Could AI undermine our (belief in the truth of) scientific theories?

The article discusses the philosophical implications of artificial intelligence (AI), especially its impact on scientific realist arguments about the predictive success of scientific theories.
The author emphasizes the significance of novel predictions in supporting the truth of scientific theories, using the formulation of the periodic table by Dmitrii Mendeleev as an illustrative example.
The role of AI in predicting properties of substances is investigated, particularly its capacity to predict novel substances without explicit theoretical training, which challenges traditional arguments for the truth of scientific theories.
A philosophical dilemma arises: if AI makes novel predictions without learning theories, either the value of predictive success in supporting theory truth was overestimated or the truth of traditional scientific theories is less certain.
The opacity problem concerns the lack of understanding of how AI models make predictions, acting as a black box that can obscure the underlying theoretical connections and processes.
The author acknowledges the evolving nature of AI and the potential of explainable AI to resolve some of the current opacity issues, but also suggests that the understanding of AI might change rapidly, keeping philosophical inquiry dynamic.
Finally, the article concludes with a note of caution and curiosity, highlighting the need for ongoing philosophical exploration as AI progresses.