The recent proliferation of ‘Generative AI (GenAI)’ has raised a significant question for both artists and scholars: can Artificial Intelligence generate legitimate art? AI models in their current form, which are various types of Machine Learning (ML) algorithms, are essentially computational models that discern a mathematical pattern from a given dataset and apply it to new inputs. Then, the underlying question here is whether ML algorithms, as computational models, can simulate some aspect of artistic imagination —textual, visual, or otherwise. I attempt to answer this question through Augustine’s theology of numbers and imagination.
In De Musica, Augustine presents a theology of ‘numbers’ (numeri), which names the principle of harmony and coherence inherent in beautiful things within the created order. He analyses the beauty of poetic rhythm and meter in terms of their ‘numberliness’ (numerositas), but this discussion can be naturally extended to other forms of art. It is by these numbers that beautiful forms please us, as the soul perceives, remembers, enacts, judges, and evaluates their ‘equality’ and ‘order’ —their truthfulness and beauty (VI.xiii.38). Imagination is the power of the memory, which by (re)ordering these numbers according to the proper context of time and order, recalls and produces mental images of beauty (VI.viii.22). So far, there are significant analogies between this theory of imagination and the operation of the ML algorithm. The latter also ‘learns’ the numbers undergirding sensible forms and reproduces them by reordering the numbers within its ‘memory.’ In this sense, these algorithms can simulate the human imagination to a certain degree.
However, for Augustine, the problem is that our sensibilities and perspectives are always limited; and as fallen creatures with finite and impure desires, we are too often not free to imagine what is truly beautiful. Only a transcendent and infinite love can free the finite imagination from its inherent drive towards falsehood. Thus, the order of one’s loves determines the truthfulness of the phantasms produced by the imagination (VI.xi.32). A sound imagination belongs to the soul in which its numbers are ordered towards the love of God and their neighbours (VI.xi.29). By practicing such imagination, one is able to express the true beauty that belongs to God and invite others to contemplative wisdom (sapientia) —like Ambrose’s hymn Deus Creator Omnium (VI.xvii.57). On the other hand, a disordered memory, which directs its desires towards the transient and the superficial rather than the eternal, produces vain phantasms and errs by mistaking them for reality (VI.xiii.42). Pride is the characteristic vice of such a disordered soul, who uses its phantasms to manipulate others for its own end instead of serving others (VI.xiii.41). This is where the AI fails: it lacks the fundamental freedom to desire, delight, or love. Its simulated imagination is always confined not only by its fundamental architecture but also by the desires of its designers and users —hence, its bias. Consequently, Generative AI cannot avoid generating false phantasms, and incapable of distinguishing them from reality, the algorithm ‘hallucinates.’
I conclude by suggesting some practical and ethical implications of this analysis. First, I contend that to regard Generative AI as a free and independent subject of artistic imagination is not only theologically untruthful but also dangerous, socially and spiritually. Then, I propose that the ethical design and use of AI must always involve a serious reflection of our own loves; since it is we human beings who love, and the AI can only reflect and amplify our own desires —whether infinitely oriented towards God and neighbours or sinfully bound by pride.
The recent proliferation of ‘Generative AI’ has raised a significant question: can AI generate legitimate art? I attempt to answer this question through Augustine’s theory of numbers and imagination. In his De Musica, numbers name the fundamental principle of beauty, and imagination is the faculty that (re)orders the numbers in one's memory to produce images of beauty. There are significant analogies between this theory of imagination and the operation of the ML algorithm here. However, for Augustine, the problem of our limited sensibilities and perspectives is critical. Our imagination is bound by our desires, and only transcendent love can free one's imagination. I argue that the simulated imagination of the AI is always bound by its limited architecture and the finite desires of its designers and users. Thus, it cannot but generate false phantasms and 'hallucinate.' I conclude by suggesting some practical and ethical implications of this analysis.