On Monday, 11th March 2019 Klaus-Robert Müller, Professor for Machine Learning at the TU Berlin emphasized within the Enquete Commission Artificial Intelligence (AI) of the German Bundestag for a stronger emphasis on the field of “explainable AI”. Particularly, he pointed to the enormous risks of common AI/ML systems in safety critical areas, e.g. medical diagnostics.

Therefore, they emphasis in research should be in the field of explainable artificial intelligence” to enable domain experts to understand why a certain machine decision has been reached. The Müller-Group published recently an interesting Nature paper:

Sebastian Lapuschkin, Stephan Wäldchen, Alexander Binder, Grégoire Montavon, Wojciech Samek & Klaus-Robert Müller (2019). Unmasking Clever Hans predictors and assessing what machines really learn. Nature Communications, 10, (1), doi:10.1038/s41467-019-08987-4, https://www.nature.com/articles/s41467-019-08987-4

The authors criticize that despite that ML has solved hard problems and reaching high accuracy, they show seemingly intelligent behavior. The authors observe that standard performance evaluation metrics can be oblivious to distinguishing diverse problem solving behaviors. Moreover, they propose their semi-automated Spectral Relevance Analysis that provides a practically effective way of characterizing and validating the behavior of nonlinear learning machines. This helps to assess whether a learned model indeed delivers reliably for the problem that it was conceived for. Most important this paper emphasizes the caution to the enormous excitement on machine “intelligence” and pledges to evaluate and judge some of these recent successes in a more nuanced manner.