I came across this excellent paper on the philosophical underpinnings of Machine Learning (as a discipline). It's well worth a read.
http://arxiv.org/abs/2604.06754
The summary (from the paper) is:
Machine learning is a style of reasoning, and is as rhetorical as any other. It
• Takes data as fact (not a core object of enquiry)
• Presumes the data is “random” (as an omnibus sanitisation protocol)
• Purports to learn representations of the world (from the “intrinsic structure of data”)
• Presumes that knowing the world suffices to control it
• Takes categories as features of the world (to avoid grappling with the hard choice)
• Avoids grappling with the tension between the individual and the aggregate
• Confuses and conflates data and information
• Valorises method above all
• Judges methods solely via canned “benchmarks”
• Makes black boxes, without providing the associated data-sheets.
• Construes its products as fully autonomous, when it is mere partial delegation.
It has honed its style of reasoning so that the style is invisible. It has thus successfully turned itself into a self-perpetuating thought-style — in other words, a “discipline”.
#ML #MachineLearning #AI #ArtificialIntelligence #assumptions #philosophy #rhetoric #NewPaper