Page 18 - FIGI - Big data, machine learning, consumer protection and privacy

logical reasons. It is widely acknowledged, though, Consumer protection laws typically involve the appli-
that all consumers are vulnerable in some respects. cation of rules, principles and procedures to give
We cannot know everything at all times. We have a consumers certain rights relating to the products
limited ability to assess risk and benefits, i.e., we are and services they purchase. These rights include:
subject to “bounded rationality.” 48
In consumer protection, the State intervenes • rights prior to purchase (pre-engagement), such
through laws and processes in what would otherwise as information about the product or service pro-
be a private relationship between consumer and pro- vided;
vider. The need for this arises from perceived asym- • the provision, quality and functioning of the prod-
metries between providers and consumers. These uct or service itself (engagement); and
may include information asymmetries, where provid- • post-purchase means of holding providers
ers have greater data, knowledge and understanding accountable (post-engagement).
than consumers. Differences in economic scale can
also result in severe asymmetries of bargaining pow- The FAT values may apply in the pre-engagement
er. In addition, the transaction costs that consumers phase, requiring notification to consumers about the
would face if they had to negotiate assurances about product or service they are getting and sometimes
every product or service they acquire are too high to securing express consent to it so that the consumer
be feasible. As a result, a purely private, negotiated can take responsibility for their decisions.
bargain between consumer and provider would be However, a substantial part of consumer protec-
one-sided. tion law operates on the premise that even if the con-
Consumer protection is formulated in various sumer is notified about and consents to a product
ways, but commonly seeks to promote the values or service on the offered terms and conditions, such
of fairness, accountability and transparency (FAT). consent alone may not adequately achieve fairness,
49
The policy debate around consumer protection in accountability and transparency. Thus, the FAT values
relation to artificial intelligence and machine learn- may also apply in the engagement phase, i.e., to the
ing concerns the capacity of algorithms and machine actual product or service itself – its safety, quality or
learning systems to reflect such values. Consumers other features and conditions of provision. Therefore,
50
may be vulnerable when dealing with services rely- consumer protection laws go further than pre-en-
ing on computer processing for numerous reasons. gagement notice and consent where notice and con-
Their functioning exceeds the comprehension of sent would not sufficiently protect the consumer and
most of the population. Their precise, digital pro- should not alleviate responsibility of the provider.
cesses and results have a “seductive precision of out- Again, FAT principles apply also in the post-en-
put.” As a result, computers and results driven by gagement phase to ensure accountability mecha-
51
them may be perceived as being objective and even nisms for securing explanations of why a given prod-
fair. Today, however, there are risks that consumers uct or service was provided in the manner it was.
will find some aspects of digital services to be unfair, They provide for consumers to have an opportunity
unaccountable and non-transparent (the opposite to contest such decisions, and a means of redress
of FAT), undermining trust between consumers and where harm has resulted. Such protections may be
service providers and so hampering the prospects applied regardless of whether the consumer has con-
for growth in digital services. sented otherwise. For instance, many countries’ laws
do not permit consumers to submit to certain types

16 Big data, machine learning, consumer protection and privacy