Talk: Just er­rors? To­wards re­flect­ive use of com­puters in so­cial science

Social scientists are increasingly applying machine learning approaches to analyse data. However, machine learning process require making procedural choices. Researchers have shown that these processes are sensitive to made choices. In the worst case, scholars may accept or reject an hypothesis due to procedural choices, not due to a phenomena existing in the data.

Nelimarkka´s talk opens up this can of worms and highlights how our scholarly practices require more methodological research but also increasing reflexivity on our research practices. What can machine learning methods learn from the decades of discussion and development which has supported other social science research methods?