Page 27 - FIGI - Big data, machine learning, consumer protection and privacy
P. 27
as opposed to a non-bank). This may also increase and data subjects to foresee consequences of their
the broader range of data available about consumers, actions”) will operate.
and so enrich and plug gaps in the data ecosystem.
These potential advantages need to be weighed in 4�2 Protecting consumers from bias and discrimi-
light of how the alternative credit market is develop- natory treatment
ing. Loans made using alternative data and automat-
ed decisions are often small (e.g., to tide someone Biased inferences and decision-making outputs
over until the end of the month), and so their results While one concern arising with big data is how input
are possibly of limited utility. The new and grow- data, such as name, age and other personal data, will
ing market in automated lending using proprietary be used and protected, another relates to the infer-
algorithms to evaluate borrowers with no traditional ences that result from processing such data. Just as
credit history is also highly innovative. Requiring new important as the accuracy of the input data is the
innovative lenders to share their lending results may manner and accuracy of the inferences big data and
deprive them of some of the benefits of their invest- machine learning will draw from it about individuals
ment and first mover advantage. In addition, such and groups, and the impact of such inferences on
firms are often entrepreneurial start-ups that may decisions. Some such inferences, which predict
117
struggle with weighty reporting obligations as they future behaviour and are difficult to verify, may
seek to grow a risky business. Some do not even rely determine how individuals are viewed and evaluated
on credit reference bureau data themselves for their and so affect their privacy, reputation and self-de-
own lending decisions (relying entirely on alterna- termination.
tive data), which may weaken the logic of reciproci- Data protection laws that govern the collec-
ty inherent in credit reference bureaus (where those tion, use and sharing of personal data typically do
supplying data are entitled to rely on the wider pool not address the outputs of machine learning mod-
of aggregated data supplied by others). 116 els that process such data. One of the concerns of
For these reasons, it is important to consider the data protection and privacy law and regulation is to
overall data environment of the financial market as prevent discrimination. Principle 5 of the High Level
it develops, both in relation to the accuracy of data Principles for Digital Financial Inclusion states that
used in automated decisions and how responsibility data should “not be used in an unfair discriminatory
for accurate data should be allocated in the formal manner in relation to digital financial services (e.g., to
credit data reporting systems and more generally. discriminate against women in relation to access to
Given the wide range of data available and its credit or insurance).” 118
varying sources and levels of reliability, there are Recent examples of inferences involving major
numerous policy dilemmas to come regarding how internet platforms concern sexual orientation, phys-
the guidelines on clarity and predictability in the ical and mental health, pregnancy, race and political
fourth General Principle Credit Reporting (“The legal opinions. Such data may be used in decisions about
and regulatory framework should be sufficiently pre- whether a person is eligible for credit. The GDPR
119
cise to allow service providers, data providers, users sets apart special categories of personal data for
tighter restrictions. While personal data is defined as
Monetary Authority of Singapore’s FEAT Principles
Principle 3. Data and models used for AIDA-driven decisions are regularly reviewed and validated for
accuracy and relevance […]
Principle 4. [Artificial Intelligence and Data Analytics]-driven decisions are regularly reviewed so that
models behave as designed and intended.
Smart Campaign’s draft Digital Credit Standards
Indicator 2�1�5�0
Underwriting data and analysis is refreshed at each loan cycle to identify changes in the client’s situ-
ation.
Big data, machine learning, consumer protection and privacy 25