{"id":2427,"date":"2025-11-10T08:46:47","date_gmt":"2025-11-10T07:46:47","guid":{"rendered":"https:\/\/www.lindacastaneda.com\/?p=2427"},"modified":"2025-11-11T10:08:13","modified_gmt":"2025-11-11T09:08:13","slug":"cuando-la-ia-desnaturaliza-la-ciencia","status":"publish","type":"post","link":"https:\/\/www.lindacastaneda.com\/en\/mushware\/cuando-la-ia-desnaturaliza-la-ciencia\/","title":{"rendered":"When AI \u201cDenaturalising\u201d Science"},"content":{"rendered":"

<\/p>\n

Across multiple disciplines \u2014genetics, neuroscience, social sciences, computational fields\u2014 a similar warning has emerged: if AI is used only to get things right<\/em> rather than to understand why<\/em>, we risk producing statistics that look<\/em> like science but do not explain it. Science does not end at the thresholds that statistics reveal; its purpose is to explore, to understand, to propose mechanisms, to infer causes, to generate refutable hypotheses, and to design interventions that work beyond<\/em> the dataset.<\/p>\n

This debate has history.\u00a0 Just to mention 2 of the authors mentioned it, Breiman<\/strong> spoke of the two cultures \u2014prediction versus inference. At the same time (and this is crucial), it is not enough to look at the algorithm; we must also examine the infrastructure<\/em> that makes it credible;\u00a0 the team of Williamson et al. (2024)<\/strong> show, for instance, how consortia, data architectures and technical apparatuses establish a data-centred epistemology that reframes educational phenomena as molecular-like associations \u201cdiscoverable\u201d through bioinformatics. That sociotechnical choreography grants authority to the algorithmic, displaces social theory, and produces an ontology in which subjects appear entirely surveyable and predictable. But this does not happen only in the social realm; it extends to how we understand the natural world \u2014indeed, to almost any complex sociomaterial context.<\/p>\n

How to integrate AI without \u201cdenaturalising\u201d science<\/strong><\/h3>\n
\"\"
Analogy: The fried egg is the classic example of protein denaturalisation.<\/figcaption><\/figure>\n
    \n
  1. Declare AI\u2019s role in one sentence.<\/strong>
    \u201cGenerates hypotheses.\u201d \u201cActs as a surrogate model to speed up simulations.\u201d \u201cPrioritises experiments.\u201d
    If you write \u201cdiscovers cause,\u201d make sure you can defend a causal design \u2014not just performance.<\/li>\n
  2. Position your work on the scientific staircase.<\/strong>
    Description \u2192 prediction \u2192 mechanism \u2192 intervention\/contrafactual.
    Specify where you are and what is missing to move upward (experiments, instruments, DAGs\/causal diagrams, control variables).<\/li>\n
  3. Triangulate with theory.<\/strong>
    Let your pattern converse with existing frameworks: does it confirm, contradict, or extend them?
    If it contradicts, state what must be revised and how you will test it out of distribution<\/em> (a different cohort, site, or team).<\/li>\n
  4. Design explanations that support decisions.<\/strong>
    Feature importance alone is not enough. What plausible mechanism does it suggest? What experiment or quasi-experiment would you run tomorrow to try to falsify it?<\/li>\n
  5. Use hybrids when appropriate.<\/strong>
    Physics- or theory-informed models, biological or organisational constraints embedded in architectures, or AI \u2192 hypothesis \u2192 experiment pipelines.<\/li>\n<\/ol>\n

    Less \u201coracle\u201d, more cumulative science.<\/p>\n

    Warning signs of denaturalisation<\/strong><\/h3>\n