M.Z. Naser

Large language model (LLM) interactions often produce dual convergence that is invisible to users. In this paper, we identify two concurrent phenomena: 1) axiological convergence, wherein users inherit preferences embedded in model training without inspection or endorsement, and 2) epistemic convergence, wherein shared infrastructure has the potential to homogenize beliefs across populations despite apparent personalization. Both processes compound since as epistemic sources concentrate, value transmission concentrates correspondingly, and as values align, epistemic filtering reinforces that alignment. We characterize the conditions under which dual convergence emerges and argue that it requires only that users optimize for convenience and that infrastructure exhibits scale economies. Fundamentally, the appearance of personalization can co-exist with delivering individualized outputs while producing aggregate homogenization toward shared epistemic and axiological baselines. We demonstrate that no individual-level intervention can break convergence without sacrificing convenience, thereby establishing this as a collective action problem that requires a structural response. We conclude that dual convergence constitutes a novel mechanism of cultural evolution that operates through accommodation as opposed to coercion, with implications that span autonomy, pluralism, and the epistemic foundations of democratic societies.

When a large language model receives a prompt, it computes a probability distribution over all possible continuations before sampling produces a single output. This paper examines the ontological and ethical status of ungenerated completions (i.e. texts that were mathematically present in the distribution but never actualized). I argue that these ungenerated texts occupy a novel ontological category since they are neither merely possible nor fully actual. Therefore, they exist as computationally determinate potentials that are precisely weighted, semantically meaningful, and accessible in principle, although not in practice. On this basis, I advance three claims. First, standard metaphysical categories fail to capture the mode of being proper to probability distributions over meaningful strings, and hence, a new category of computational abstracta is required. Second, the structured space of ungenerated completions is metaphysically more fundamental than any particular generated completion because it carries most of the probability mass, encodes more of the model’s structure, and preserves a richness of potential responses that any single realization necessarily truncates. Third, I sketch initial implications for what I term modal ethics. Insofar as the distribution assigns nontrivial probability to completions of moral significance, our practices of prompting and sampling are subject to normative assessment. What remains ungenerated can matter morally because foreclosing or enabling particular possibilities has consequences for actual agents, and this reveals a broader class of philosophical problems that could not have been formulated before the advent of large language models.

The philosophical notion of possible worlds has long been debated as a metaphysical abstraction. Machine learning (ML) can present an empirically grounded form of this plurality when multiple models trained on identical data achieve equivalent performance while inferring incompatible internal representations. This paper develops an algorithmic lens on this phenomenon by treating such cases as evidence-compatible (epistemic) possible worlds. We demonstrate how competing ML and causal algorithms can identify mutually incompatible causal graphs, each statistically credible yet diverging in mechanistic, interventional, and counterfactual commitments. We also show how this phenomenon extends to large language models (LLMs), where hallucinations can be interpreted not only as errors but as internally coherent alternative textual world-states sampled from the same parameterized distribution, with varying alignment to our reference world. By framing models as evidence-compatible worlds, we provide an empirical grounding on underdetermination (a modernized Duhem–Quine thesis) and argue that accuracy alone cannot adjudicate between incompatible explanations. We conclude by outlining the ethical implications of privileging one algorithmic narrative over rivals that remain comparably supported by the data.

This paper examines the paradoxical decline in engagement with philosophy of science among engineers precisely when machine learning (ML) systems are increasingly performing complex epistemological functions in engineering practice. We identify how philosophical naivety, characterized by the uncritical adoption of reductive frameworks regarding consciousness, intelligence, and ethics, creates tangible organizational and technical liabilities. We then demonstrate how conceptual limitations in engineers’ philosophical foundations lead to three primary flaws: 1) ontological misclassification of system capabilities, 2) ethical blind spots in ML system design and application, and 3) inadequate epistemological approaches and hidden philosophical commitments for interpreting model outputs. Thus, we argue that renewed engagement with the philosophy of science is not merely academic but necessary for engineers to maintain epistemic authority and responsibility in an era where engineering judgment is increasingly delegated to or mediated by ML systems. In response, we propose a technical-philosophical framework integrating perspectives from philosophy of mind, ethics, epistemology, and engineering to address these shortcomings systematically.

Causality and explanation are fundamental, yet often underacknowledged, in engineering. While engineering education emphasizes design principles, codes of practice, and technical considerations, the explicit notion of causality is typically buried under layers of empirical formulas, rule-based guidelines, and prescriptive procedures. This paper examines how causality remains elemental in engineering practices and why codal provisions and curricula often treat it only implicitly. This paper also explores the interplay between causality and machine learning (ML), which excels at detecting associations, but questions still arise regarding its ability to explain and capture causal relations. In a series of case studies, we argue that building codes conventionally distill collective engineering wisdom and embed causal reasoning in simplified yet codified rules. However, this embedding is seldom articulated as “causality,” which leads to potential gaps in how engineers explain their decisions. Thus, this paper advocates for a renewed focus on causality in engineering by bridging philosophical inquiry, engineering methodology, and ML.