Editing Essentialism (section)

===Machine learning===

Pelillo argues that traditional [[machine learning]] techniques often align with an essentialist paradigm by relying on [[feature (machine learning)|features]] - properties assumed to be essential for [[classification]] tasks. For instance, [[pattern recognition]], which attempts to extract essential attributes from data, is described as inherently essentialist since it presupposes that objects have stable, identifiable essences that define their categories. This perspective extends to [[similarity learning|similarity-based]] approaches, which use [[prototype theory]] to establish relationships within data by grouping instances around central prototypes that exhibit the "essence" of a category.<ref>{{cite journal|last=Pelillo|first=M.|title=Introduction: The SIMBAD Project|journal=Similarity-Based Pattern Analysis and Recognition|editor=M. Pelillo|year=2013|volume=Advances in Computer Vision and Pattern Recognition|pages=1–10|doi=10.1007/978-1-4471-5628-4_1}}</ref>

Expanding on this, Pelillo and Scantamburlo highlight that certain machine-learning scenarios, such as when data is highly dimensional or features are poorly defined, challenge the essentialist framework. They advocate for alternative paradigms that consider relational and [[Context (linguistics)|context]]ual [[information]] instead of isolated feature analysis. This relational focus aligns with anti-essentialist stances, which view categories as dynamic and context-dependent rather than fixed.<ref>{{cite journal|last=Pelillo|first=M.|author2=Scantamburlo, T.|title=How Mature Is the Field of Machine Learning?|journal=AI*IA 2013: Advances in Artificial Intelligence|year=2013|volume=8249|pages=121–132|doi=10.1007/978-3-319-03524-6_11}}</ref>

Šekrst and Skansi build on these ideas, noting that [[supervised learning]], by utilizing labeled [[dataset]]s, reflects essentialist tendencies since it relies on predefined human-defined categories. However, they argue that this does not commit machine learning to an ontological stance on essentialism. Instead, they propose that the categories used in supervised learning are human-constructed in [[feature selection]] processes and reflect [[epistemology|epistemological]] practices rather than [[metaphysics|metaphysical]] truths. Similarly, [[unsupervised learning]]'s [[Cluster analysis|clustering]] and similarity-based approaches often resemble prototypical reasoning but do not inherently affirm or deny essentialism, focusing instead on [[pragmatics|pragmatic]] task performance.<ref>{{cite journal|last=Šekrst|first=K.|author2=Skansi, S.|title=Machine Learning and Essentialism|journal=Philosophical Problems in Science (Zagadnienia Filozoficzne w Nauce)|year=2022|volume=73|pages=171–196|url=https://philpapers.org/rec/EKRMLA|doi=}}</ref>