When do algorithms violate the law
I enjoyed reading the judgement of the Federal Court of Australia on whether Westpac Banking Corporation’s computer operated home loan approval system (known as the automated decision system or ADS) violated Australia’s responsible lending laws. The judgement is fun to read, and that might itself be enough reason to read it since delightful court judgements are relatively rate. More importantly, this issue of evaluating algorithms for compliance is going to become increasingly important in the years to come.
The court threw out the case basically on the ground that the Australian Securities and Investments Commission (ASIC) had not done its homework well enough.
[ASIC] does not allege that the alleged defects in the ADS resulted in Westpac extending loans to any consumers who it ought to have found would be unable to meet their financial obligations under the credit contracts or who would be able to do so only in circumstances of substantial hardship. ASIC did originally make several such allegations in relation to specified loans but it abandoned these on the day before the trial commenced. This then is a case about the operation of the responsible lending laws without any allegation of irresponsible lending.
ASIC was claiming that Westpac’s ADS violated the responsible lending laws simply because the rules in the algorithm ignored some data or used imperfect measures for some variables. The court rejected this approach:
It is not enough to point to an individual rule in the ADS and to submit that it does not comply with Div 3. Westpac’s entire system (including manual assessment where referral is triggered) must be examined, and compliance with Div 3 gauged that way.
Of course, the Court is right on this point, and therein lies the challenge in regulating a financial world that is increasingly run on algorithms. It appears to me that financial sector regulators are by and large unprepared for this challenge.
Posted at 7:55 pm IST on Wed, 28 Aug 2019 permanent link
Comments