An article by the Allen Institute of Artificial Intelligence highlights a pending issue: machines do not really understand what humans write or read.

Surely we all intuit or know them, but it should be remembered please. Because in the very near future, perhaps in Europe can legislate and allow decisions in the judicial, economic and social fields. An issue that has to be very seriously addressed.

And this raises an important question whether these models have truly acquired robust commonsense capabilities or whether they rely on spurious biases in the datasets that lead to an overestimation of the true capabilities of machine commonsense.

External link:

NLP Artificial Intelligence algorithms tracing Bias AIAudit Ethics eulaws