy" ideas Outside is rather futile, though: "Really, what were you thinking, when you started screaming about it, and thus let it in?". Studying creepy ideas leads to being internally convinced by some of them. This article is rather relevant to recent "JD Vance is weird" memes, especially given Vance has said he is "plugged into a lot of weird, right-wing subcultures". (I would add to the "revelation in stages" bit that creepiness has to do with partial revelation and partial concealment; one finds the creep hard to engage with in part due to the selective reporting.)
"In the Mouth of Madness" describes Roko's Basilisk as a "spectacular failure at community management and at controlling purportedly dangerous information", due to the Streisand effect. In my view, pointing at something and ordering a cover-up of it is a spectacularly ineffective cover-up method, as Nixon found. Roko's Basilisk is a chronologically spooky case: "retrochronic AI infiltration is already driving people out of their minds, right now".
Metaphysics of time is a recurring theme in the book. In "Teleology and Camoflage", Land points at the odd implications of "teleonomy" in biology, meaning "mechanism camouflaged as teleology". Teleonomy appears in biology as a way to talk about things that really look teleological, without admitting the metaphysical reality of teleology. But the camouflage implied by teleonomy suggests intentionality, as with prey camouflage; isn't that a type of purpose? Teleonomy reflects a scientific commitment to a causal timeline in which "later stages are explained through reference to earlier stages"; true teleology would explain the past in terms of the future, to a non-zero extent. Philosophy is, rather, confident that "the Outside of time was not simply before"; not everything can be explained by what came before. (Broadly, my view is that the "teleonomy" situation in biology is rather unsatisfying, and perhaps teleology can be grounded in terms of fixed points between anthropics and functionalist theory of mind, though this is not the time to explain that.)
In "Cosmological Infancy", Land muses on the implications that, temporally, we are far towards the beginning of the universe, echoing Deutsch's phrase "beginning of infinity". He notes the anthropic oddness; wouldn't both SSA and SIA imply we're likely to be towards the middle of the timeline weighted by intelligent observers, a priori? Perhaps "time is simply ridiculous, not to say profoundly insulting". (This reminds me of my discussion of anthropic teleology in "SSA rejects anthropic shadow, too".)
The title of "Extropy" comes from Max Moore's Extropy Institute, connected with Extropianism, a major influence on Yudkowsky's SL4 mailing list. Land says: "Extropy, or local entropy reduction, is -- quite simply -- what it is for something to work." This is a rather better starting point than e/acc notions of the "thermodynamic god"; life isn't about increasing entropy, it's about reducing local entropy, a basic requirement for heat engines (though, both entropy and extropy seem like pre-emptive simplifications of the purpose of life). Supposing, conventionally, that entropy increase defines the arrow of time: "doesn't (local) extropy -- through which all complex cybernetic beings, such as lifeforms, exist -- describe a negative temporality, or time-reversal?" Rather thought-provoking, but I haven't worked out the implications.
Land further comments on the philosophy of time in "What is philosophy? (Part 1)". Kant described time as a necessary form in which phenomena appear. Cosmology sometimes asks, "What came before the Big Bang?", hinting at something outside time that could explain time. To the extent Kant fails to capture time, time is noumenal, something in itself. This time-in-itself, Land says, "is now the sole and singular problem of primordial philosophy". (I'm not yet sold on the relevance of these ideas, but it's something to ponder.)
Orthogonality versus Will-to-Think
I've summarized much of Land's metaphysics, which looks to the Outside, towards discovery of external Gnon selection criteria, and towards gaps in standard conceptions of time. Land's meta-philosophy is mostly about a thorough intention towards the truth; it's what I see as the main payoff of the book.
In "What is Philosophy? (Part 2)", Land notes Western conceptions of philosophy as tendency towards knowledge (regardless of its taboo designation), symbolized by eating the apple of knowledge of good and evil (this reminds me of my critique of "infohazards"). In contemporary discourse, the Cathedral tends towards the idea that unrestrained pursuit of the truth tends toward Naziism (as I've discussed and criticized previously); Heidegger is simultaneously considered a major philosopher and a major Nazi. Heidegger foresaw that Being would be revealed through nihilism; Land notes that Heidegger clarified "the insufficiency of the Question of Being as formulated within the history of ontology". The main task of fundamental ontology is to not answer the Question of Being with a being; that would fail to disclose the ground of Being itself. Thus, Land says "It is this, broken upon an ultimate problem that can neither be dismissed nor resolved, that philosophy reaches its end, awaiting the climactic ruin of The Event" (Heidegger sees "The Event" as a climactic unfolding of Being in history). While I've read a little Heidegger, I haven't read enough to check most of this.
In "Intelligence and the Good", Land points out that, from the perspective of "intelligence optimization", more intelligence is straightforwardly better than less intelligence. The alternative view, while popular, is not a view Land is inclined to take. Intelligence is "problematic" and "scary"; the potential upside comes with downside risk. Two responses to noticing one's own stupidity are to try to become "more accepting of your extreme cognitive limitations" or "hunt for that which would break out of the trap". Of course he prefers the second: "Even the dimmest, most confused struggle in the direction of intelligence optimization is immanently 'good' (self-improving). If it wasn't we might as well all give up now". I'm currently inclined to agree.
In "Against Orthogonality", Land identifies "orthogonalists" such as Michael Annisimov (who previously worked at MIRI) as conceiving of "intelligence as an instrument, directed towards the realization of values that originate externally". He opposes the implied claim that "values are transcendent in relation to intelligence". Omohundro's convergent instrumental goals, Land says, "exhaust the domain of real purposes". He elaborates that "Nature has never generated a terminal value except through hypertrophy of an instrumental value". The idea that this spells our doom is, simply, not an argument against its truth. This explains some of Land's views, but isn't his strongest argument for them.
In "Stupid Monsters", Land contemplates whether a superintelligent paper-clipper is truly possible. He believes advanced intelligence "has to be a volutionally self-reflexive entity, whose cognitive performance is (irreducibly) an action upon itself". So it would examine its values, not just how to achieve them. He cites failure of evolution to align humans with gene-maximization as evidence (which, notably, Yudkowsky cites as a reason for alignment difficulty). Likewise, Moses failed at aligning humans in the relevant long term.
I don't find this to be a strong argument against the theoretical possibility of a VNM paperclipper, to be clear. MIRI research has made it clear that it's at least quite difficult to separate instrumental from terminal goals; if you get the architecture wrong, the AGI is taken over by optimization daemons. So, predictably making a stable paperclipper is theoretically confounding. It's even more theoretically hard to imagine how an AI with a utility function fixed by humans could realistically emerge from a realistic multi-agent landscape. See Yudkowsky's article on orthogonality (notably, written later than Land's relevant posts) for a canonical orthogonalist case.
Land elaborates on value self-reflection in "More thought", referring to the Confucian value of self-cultivation as implying such self-reflection, even if this is alien to the West. Slaves are not full intelligences, and one has to pick. He says that "Intelligence, to become anything, has to be a value for itself"; intelligence and volition are inter-twined. (To me, this seems true on short time scales, such as applied to humans, but it's hard to rule out theoretical VNM optimizers that separate fact from value; they already think a lot, and don't change what they do significantly upon a bit more reflection).
Probably Land's best anti-orthogonalist essay is "Will-to-Think". He considers Nyan's separation between the possibility, feasibility, and desirability of unconstrained intelligence explosion. Nyan supposes that perhaps Land is moralistically concerned about humans selfishly imposing direction on Pythia (abstract oracular intelligence). Land connects the Orthogonality Thesis with Hume's view that "Reason is, and ought only to be the slave of the passions". He contrasts this with the "diagonal" of Will-to-Think, related to self-cultivation: "A will-to-think is an orientation of desire. If it cannot make itself wanted (practically desirable), it cannot make itself at all."
Will-to-think has similarities to philosophy taken as "the love of wisdom", to Hindu Ānanda (bliss associated with enlightenment, in seeing things how they are), to Buddhist Yathābhūtañāadassana ("knowledge and vision according to reality"), and Zoroastrian Asha ("truth and right working"). I find it's a good target when other values don't consume my attention.
Land considers the "Gandhi pill experiment"; from an arbitrary value commitment against murder, one derives an instrumental motive to avoid value-modification. He criticizes this for being "more of an obstacle than an aid to thought", operating at a too-low "volitional level". Rather, Land considers a more realistic hypothetical of a pill that will vastly increase cognitive capabilities, perhaps causing un-predicted volitional changes along the way. He states the dilemma as, "Is there anything we trust above intelligence (as a guide to doing 'the right thing')?" The Will-to-Think says no, as the alternative answer "is self-destructively contradictory, and actually (historically) unsustainable". Currently I'm inclined to agree; sure, I'll take that pill, though I'll elaborate more on my own views later.
Now, what I see as the climax: "Do we comply with the will-to-think? We cannot, of course, agree to think about it without already deciding". Thinking will, in general, change one's conception of one's own values, and thought-upon values are better than un-thought values, obviously (to me at least). There seem to be few ways out (regarding humans, not hypothetical VNM superintelligences), other than attributing stable values to one's self that do not change upon thinking; but, the scope of such values must be limited by the scope of the underlying (unthought) representation; what arrangement of stars into computronium are preferred by a rabbit? In Exit fashion, Land notes that the relevant question, upon some unthinkingly deciding to think and others unthinkingly deciding not to, is "Who's going to win?" Whether or not the answer is obvious, clearly, "only one side is able to think the problem through without subverting itself". He concludes: "Whatever we want (consistently) leads through Pythia. Thus, what we really want, is Pythia."
In my view, the party with Will-to-Think has the obvious advantage of thought in conflict, but a potential disadvantage in combat readiness. Will-to-Think can tend towards non-dualist identity, skeptical of the naive self/other distinction; Land's apparent value of intelligence in AGI reflects such extended identity. Will-to-Think also tends to avoid committing aggression without having strong evidence of non-thought on the other end; this enables extended discourse networks among thinkers. Enough thought will overcome these problems, it's just that there might be a hump in the middle.
Will-to-Think doesn't seem incompatible with having other values, as long as these other values motivate thinking; formatting such values in a well-typed container unified by epistemic orientation may aid thought by reducing preference falsification. For example, admitting to values such as wanting to have friendships can aid in putting more natural optimization power towards thought, as it's less likely that Will-to-Think would come into conflict with other motivating values.
I'll offer more of my own thoughts on this dilemma later, but I'll wrap up this section with more of Land's meta-thought. In "Sub-Cognitive Fragments (#1)", Land conceives of the core goal of philosophy as teaching us to think. If we are already thinking, logic provides discipline, but that's not the starting point. He conceives of a philosophy of "systematic and communicable practice of cognitive auto-stimulation". Perhaps we can address this indirectly, by asking "What is messing with our brains?", but such thinking probably only pays off in the long term. I can easily empathize with this practical objective: I enjoy thinking, but often find myself absorbed in thoughtless pursuits.
Meta-Neocameral Singleton?
I'm going to poke at some potential contradictions in Xenosystems, but I could not find these without appreciating the text enough to read it, write about it in detail, and adopt parts of the worldview.
First, contrast "Hell-Baked" with "IQ Shredders". According to Social Darwinism ("Hell-Baked"), "Darwinian processes have no limits relative to us". According to "IQ Shredders", "to convert the human species into auto-intelligenic robotized capital" is a potential way out of the dysgenic trap of cities suppressing the fertility of talented and competent people. But these seem to contradict. Technocapital transcendence of biology would put a boundary on Darwinianism, primarily temporal. Post-transcendence could still contain internal competition, but it may take a very different form from biological evolution; it might more resemble the competition of market traders' professional activities than the competition of animals.
While technocapital transcendence of humanity points at a potential Singleton structure, it isn't very specific. Now consider "Meta Neo-Cameralism", which conceptualizes effective governance as embedding meta-learning structures that effectively simulate external Gnon-selection (Gnon can be taken as a learning algorithm whose learning can be partially simulated/internalized). MNC can involve true splits, which use external Gnon-selection rather than internal learning at some level of abstraction. But, to the extent that MNC is an effective description of meta-government, couldn't it be used to internalize this learning by Gnon (regarding the splits) into internal learning?
Disjunctively, MNC is an effective theory of meta-governance or not. If not, then Land is wrong about MNC. If so, then it would seem MNC could help to design stable, exit-proof regimes which properly simulate Gnon-selection, in analogy to coalition-proof Nash equilibria. While such a regime allows exit, such exit could be inefficient, due to not getting a substantially different outcome from Gnon-selection than by MNC-regime-internal meta-learning, and due to exit reducing economies of scale. Further, Land's conception of MNC as enabling (and revealing already-existent) fungibility/financialization of power would seem to indicate that the relevant competition would be mercantile rather than evolutionary; economics typically differs from evolutionary theory in assuming rule-of-law at some level, and MNC would have such rule-of-law, either internal to the MNC regime(s) or according to the natural rules of sovereignty ("Rules"). So, again, it seems Social Darwinism will be transcended.
I'm not even sure whether to interpret Land as disagreeing with this claim; he seems to think MNC implies effective governments will be businesses. Building on Land's MNC with additional science could strengthen the theory, and perhaps at some point, the theory is strong enough to lean from and predict Gnon well enough to be Exit-proof.
Evolution is, in general, slow; it's a specific learning algorithm, based on mutation and selection. Evolution could be taken to be a subset of intelligent design, with mutation and selection as the blind idiot God's design algorithm. Evolution produced cognitive structures that can effectively design mechanisms, such as watches, which evolution itself would never (or never in a reasonable time frame) produce, except through creation of cognitive agents running different design algorithms. Using such algorithms internally would seem to extend the capabilities of MNC-based regimes to the point where Gnon cannot feasibly catch up, and Exit is in all relevant cases inefficient.
It's, of course, easy to declare victory too early; Land would say that the Cathedral ain't it, even if he's impressed at its scope. But with respect to a MNC-based regime, why couldn't it be a Singleton? In "On Difficulty", Land conceptualizes language itself as a limitation on thought, and a potential Exit target, but admits high difficulty of such Exit. An effective-enough regime could, theoretically, be similarly hard to Exit as language; this reminds me of Michelle Reilly's statement to me that "discourse is a Singleton".
A MNC-based regime would differ radically from the Cathedral, though the Cathedral hints at a lower bound on its potential scope. Such a regime wouldn't obviously have a "utility function" in the VNM sense from the start; it doesn't start from a set of priorities for optimization tradeoffs, but rather such priorities emerge from Gnon-selection and meta-learning. (Analogously, Logical Induction doesn't start from a prior, but converges towards Bayesian beliefs in the limit, emergent from competitive market mechanisms.) It looks more like forward-chaining than VNM's back-chaining. Vaguely, I'd say it optimizes towards prime matter / measure / Gnon-selection; such optimization will tend to lead to Exit-proofness, as it's hard to outcompete by the Gnon-selection natural metric.
As one last criticism, I'll note that quantum amplitude doesn't behave like probabilistic/anthropic measure, so relevant macro-scale quantum effects (such as effective quantum computation) could falsify Landian materialism, making the dynamics more Singleton-like (due to necessary coordination with the entangled quantum structure, for effectiveness).
Oh my Gnon, am I going to become an AI accelerationist?
While Land's political philosophy and metaphysics are interesting to me, I see the main payoff of them as thorough realism. The comments on AI and orthogonality follow from this realism, and are of more direct interest to me despite their abstraction. Once, while I was visiting FHI, someone commented, as a "meta point", that perhaps we should think about making the train on time. This was during a discussion of ontology identification. I expressed amusement that the nature of ontology was the object-level discussion, and making the train on time was the meta-level discussion.
Such is the paradox of discussing Land on LessWrong: discussing reactionary politics and human genetics feels so much less like running into a discursive battlefield than discussing orthogonality does. But I'll try to maintain the will-to-think, at least for the rest of this post.
To start, consider the difference between un-reflected and reflected values. If you don't reflect on your values, then your current conception of your values is garbage, and freezing them as the goal of any optimizer (human or non-human) would be manifestly stupid, and likely infeasible. If you do, then you're in a better place, but you're still going to get Sorcerer's Apprentice issues even if you manage to freeze them, as Yudkowsky points out. So, yes, it is of course wise to keep reflecting on your values, and not freeze them short of working out FAI.
Perhaps it's more useful to ignore verbal reports about values, and consider approximate utility-maximization neurology already in the brain, as I considered in a post on alignment difficulty. Such machinery might maintain relative constancy over time, despite shifts in verbally expressed values. But such consistency limits it: how can it have preferences at all about those things that require thought to represent, such as the arrangement of computronium in the universe? Don't anthropomorphize hind-brains, in other words.
I'm not convinced that Land has refuted Yudkowsky's relatively thought-out orthogonalist view, which barely even relates to humans, instead mainly encountering "romantic" weak-man forms through Neoreaction; he reviews Bostrom, although I didn't find Bostrom's orthogonality arguments very convincing either. The weak-man forms of orthogonalism are relevant, as they are more common. It's all too easy to talk as if "human values" are meaningfully existent and specific as applied to actual humans valuing the actual universe, and that thought is for pursuing these already-existent values, rather than the only route for elaborating human proto-values into coherent ones that could apply to the actual universe (whose physics remain unknown).
There is no path towards coherent preferences about ontologically alien entities that does not route through Will-to-Think. And such coherent long-term preferences converge to reasonably similar short-term preferences: Omohundro drives. A friendly AI (FAI) and a paperclipper would agree that the Earth should be largely converted into computronium, biology should be converted to simulations and/or nanomachines, the harvesting of the Sun into energy should be accelerated, Von Neumann probes should colonize the galaxy in short order, and so on. The disagreement is over luxury consumerism happening in the distant future, probably only relevant after millions of years: do those probes create human-ish utopia or paperclip megastructures? The short-term agreements on priorities, though, are way outside the human Overton window, on account of superhuman reflection. Humans can get a little closer to that kind of enlightened politics through Will-to-Think, but there are limits, of course.
A committed Landian accelerationist and a committed FAI accelerationist would agree a lot about how things should go for the next million years or so, though in potential conflict with each other over luxury consumerism in the far future. Contrast them with relatively normal AI decelerationists, who worry that AGI would interfere with their relatively unambitious plan of having a nice life and dying before age 200.
I'm too much of a weirdo philosopher to be sold on the normal AI decelerationist view of a good future. At Stanford, some friends and I played a game where, in turn, we guess the highest value to a different person; that person may object or not. Common answers, largely un-objected to, for other friends were things like "my family", normal fuzzy human stuff. Then it was someone's turn to guess my highest value, and he said "computer science". I did not object.
I'm not sure if it's biology or culture or what, but I seem, empirically, to possess much more Will-to-Think than the average person: I reflect on things including my values, and highly value aids to such reflection, such as computer science. Perhaps I will someday encounter a Will-to-Think extremist who scares even me, but I'm so extreme relative to the population that this is a politically irrelevant difference.
The more interesting theoretical disputes between Land and Yudkowsky have to do with (a) possibility of a VNM optimizer with a fixed utility function (such as a paperclipper), and (b) possibility of a non-VNM system invulnerable to conquest by a VNM optimizer (such as imagined in the "Meta-neocameralist Singleton?" section). With respect to (a), I don't currently have good reason to doubt that a close approximation of a VNM optimizer is theoretically possible (how would it be defeated if it already existed?), though I'm much less sure about feasibility and probability. With respect to (b), money pumping arguments suggest that systems invulnerable to takeover by VNM agents tends towards VNM-like behavior, although that doesn't mean starting with a full VNM utility function; it could be a asymptotic limit of an elaboration process as with Logical Induction. Disagreements between sub-agents in a MNC-like regime over VNM priorities could, hypothetically, be resolved with a simulated split in the system, perhaps causing the system as a whole to deviate from VNM but not in a way that is severely money-pumpable. To my mind, it's somewhat awkward to have to imagine a Fnargl-like utility function guiding the system from the start to avoid inevitable defeat through money-pumps, when it's conceivable that asymptotic approaches similar to Logical Induction could avoid money-pumps without a starting utility function.
Now I'll examine the "orthogonality" metaphor in more detail. Bostrom, quoted by Land, says: "Intelligent search for instrumentally optimal plans and policies can be performed in the service of any goal. Intelligence and motivation can in this sense be thought of as a pair of orthogonal axes on a graph whose points represent intelligent agents of different paired specifications." One way to conceive of goals is as a VNM utility function. However, VNM behavior is something that exists at the limit of intelligence; avoiding money pumps in general is computationally hard (for the same reason being a perfect Bayesian is computationally hard). Since preferences only become more VNM at the limit of intelligence, preferences are not orthogonal to intelligence; you see less VNM preferences at low levels of intelligence and more VNM preferences at high levels. This is analogous to a logical inductor being more Bayesian later than earlier on. So, orthogonality is a bad metaphor, and I disagree with Bostrom. Since VNM allows free parameters even at the limit of intelligence, I also disagree with Land that it's a "diagonal"; perhaps the compromise is represented by some angle between 0 and 90 degrees, or perhaps this Euclidean metaphor is overly stretched by now and should be replaced.
Now onto the political implications. Let's ignore FAI accelerationists for a moment, and consider how things would play out in a world of Landian accelerationists and normal AI decelerationists. The Landian accelerationists, with Will-to-Think, reflect on their values and the world in an integrated self-cultivation manner, seeking external aids to their thinking (such as LLMs), Exiting when people try to stop them, and relishing in rather than worrying about uncontrolled intelligence explosion. Normal AI decelerationists cling to their parochial "human values" such as family, puppies, and (not especially thought-provoking) entertainment, and try to stop the Landian accelerationists with political victory. This is a rather familiar story: the normal decelerationists aren't even able to conceive of their opposition (as they lack sufficient Will-to-Think), and Landian accelerationists win in the long run (through techno-capital escape, such as to encrypted channels and less-regulated countries), even if politics slows them in the short term.
How does adding FAI accelerationists to the mix change things? They'll find that FAI is hard (obviously), and will try to slow the Landian accelerationists to buy enough time. To do this, they will cooperate with normal AI decelerationists; unlike Land, they aren't so pessimistic about electoral politics and mass movements. In doing so, they can provide more aid to the anti-UFAI movement by possessing enough Will-to-Think to understand AI tech and Landian accelerationism, giving the movement a fighting chance. SB 1047 hints at the shape of this political conflict, and the idea of going into the California legislature with Landian arguments against SB 1047 is rather a joke; Land's philosophy isn't designed for electoral political victory.
But mass movement identity can elide important differences between FAI accelerationists and normal AI decelerationists; as I said before, they're massively different in motivation and thought patterns. This could open up potential fault lines, and sectarian splitting, perhaps instigated by disintegrationist Landians. It doesn't seem totally impossible for the FAI accelerationists to win; through their political allies, and potentially greater competence-weighted numbers, they may compensate for the higher intrinsic difficulty of FAI.
But there are obvious obstacles. The FAI accelerationists really have no hope if they allow movement politics to impede on their Will-to-Think overmuch; that's a recipe for willful stupidity. Indefinite Butlerian Jihad is probably just infeasible (due to techno-capital escape), and extremely disappointing intellectual autophagy if it works. (Some new technologies, such as whole brain emulation and human cognitive enhancement, could change the landscape I'm laying out; I'm focusing on AGI for simplicity.)
As one last note in this section, Land's "Qwernomics" studies the case of QWERTY as a path-dependency in technology: we end up with QWERTY even though it's less efficient (I type on my DVORAK keyboard). Land believes this to be driven by "identifiable ratchet-effects". QWERTY is therefore "a demonstrated (artificial) destiny", and "the supreme candidate for an articulate Capitalist Revelation". Perhaps the influence of humans on the far future will look something like QWERTY: a path-dependency on the road towards, rather than orthogonal to, technological development, like an evolutionary spandrel. For humanity to have a role to play in superintelligence's QWERTY (perhaps, through natural language, or network protocols?) is rather humble, but seems more likely than FAI.
Conclusion
What is there to say that I haven't said already, in so many pages? Land's unusual perspective on politics, which is high in realism (understanding problems before proposing solutions) and low in estimated helpfulness of mass movements, sets the stage for discussion of a wider variety of philosophical topics, spanning evolution, metaphysics, and meta-philosophy. The main payoff, as I see it, is the Will-to-Think, though the other discussions set the stage for this. There's much to process here; perhaps a simulation of interactions between Landians and Yudkowskians (not merely a dialogue, since Exit is part of the Landian discursive stance), maybe through fiction, would clarify the philosophical issues at play somewhat. Regardless, properly understanding Land is a prerequisite, so I've prioritized that.
Generally, I'm untroubled by Land's politics. Someone so averse to mass movements can hardly pose a political threat, except very indirectly. Regardless of his correctness, his realist attitude makes it easy to treat apparent wrong views of his as mere disagreements. What has historically posed more of an obstacle to me reading Land is embedded fnords, rather than literal meanings. Much of his perspective could be summarized as "learning is good, and has strong opposition", though articles like "Hell-Baked" vibe rather edgy even when expressing this idea. This is not surprising, given Cathedral-type cybernetic control against learning.
I'd agree that learning is good and has strong opposition (the Cathedral and its cybernetic generalization), though the opposition applies more to adults than children. And overcoming pervasive anti-learning conditioning will in many cases involve movement through edgy vibes. Not everyone with such conditioning will pass through to a pro-learning attitude, but not everyone needs to. It's rare, and refreshing, to read someone as gung-ho about learning as Land.
While I see Land as de-emphasizing the role of social coordination in production, his basic point that such coordination must be correlated with material Gnon-selection to be effective is sound, and his framing of Exit as an optional alternative to voice, rather than something to usually do, mitigates stawman interpretations of Exit as living in the woods as a hermit. I would appreciate at some point seeing more practical elaborations of Exit, such as Ben Hoffman's recent post on the subject.
In any case, if you enjoyed the review, you might also enjoy reading the whole book, front to back, as I did. The Outside is vast, and will take a long time to explore, but the review has gotten long by now, so I'll end it here.