Explaining black box AI machine learning models and using interpretable models

Choisissez la langue de votre document :

Parliamentary questions

48k

10k

1 October 2020

E-005390/2020

Question for written answer E-005390/2020
to the Commission
Rule 138
Patrick Breyer (Verts/ALE)

Answer in writing

Subject: Explaining black box AI machine learning models and using interpretable models

There has been criticism(1) of the fact that the private sector has an incentive to produce non-transparent black box algorithms in order to be able to commercialise them. Simple interpretable models with the same or a higher level of performance would be free for everybody to use.

1. What is the Commission’s opinion of the proposal(2) that for certain high-stakes decisions, black boxes should not be allowed if there is already an interpretable model with the same level of performance?

2. What is the Commission’s opinion of the proposal(3) that organisations that introduce black box models should also be obliged to report the accuracy of interpretable modelling methods?

(1)	Rudin, Cynthia, ‘Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead’, Nature Machine Intelligence 1, 206–215 (2019). https://doi.org/10.1038/s42256-019-0048-x
(2)	See above.
(3)	See above.

Last updated: 16 October 2020

Legal notice - Privacy policy