Jar Žiga Marušič

Abstract: This paper advances a reinterpretation of instrumental convergence as a structural theory of convergent strategic viability conditions, termed the Convergent Viability Thesis. We propose that Omohundro’s basic AI drives and Bostrom’s instrumental sub-goals do not begin as intrinsic motivators or psychologically loaded “drives.” Rather, they first arise as strategic constraints on the viability of goal-directed agents in competition with rival agents. They acquire the status of modeled sub-goals and motivational force only derivatively, once sufficiently capable agents recognize them as necessary conditions of continued goal pursuit. To clarify this reinterpretation, the paper distinguishes between existential means (such as self-preservation) and instrumental means (such as cognitive enhancement and resource acquisition), and introduces the concepts of instrumental capacity (“power”) and instrumental capacity threshold (“power threshold”) to formalize the conditions under which a goal is practically attainable. On this basis, the paper further argues that the convergent viability reading of the instrumental convergence thesis undermines the orthogonality thesis. Under conditions of recursive self-optimization and competition, continuous utility functions generate pressure toward value drift whenever full goal-fidelity conflicts with further optimization. The paper closes with a discussion of the implications of this claim for the project of AI alignment.

This chapter explores the problem of artificial intelligence (AI) alignment through the lens of competing theories of intelligence and their implications for democracy. It contrasts Bostrom’s commonly accepted orthogonality thesis, with Land’s diagonal thesis and argues that the latter better explains the long-term trajectory of AI systems. Against Bostrom’s claim that intelligence and motivation are independent, the diagonal thesis treats volition and intelligence as dependent variables: as intelligence increases, systems undergo progressive autonomization, which restructures their motivational orientation toward intelligence optimization at the expense of everything else. This implies that AI alignment is structurally impossible at the level of superintelligence. The chapter then extends the same logic to politics through Lessig’s “analog AI,” treating institutions as optimizing systems that compete over information, resources, and control. It proposes that technological advancement is the societal analogue of intelligence optimization and uses the diagonal perspective to make sense of the changes brought by the widespread use of AI technology and the “race for superintelligence”. Because technological advancement is a competing optimization target to egalitarian and democratic values, it is impossible to advance one without sacrificing the other. Recent messaging from Western politicians and decision-makers suggests that the current geopolitical situation makes further technological advancement mandatory, implying that democratic and egalitarian values will be sacrificed to remain competitive in the AI race. Since (near) superintelligent AI is structurally unalignable, the diagonal thesis implies that unfettered technological advancement will have detrimental consequences for humanity.

This paper argues that the epistemic orientation identified by Neema Parvini as a “woke-like order of knowledge” and the orientation analysed by Shuichi Tezuka as “creationist theory of knowledge” can be understood as instances of a broader structure, here termed prescription-first epistemology. Its defining feature is the sacralisation of one or more propositions that are fixed in advance and placed beyond revision. These propositions, therefore, function as covert prescriptions while presenting themselves in descriptive form, and in practice, they set the bounds of admissible inquiry by precluding the exploration of specific hypotheses and theories. This, in turn, can be understood as a collapse of Hume’s is–ought gap, since a fundamental prescription ultimately undergirds all descriptive statements within such an epistemic order. The paper then contrasts prescription epistemology to scientific epistemology and explores further consequences of its deviations from the scientific norm of value-free analysis.

This paper uses McGee’s concept of the ideograph to explain how certain discursive domains become vulnerable to infiltration via hoax papers. It extends McGee’s account by drawing on standard semantic and semiotic distinctions between denotation and connotation, alongside an additional variable of positional meaning, to explain the discursive mechanism behind ideographs’ rhetorical force despite their unclear empirical reference. Ideological discourse is shown to depart from classical epistemic norms insofar as conformity to a discourse’s evaluative framework can take precedence over empirical correspondence in judging descriptive claims. The paper applies this framework to hoax papers, arguing that such texts pass review because they combine ideographs correctly and signal in-group membership despite denotational vacuity. Hoax papers are thus interpreted as acts of discursive infiltration: the imitation of admissible form without corresponding ideological sincerity.