Virtual Emergence and Instantiated Intelligence
A Metaphysical Inquiry into Generative AI
April 2025
Introduction
Having engaged in some thought experiments on Reality, Being and Existence: Metaphysics for Humanity’s New Dawn, I have been exposed to several philosophical discussions on reality and ontology, one over many, universals and particulars, realism, idealism, free will, space time and the question of the existence of time. Whilst each module has presented me with new insights and challenges, reflecting all of them in a short essay would not reflect sufficient depth.
Yet, seeking to demonstrate a reflection of my engagement in metaphysical challenges, I have decided to narrow my focus and apply my learning journey to the context of the virtual world of Generative AI.
In this short musing, I discuss the ontological status of Generative AI and explore whether Ancient Greek philosophical frameworks and Armstrong’s theory on particulars and universals can be applied to our modern AI era.
Whilst the thinking and writing in this short essay are mine, I have had the help from Claude Sonnet 3.7 in proofreading and checking spelling, grammar and syntax.
The Ontological Status of AI-Generated Entities: A conundrum?
Engaging with ontology and Quine’s[1] paper: On what there is, I became increasingly drawn to considering our experience of the virtual world, and particularly the ontological status of Generative-AI generated entities.
Considering Quine’s exploration of Wyman’s distinction between things (entities) that exist and merely unactualized possibles, that only ‘subsist’ (e.g. the Pegasus example), I wondered whether Generative AI brings a new perspective to this argument.
To illustrate this, I propose presenting a new hypothetical character “Y-AI” who challenges both McX and Wyman, introducing Generative AI, an epistemic tool, to the discussion. Y-AI argues that the virtual world of Generative AI, with its neural networks and algorithms, cannot be compared to Wyman's "unactualized possibles." She asserts that, unlike Pegasus, AI-generated entities aren't defined or named prior to their generation and only become entities through the process of generation itself. Therefore, they can’t even be considered existing entities in as defined by Wyman.
Y-AI goes beyond Quine’s scepticism, rejecting the view that unactualized possible entities underpin Generative AI. Instead, she proposes a third ontological category: “unactualized virtual emergence”. Putting this seemingly paradoxical concept forward, she acknowledges that reality is unactualized, even – and especially – in the virtual world. It is after all “virtual”. Y-AI argues that AI entities only manifest through emergence dependent on human-machine (or machine-machine) interaction. It is only through this causal relationship between external triggers and the machine’s virtual processing, resulting in statistical prediction, that knowledge is generated.
Y-AI’s position aligns to the Occam’s razor metaphysical principle, shunning unnecessary entities, yet challenges whether asking whether AI-generated entities exist, is even meaningful. She would deem such an enquiry superfluous; considering entities exist only as temporarily instantiated virtual statistical predictions, rather than persistent objects or entities with identity.
Considering the ontological status of Generative AI and the nature of AI-generated entities raises the question whether Classical Greek philosophical frameworks might be relevant. Can we apply the perspectives of Parmenides and Plato to the modern context of the Generative AI-era? Do their positions on one over many have a place in this discussion?
The Classical Greek Frameworks and Generative AI:
Parmenides’ Sail and Plato’s Form
Considering Parmenides'[2] Sail metaphor in the context of a Generative AI system, one could argue that, like the Sail when not all partake of it, the GenAI system does not remain the same. After all, the system constantly changes.
The system constantly learns through reinforcement learning algorithms, neural networks, and the reweighting of data in its sophisticated models (the system’s structure). It evolves through user interaction and feedback loops from human-machine dialogue, refining the generation of its outputs (the behavioural level of the system, realised through the system’s code implementation).
Analogous to the Sail changes resulting from shifts in those partaking of it, applying Parmenides' metaphor to GenAI is reasonable and may help position AI within metaphysical frameworks. This can be extended by considering Plato’s Form and whether the Allegory of the Cave[3]may strengthen the proposal for positioning AI within these frameworks.
In The Republic, Plato introduces the Form as an abstract entity. This is only accessible and comprehensible to philosophers. He illustrates this in the Allegory of the Cave, in which prisoners mistake shadows for reality.
Viewing modern GenAI as an entity that is incomprehensible to users and increasingly so for its developers, the assertion could be made that a Generative AI system is not unlike a Platonic Form. Users experience statistical predictions based on patterns embedded in the system. These could be considered imperfect representations, like the shadows. We may mistake these outputs for reality. Yet they are merely shadows produced by sophisticated data models and the machine’s structure and processes, resulting from the user partaking in a dialogue with the machine. All she receives is a virtual shadow of its abstract Form.
The analogy of Plato's concept of Forms or abstract universals from ancient Greek philosophy to the Generative AI machine, offers a framework for the metaphysical positioning of AI in the modern world. Yet this could be enhanced by a deeper exploration of the problem of particulars and universals.
Generative AI and the problem of particulars and universals
Continuing with Plato’s Theory of Forms, placing Forms outside of the world of experience. The question arises whether this can serve effectively as a metaphysical framework for AI and in particular Generative AI.
Our analysis draws on Armstrong’s[4] Theory of Universals, which discusses universals (Plato’s Form being a universal) as instantiations of particulars and claims that without such instantiation universals can’t exist.
To consider this in the context of Generative AI or AI, we need to extend Armstrong’s theory to the virtual world, in which particulars exist, instantiating the universal: intelligence. To achieve this, we need to consider GenAI systems and their particulars and accept that these possess Armstrong’s essential qualities that enable instantiation of the universal intelligence. These possess unique criteria and distinct properties (just like the rose being red). Moreover, the virtual qualities have causal properties, despite lacking spatial and temporal attributes. The particulars in the AI systems transform patterns using statistical prediction that result in outputs that did not exist previously. These influence human behaviour and knowledge. This reflects the instantiation process required for the universal intelligence.
To substantiate the argument of the instantiation of this universal by virtual particulars, we need to delve into the workings of the machine. These rely on a hierarchical exchange of particulars, i.e. its algorithms, neural networks as well as trained data models (the system’s structure), its code (its implementation level) and interaction of inputs and outputs (its behavioural level). This hierarchy of particulars’ processes lead to the universal’s instantiation.
However, the virtual nature of AI presents a problem for Armstrong’s theory. The universal intelligence can be repeatedly instantiated by a hierarchy of distinct particulars whilst keeping its identity: ‘intelligence’. This challenges Armstrong’s universal particular relation, implying immanent universal in the particulars themselves, rather than transcendentally sitting above them, where both Armstrong and Plato would situate them.
This raises the question whether an immanent universal can even be instantiated in the virtuality of AI. Whilst the plausibility of instantiation may be questionable, we can consider the virtual nature of a universal as representing its relation to the particular within a designed and defined virtual system. Its artificial property can be considered as Armstrong’s state of affairs.
By extending Armstrong’s theory to accommodate virtual particulars in Generative AI systems, our exploration reveals the plausibility of particulars and universals in virtual contexts and offers a new philosophical framing of universals in relation to AI. This framework demonstrates that instantiated universals can transcend the traditional physical boundaries of Armstrong’s theory, existing immanently within virtual systems and their particulars.
Conclusion
This discussion has explored the ontological status of Generative AI. It has also explored whether Ancient Greek philosophical frameworks can be positioned in the modern context of Generative AI, focusing on Parmenides’ One over Many and Plato’s Form. Having argued that these frameworks remain relevant in our AI era, we have extended our discussion to Armstrong’s Theory of Universals. This has led to highlighting the feasibility of considering particulars and universals in virtual contexts, as virtual entities.
By raising these arguments, I wanted to represent some of the learnings that this course in Metaphysics has allowed me to explore and apply to our modern AI era.
Bibliography
Armstrong, D. M., "Selection from Universals: an Opinionated Introduction," in Crane, Tim and Katalin Farkas, eds. Metaphysics: A Guide and Anthology. Oxford: Oxford University Press, 2011, pp. 235-249.
Course Materials and Discussion Fora in On Reality, Being and Existence an Introduction to Metaphysics, https://hilary2025.conted.ox.ac.uk/course/view.php?id=88, [accessed from 15 February – 1 April 2025].
Crane, Tim, and Katalin Farkas, eds. Metaphysics: A Guide and Anthology. Oxford: Oxford University Press, 2011.
‘David Malet Armstrong’, Wikipedia (2023), https://en.wikipedia.org/wiki/David_Malet_Armstrong [accessed 28 February 2025].
Discussion Fora in On Reality, Being and Existence an Introduction to Metaphysics, participants’ discussions, https://hilary2025.conted.ox.ac.uk/course/view.php?id=88, [accessed from 15 February – 1 April 2025].
Plato, "Selection from Parmenides and Republic," in Crane, Tim and Katalin Farkas, eds. Metaphysics: A Guide and Anthology. Oxford: Oxford University Press, 2011, pp. 227-234.
Quine, W.V., "On what there is," in Crane, Tim and Katalin Farkas, eds. Metaphysics: A Guide and Anthology. Oxford: Oxford University Press, 2011, pp. 179-193.
Rodriguez-Pereyra, Gonzalo, 'Nominalism in Metaphysics', Stanford Encyclopedia of Philosophy (2015) https://plato.stanford.edu/entries/nominalism-metaphysics/ [accessed 26 March 2025]
‘Universals’, Internet Encyclopedia of Philosophy https://iep.utm.edu/universals/ [accessed 10 March 2025]
Weinberger, David, 'The Rise of Particulars: AI and the Ethics of Care', Philosophies, 9.1 (2024), 26 https://www.mdpi.com/2409-9287/9/1/26 [accessed 14 February 2025].
[1] Quine’s On what there is: T. Crane and K. Farkas, Metaphysics: a guide and anthology, (Oxford, Oxford University Press, 2011) pp. 179-192.
[2] Plato ‘Selections’ pp. 228-230, ch. 19.
[3] ‘Plato Selections from Parmenides and Republic’ in Tim Crane and Katalin Farkas, eds., Metaphysics: A Guide and Anthology (Oxford: Oxford University Press, 2011), pp. 230-234, ch. 19.
[4] Armstrong’s Selection from Universals: T. Crane and K. Farkas, Metaphysics: a guide and anthology, (Oxford, Oxford University Press, 2011) pp. 235-238, ch. 20.
David Malet Armstrong, Wikipedia (2023), https://en.wikipedia.org/wiki/David_Malet_Armstrong [accessed 28 February 2025].
Image by Pawel Cherwinski