Limits of Automatization

Workshop at the University of Tübingen

Machine Learning (ML) is the study and developmentof computational methods which help to improve the performance of a machineby enabling it to acquireknowledge from experience. Learning therefore has to be mechanized. The ML approach overlaps with the inductive taskwell knownin the empirical sciences. In fact ML even aims forautomatization in the scientific processes itself, thereby reducing or evenreplacing human intellectual work to some extent in the actual practice of science. The promise of some people in the field of ML research is that automatization of the scientific process will increase the rate of discovery and the productivity of sciencein general. At least one tacit assumption seems to support this thesis, namely the idea that ML would range at the same methodological level as theory and experiment.

The workshop aims at exploring the limits of automatization in ML research in general and with respect to the scientific process. To what extent do human decisions still play a crucial role for thesuccess of ML? How much does the epistemological power of ML algorithms depend on prior assumptions built right into the algorithms? Do we need to develop a more nuanced view of the relation between human judgment, expertise and machine learning algorithms? Finally, how does this intricate entanglement affect questions of moral responsibility and the accountability of human agents using this technology? For your participation please send an email to info(at) explaining your motivation to participate. Please note that participants are expected to stay for the entire workshop and to engage in discussion.



  • Dr. Niels Weidtmann (FORUM SCIENTIARUM, University of Tübingen)
  • Michael Hermann (FORUM SCIENTIARUM, University of Tübingen)

Further Information (PDF)