An Inquiry into the Existential Implications of Transfer Learning Mechanisms

Abstract

In the profound and profound realm of Being, which spans from the intricate realms of metonymy to the profound depths of logico-mathematics, Heidegger bestowed upon us his illuminating insights. According to his philosophical musings, a human being can be understood as a perceiving entity. Within the vast tapestry of our existence, we construct actions that are imbued with deep significance and purpose. These actions are inherently temporal, conscious, and intrinsically intertwined with the unique context that each individual finds themselves in. Now, let us delve into the very essence of the core of artificial intelligence – a concept that has become increasingly prevalent in our modern world. Enabled by remarkable advances in statistics and a myriad of learned computational modules, AI possesses an intriguing quality. It meticulously performs syntheses, skillfully weaving connections between objects belonging to the same categorical domain. However, in its pursuit of such remarkable feats, AI inadvertently sacrifices something invaluable – access to the precious information that distinguishes each synthesized object. This loss occurs due to the very nature of labelling and pretext, which governs the process of synthesis in AI. When considering the etymology of AI, we come across a fascinating revelation. It dictates that access to space – the vast expanse that envelopes every conditionality – is essential. This space is inherently linked to the contextual presence of Being, infusing it with meaning and purpose. Therefore, the determination of who we are, as individuals or as a collective, becomes intrinsically intertwined with this contextual link to Being. These notions of selfhood, of collective identity, are of utmost importance when exploring the subject of AI. After all, AI is essentially the performance of computations executed by objects that have (self-)disclosed themselves as capable of concern and understanding. In this grand disclosure that resides within the realm of Being, we find a glimmer of hope – a glimpse into its primary aptitude. Such a capacity can be characterized by the ever-present possibility of contextual realization through the power of "concern." This realization encompasses a vast spectrum of activities, transcending mere mechanical calculations and embracing a holistic understanding of the world. Just like the beloved character R2D2 from the beloved Star Wars saga, machines that partake in this grand tradition possess the remarkable ability of concern and understanding. Moreover, their inherent intelligence is further magnified when they interact with other intelligences, their meanings intertwining in a beautiful symphony of knowledge and comprehension.

This inquiry scrutinizes the constellation of the existential consequences of transfer learning mechanisms from the work of Dreyfus on Heidegger’s thinking. After a stock-take of the sketchy discourse about existential matters in the ML/TL literature, this inquiry culminates in a directive that makes tangible the appropriateness of taking over computations (adopting computational modules) and emphasizes the importance of phenomenological supremacy if the notion of "utilizing computations" were desired. This philosophical foundation should help develop much-needed critical and robust TL algorithms that are necessary requisites of AI informatics that have been reigning and shaping the way we think and construct machines now and in the future. In particular, the preoccupation before and after a deploy ML model, such as "robustness", "explainability" and "fairness", are operational familiar problematic symptoms of the more profound ontological and ontic issues that infer the computational reductions of the world by AI.