Australian AI ethics principles: NAB strengthens its data ethics

NAB’s AI solution

To help customers, NAB piloted facial recognition technology (FRT) allowing them to digitally verify their identity.

The FRT technology allowed customers to take a picture of their identification (ID) document (for example a drivers licence) on their mobile. They could then provide images or videos of themselves. The FRT would then compare the image on their ID document with the one they supplied to verify their identity.

NAB worked with an external supplier that uses FRT software to devise this solution.

Considering ethics when purchasing AI products

Before NAB decides to acquire and use AI products developed elsewhere it:

• investigates potential harm to its customers of implementing this technology
• checks the technology doesn’t unjustly impact its customers
• requires providers to design their data system so others can audit it.  

NAB has a process for reviewing all data analytic and AI projects from an ethical viewpoint before implementation. These reviews check that the technology is responsible, sustainable and explainable to affected customers.

The review focuses on:

• addressing inherent bias
• promoting transparency
• implementing explainable AI and deep learning techniques
• ensuring customers have given their consent to the intended application
• ensuring staff have completed relevant privacy impact and data ethics assessments
• identifying ways customers can challenge use of their data or AI-enabled decisions where there is a significant impact.

These ethical reviews ultimately build trust with customers that the bank uses their data ethically and securely.

Putting AI ethics principles into practice

NAB participated in the Australian AI Ethics Principles pilot to enhance its existing data ethics processes. This participation also allowed the bank to incorporate the Australian AI Ethics Principles into its existing framework. It focused on the subset of principles that it wanted to strengthen in its existing data ethics processes.

NAB applied its own data ethics principles alongside the Australian AI Ethics Principles when considering the FRT solution.

Principle 1: Human, social and environmental wellbeing

NAB identified all the potential impacts of the FRT solution on:

• its customers
• society
• the environment
• its business.

NAB considered customers would significantly benefit from the technology, especially those restricted from travelling outside due to COVID-19. They could easily and conveniently use their mobile to verify their identity.

The FRT solution allowed NAB to balance its obligations to conduct diligent ID checks while delivering time-saving benefits to customers. With increased use, this technology could contribute to reduced carbon emissions with fewer in-person visits to the bank.

Principle 2: Human-centred values

NAB checked that the FRT solution was based on human-centred values that represented all of NAB’s customers. The bank required suppliers to provide assurances on whether:

• the facial recognition data the solution was trained on was representative
• the supplier adequately tested all demographic groups and could supply results for consideration
• there was evidence showing the solution did not impact customers unfairly
• there was evidence showing that there were no disproportionately higher error rates for certain groups over others.

Principles 4, 6, 7 and 8: Privacy protection and security, Transparency and explainability, Accountability and Contestability

NAB considered the FRT solution a fairly low-risk use of AI. This is because the solution analyses data sets to compare images with documents customers provide.

Despite being a low-risk application, NAB intentionally chose not to completely automate the solution‘s deployment. It configured the FRT solution to refer all verification requests the system couldn’t verify  to a human operator (a banker) rather than rejecting them.

To ensure the system was delivering accurate results and beneficial outcomes, NAB:

• monitored all customer verification requests when they piloted the FRT
• called customers who needed assistance to complete the verification process
• reviewed all verification outcomes to ensure customers had a positive experience.

NAB implemented processes to review and audit the FRT solution outputs as needed. NAB sought assurances from its AI supplier that the solution was auditable (fully or partially). This inquiry helped to ensure NAB didn’t use technology it could not adequately explain internally or to customers and regulators.

Where NAB acquires AI solutions, its ethical assessment processes help identify how the solution makes decisions. NAB expects AI vendors to provide transparent disclosure on the inner workings of proposed solutions (subject to confidentiality as required). Where this disclosure is not available or of poor quality, NAB has chosen to go with another supplier or develop the technology in-house.

Under contracts, NAB could also restrict suppliers from using captured images for any purposes other than the original purpose for which the image was supplied. This would further protect privacy by ensuring personal information is not used for ‘secondary purposes’ as required by Australia’s privacy laws.

Often the pure machine power behind AI can lead to personal information being used in ways it was not originally intended. To protect privacy, NAB only uses personal information for purposes the customer has endorsed. In this situation, NAB endorsed the use of the FRT solution to verify the customer’s identity with their consent. If NAB were to use the FRT for any other purpose, they would need to seek the customer’s consent for this secondary purpose.

Benefits and impacts

Following the pilot, NAB enhanced its data and AI ethics frameworks by:

• merging its data and AI ethics assessment process
• boosting resources to raise internal awareness (for example on the benefits of managing ethical impacts for NAB and its customers)
• facilitating multi-disclipinary forums across the bank with subject matter experts, tech developers and data scientists to discuss ethical risks (diverse views can surface impacts across various scenarios)
• developing mechanisms to measure and monitor use cases after they go live.

NAB had a positive experience discussing ethics with its FRT vendor. The vendor was very accommodating in addressing NAB’s queries about ethical impacts and the transparency of the FRT solution. This has not always been the case with other vendors. NAB shared the following observation from interactions with other AI vendors:

‘The ethical AI questions we asked appeared to be a first for some vendors who were unable to answer them without deeper preparation.’

Vendors commented that NAB was asking more questions than other potential clients. NAB’s response to this is:

‘We see this as a good thing. But also see this as an opportunity for organisations at large to all come together to seek higher standards. If other large organisations sought these conversations, AI developers will become more prepared to explain how they consider ethics in the systems they develop.’

NAB commits to leveraging AI ethically to positively benefit its customers, employees and its organisation. This commitment drives NAB to engage with third parties that share similar perspectives.

Contact NAB

If you have any questions about this example or are interested to learn about NAB’s data and AI ethical processes, please email data.ethics@nab.com.au