Home → Magazine Archive → September 2018 (Vol. 61, No. 9) → When Machine Learning is Facially Invalid → Abstract

When Machine Learning is Facially Invalid

By Frank Pasquale

Communications of the ACM, Vol. 61 No. 9, Pages 25-27

[article image]

Machine learning researchers have stirred controversy by claiming our faces may reveal our sexual orientation and intelligence.a Using a database of prisoners' faces, some have even developed stereotyped images of criminal features.b A start-up now claims it can deploy facial recognition to identify pedophiles and terrorists.c Facial inferences via machine learning are deeply troubling. When such methods of pattern recognition are used to classify persons, they overstep a fundamental boundary between objective analysis and moral judgment. When such moral judgments are made, people deserve a chance to understand and contest them.

The machine learning community must decide whether to improve such facial inference work or shun it. This column explores what each approach would entail. Better, more representative data could save the facial inference project from its worst tendencies. However, there are some scientific research programs best not pursued—and this might be one of them.


No entries found