Research Insights by Jacqui O'Sullivan https://padlet.com/josullivan0308/fzqav7fdvs7xttfm en-us 2022-10-13 22:20:04 UTC 2026-03-01 19:42:14 UTC hello@padlet.com Research Q1 trammtran8 https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339475211 investigate barriers that constrain transparency & fairness in loan services through primary research on current data & regulations

a)Investigate the use of AI and privacy laws within the EU and US to further explore potential strategies that ANZ can incorporate within Australia’s jurisdiction.]]> 2022-10-13 22:22:51 UTC https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339475211 Research Q2 trammtran8 https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339475281 Explore consumer behaviour and attitudes towards the use of AI algorithms when collecting, storing and analysing their data to determine the importance of transparency and fairness as fundamental ethical principles.

]]> 2022-10-13 22:23:00 UTC https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339475281 Research Q3 trammtran8 https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339475484 Research and investigate case studies and literature reviews on the use of AI within the banking industry to determine any presence of discrimination in home loan frameworks. 


]]> 2022-10-13 22:23:21 UTC https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339475484 Research Q4 trammtran8 https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339475619 Review ANZ’s current privacy policies to understand the utilisation of AI in ANZ’s processes to propose policy changes aiming to embed ethical principles of transparency and fairness.

]]> 2022-10-13 22:23:37 UTC https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339475619 josullivan0308 https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339477851 lack of governance of data collection in Aus means that customers ability to 'select a product that best meets their privacy preferences' is undermined (ACCC 2018)]]> 2022-10-13 22:27:16 UTC https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339477851 josullivan0308 https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339479573 governments in Aus & internationally have proposed regulatory advances like the 'consumer data right', social media regulation, a regulatory organisation of some sort]]> 2022-10-13 22:30:30 UTC https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339479573 trammtran8 https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339479683 the focus of consumers and regulators upon opaque, excessive and unreasonable collection, use and sharing of revealing information about individuals is causing businesses to undertake expensive rearchitecting of their data systems, data handling processes, and data project governance and assurance. (Nicholls et al., 2022)

]]> 2022-10-13 22:30:41 UTC https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339479683 trammtran8 https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339480366 Not enough to simply comply with data privacy law → trust and privacy concerns go hand in hand. If consumers lack trust in businesses (due to lack of transparency) → this will constrain the promotion of fair and transparent practices of AI within the organization. Need to also think about CSR and ethics in the digital age 
(Nicholls et al., 2022)

]]> 2022-10-13 22:31:58 UTC https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339480366 https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339480763 In the GDPR citizens are granted the right to explaination for decisions made by algo and information on logic involved. - such as Credit decisions, insurance pricing.]]> 2022-10-13 22:32:40 UTC https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339480763 Exacerbate existing discrimination https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339481100 1. Compound existing biases found in the data
2. Training models with biased data
3. Identification of spurious correlations]]> 2022-10-13 22:33:18 UTC https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339481100 Potential support for solution/recommendations trammtran8 https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339481543 sustainable future in AI innovation in the financial sector is to support a proactive regulatory approach prior to any financial harm occurring. This proactive approach should implement rational regulations that embody jurisdiction-specific rules in line with carefully construed international principles (Trudy et al., 2020) 

aka, for ANZ to be proactive - but how?]]> 2022-10-13 22:34:05 UTC https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339481543 josullivan0308 https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339481721 a central issues of consumers is they feel a lack of control over their own data with the current privacy settings & options available to them (Goggin et al 2019). authors propose the need for 'better policy' to respond to these concerns]]> 2022-10-13 22:34:25 UTC https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339481721 josullivan0308 https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339482591 70% of Australians consider protection of personal info a major concern (oaic survey). 83% say they would like the government to do more to protect their data privacy (2020 community attitudes survey)]]> 2022-10-13 22:35:54 UTC https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339482591 https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339482811 Blacked box nature of AI models raises regulatory challenge in transparency and auditing of finance use cases such as lending. Difficult or impossible to audit machine learning mode.

Lack of explainability disallows supervisors to follow process that led to biased outcomes. ]]> 2022-10-13 22:36:18 UTC https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339482811 trammtran8 https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339483133 Transparency in data: 
examining all aspects of the accuracy of accumulated information ensuring its currency, relevance and completeness. Information that is incomplete, out of date, or ostensibly irrelevant should not be inputted into an AI system in order to derive vital consumer information or make consumer related decisions. (Trudy et al., 2020)]]> 2022-10-13 22:36:51 UTC https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339483133 Explainability https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339483289 Def: The underlying drivers of AI's decision, in other words, why and how the model generates certain results.

lack of trust of users and supervisors around AI applications because of lack of ‘explicit declarative knowledge.’]]> 2022-10-13 22:37:08 UTC https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339483289 josullivan0308 https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339483317 aus government generally more trusted with consumers' personal info than individual corporations or businesses -> in which case solutions/regulations should come from the government or any actions ANZ makes be backed by the government in some way]]> 2022-10-13 22:37:10 UTC https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339483317 https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339486887 EU strategy: establish risk-based legal framework for AI according to high-level ethical standards imposing liability but also providing enough flexibility for businesses to be practical and not held back by legality in innovating new business models with AI tech.

AI technology should not be regulated but the risks incurred by AI usage, especially high-risk use case needs legal safeguards but the low-risk types should be self-regulatory.]]> 2022-10-13 22:43:11 UTC https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339486887 Why Machine Learning? https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339487748 Financial institutions need credit scores typically based on borrowers’ payment history to ensure efficient market behavior. The advent of big data has provided more information to consider. Machine learning is one tool to extract information about prospective borrowers]]> 2022-10-13 22:44:42 UTC https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339487748 Q2 https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339489470 2022-10-13 22:47:25 UTC https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339489470 Q3 https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339491150 2022-10-13 22:50:08 UTC https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339491150 Data to be used https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339492127 1. Disparate impact: A creditor employs facially neutral policies or practices that have an adverse effect or impact on a member of a protected class unless it meets a legitimate business need that cannot reasonably be achieved by means that are less disparate in their impact

The second half of the definition provides lenders the ability to use metrics that may have correlations with protected class elements so long as it meets a legitimate business need, and there are no other ways to meet that interest that have a less disparate impact

2. Banks use long-standing loan records to train artificial intelligence, but these lending decisions are all to some degree detrimental to specific populations]]> 2022-10-13 22:51:44 UTC https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339492127 josullivan0308 https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339492902 ANZ sometimes shares consumer info with third parties 'this helps us make an informed decision about your application' (privacy policy)]]> 2022-10-13 22:53:03 UTC https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339492902 Q3 https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339493019 2022-10-13 22:53:15 UTC https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339493019 josullivan0308 https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339493619 'explorations in governance are at an early stage in all jurisdictions' -> must promote development of governance that matches the level of innovation of new technologies! (human rights ai paper 2019)]]> 2022-10-13 22:54:24 UTC https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339493619 josullivan0308 https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339493988 having such an organization would have economic benefit by avoiding the development & failed deployment of systems which were found to be unfairly biased -> this would be better anticipated. ould also avoid brand damage & financial loss by better identifying the risk & beginnings of AI discrimination]]> 2022-10-13 22:55:06 UTC https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339493988 Proxy Discrimination https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339494049 When AI uncovers a statistical correlation between a certain behavior of an individual and their likelihood to repay a loan, that correlation is actually being driven by two distinct phenomena: the actual informative change signaled by this behavior and an underlying correlation that exists in a protected class. They argue that traditional statistical techniques attempting to split this impact and control for class may not work as well in the new big data context.]]> 2022-10-13 22:55:14 UTC https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339494049 https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339494983 More investments, more algorithms
The key elements of artificial intelligence innovation are (1) data; (2) computing power; and (3) algorithms. The need for better, cleaner data and more computing power requires significant funding. ]]> 2022-10-13 22:56:48 UTC https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339494983 Current Effort https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339496908 The scientific research in this field can broadly be categorized into two groups: One group of researchers seeks to achieve fairness by modifying the underlying data. The other group aims to achieve fairness by modifying the classifiers.]]> 2022-10-13 23:00:20 UTC https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339496908 Q2 https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339496956 2022-10-13 23:00:25 UTC https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339496956 trammtran8 https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339497428 "Machine learning has been in ANZ for 20 years in risk management in the construct of what we call application scoring." - ANZ's Chief Risk Officer, Jason Humphrey
Humphrey said for years the blue bank has been capturing information such as a customer's profile and how they perform over 18-24 months to determine whether they, for example, pay on time, and build statistical models around those attributes.
Techniques have been used for a while, but it is due to the complexity of these digital models that ANZ had never had the compute power to be able to run them - but now they do. 

ANZ has been working on automating home loan approvals, with Humphrey describing it as, traditionally, a very document-driven process. (Barbaschow, 2021)]]> 2022-10-13 23:01:16 UTC https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339497428 https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339497508 2022-10-13 23:01:25 UTC https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339497508 Case of Apple and Goldman Sachs https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339498201 The issue of unintended bias from credit algorithms rose to prominence in late 2019 when Apple and Goldman Sachs ran into trouble when their credit card was accused of offering smaller lines of credit to women than to men.

]]> 2022-10-13 23:02:40 UTC https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339498201 https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339498313 2022-10-13 23:02:54 UTC https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339498313 Insufficient design of AI https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339499453
  • Gives undue weight to a particular data set
  • Relies on a data set that is incomplete, out of date or incorrect, or
  • Uses a data set that is affected by selection bias—that is, where the data set is not representative of a population so may ultimately favor one group over another

    • ]]>     2022-10-13 23:05:14 UTC https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339499453     https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339500609 1. Working with other organisations
    2. Performing administrative and operational tasks
    3. Marketing product and services
    4. Considering any concerns or complaints you raise against ANZ
    5. Preventing or investigating any actual or suspected fraud, unlawful
    activity or misconduct. 
    6. Identifying you or establishing your tax status
    7. As required by laws, regulations, codes of practice and external
    payment systems. 
    (privacy policy)]]>     2022-10-13 23:07:07 UTC https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339500609     Q3 https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339500974 2022-10-13 23:07:44 UTC https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339500974 Logic for Q3 https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339504136 1. Why using AI & Machine Learning?
    2. What's the problem now?
    3. What has been done?
    4. What needs to be done?]]>     2022-10-13 23:12:15 UTC https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339504136     1. Why using AI & Machine Learning? https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339504676 2022-10-13 23:13:10 UTC https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339504676 2. What's the problem now? https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339504912 2022-10-13 23:13:36 UTC https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339504912 3. What has been done? https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339505415 2022-10-13 23:14:22 UTC https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339505415 Case of the Commonwealth Bank of Australia https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339506441 1. The CommBank app uses information the same way for all users.  Customers can provide negative or positive feedback on Bill Sense directly through the app.

    2. CBA also has an Ethical AI service to investigate their models for any disparities (for example, to check a model is not behaving undesirably for different groups of customers). ]]>     2022-10-13 23:16:14 UTC https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339506441     https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339506603 In the US, A set of existing metrics, including income, credit scores (FICO), and data used by the credit reporting bureaus, are deemed acceptable despite having substantial correlation with race, gender, and other protected classes.
    Has a connection with disproportionate denials of home loans for Black and Latino borrowers.]]>     2022-10-13 23:16:33 UTC https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339506603     4. What needs to be done by ANZ? https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339506749 Our recs.]]> 2022-10-13 23:16:48 UTC https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339506749 INSIGHT 1: regulatory perspective trammtran8 https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339532964 Australian government does not have strict regulations surrounding the use of AI. But governments both in AUS and internationally have proposed advances to this. ]]> 2022-10-13 23:54:27 UTC https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339532964 INSIGHT 2: consumer perspective trammtran8 https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339537858 Consumers have a lack of control and understanding in their data and seek governments and regulatory bodies to change this.

    Banks need to explain their the way they use consumer data using AI (i.e. be more transparent)

    Link to TRANSPARENCY principle]]>     2022-10-14 00:00:43 UTC https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339537858     INSIGHT 3: Problems of AI and what should be done trammtran8 https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339544173 AI promotes more efficient market transactions, however, there are key problems that ANZ should address: 
    1) Use of AI exacerbates discrimination 
    2) Banks use longstanding information (historical and proxy bias) 
    3) Insufficient design of AI
    Two case studies from Apple and Commbank. 
    Relate to home loan services in ANZ.

    Link to FAIRNESS principle]]>     2022-10-14 00:08:56 UTC https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339544173     Potentional Solutions: trammtran8 https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339546576 1) Government models for businesses to utilise to assess their use of AI - making application to gov't to test their product in a safe space - 'sandbox'
    2) Ethical AI to check on financial AI ]]>     2022-10-14 00:11:47 UTC https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339546576     INSIGHT 4: ANZ's AI usage trammtran8 https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339555021 ANZ is using automated technology for loan approvals. They are using records from the past, so cannot avoid historical bias. Lack of clear AI policy = lack of transparency.

    Potential solution: develop an AI policy that promotes transparency and fairness.]]>     2022-10-14 00:20:04 UTC https://padlet.com/josullivan0308/fzqav7fdvs7xttfm/wish/2339555021