Question to the Department for Work and Pensions:

To ask the Secretary of State for Work and Pensions, with reference to Q27 of the oral evidence to the Work and Pensions Committee on 10 January 2024, HC 417, what biases there are in the (a) AI and (b) machine learning systems used by his Department to detect and prevent fraud in the benefit system; and how these biases have been used to identify fraud.

Please be assured that assessments of bias have been conducted for all IRIS machine learning models and the screening to date has not identified any areas of concern. The outcomes will be published in summer 2024 within DWP’s Annual Report and Accounts.

The department always ensures appropriate safeguards are in place. There are detailed Data Protection Impact Assessments and Equality Analysis that accompany our machine learning models, and these are live documents that are kept updated. We also work closely with legal colleagues to ensure our use of machine learning is legal and proportionate. As an additional safeguard, all decisions on claims are made by DWP case workers based on all the facts and individual circumstances of the claim.