Tesi etd-03272020-115634
Link copiato negli appunti
Tipo di tesi
Dottorato
Autore
GIKAY, ASRESS ADIMI
URN
etd-03272020-115634
Titolo
Automated Consumer Credit Scoring in the European Union and the United States—Perspectives from Consumer Credit Law
Settore scientifico disciplinare
Istituto di Diritto, Politica e Sviluppo
Corso di studi
SCIENZE POLITICHE - INDIVIDUAL PERSON AND LEGAL PROTECTIONS
Commissione
Membro Prof.ssa PONCIBÒ, CRISTINA
Membro Prof.ssa AMATO, CRISTINA
relatore COMANDE', GIOVANNI
Membro Prof.ssa SGANGA, CATERINA
Membro Prof.ssa AMATO, CRISTINA
relatore COMANDE', GIOVANNI
Membro Prof.ssa SGANGA, CATERINA
Parole chiave
- AI Driven Credit Scoring
- Algorthmic Accountability
- Automated Consumer Credit Scoring
- Black Box
- Discrimination and Bias Credit Scoring
- Transparency
Data inizio appello
22/06/2020;
Disponibilità
completa
Riassunto analitico
The existing law and technology literature generally emphasize on the challenges of automated consumer credit scoring and tends to propose legal solutions that are either unrealistic or drastic, that could deter innovation while not making the consumer better off. Moreover, by comingling various automated decision making with different policy concerns and by ignoring the potential solutions for the challenges presented by automated credit scoring available in the existing consumer financial services law, the current legal literature creates an unnecessary confusion with regard to the appropriate legal solutions that can be adopted.
This thesis posits that any regulatory regime for automated consumer credit scoring should depart from the assessment of whether the existing consumer financial law provides a regulatory basis for the phenomenon. Adopting this approach which presumes that law is technology neutral and is able to adapt to emerging technology-based legal challenges is compelling for two reasons.
First, the fundamental basis of consumer financial protection, namely information paradigm, financial literacy and limiting contractual autonomy that are three of the most prevalent regulatory tools are substantially useful tools to regulating automated credit scoring notwithstanding the limitations they suffer. Second, consumer credit law requires financial institutions to comply with certain standards in their decision making such as ensuring accuracy and observing anti-discrimination laws. In the US, these standards are being applied to automated consumer credit scoring. Thus, legal solutions proposed to tackle the challenges of automated consumer credit scoring should not jump to the idiosyncrasies of phenomenon without closely examining the existing consumer financial services law and its responses.
Based on the in-depth analysis of the literature, law, judicial decisions, empirical evidence and industry practice, this thesis proposes six anatomies of the regulation of automated consumer credit scoring. These are (1) technology neutrality as a starting point, (2) separating different areas of automated decision making (3) differentiating between deterministic algorithmic scoring and machine learning-driven scoring, (4), separating different types of consumer credits, (5) creating rules for disclosure of trade secret protected algorithms and (6) and pre-use algorithmic auditing.
Within the framework of adopting technologically neutral regulation that protects the consumer without stifling innovation, the thesis concludes that (a) complete transparency, explainability and auditability in machine learning-driven scoring may not be achieved. Hence, in the field of consumer credits of high socio-economic significance, machine learning-scoring should be banned while it should be permitted in small loan underwritings with lower socio-economic significance.
In the light of the argument that algorithmic transparency is hindered by trade secret law, the thesis argues that rules for disclosure of trade secret protected algorithms should be created. Nevertheless, based on economic analysis of patent regime for algorithms, the thesis rejects the possibility of creating mandatory patent regime for algorithms. Furthermore, compelling businesses to disclose their algorithm in unprotected environment should be rejected due to its adverse effect on innovation.
Within the context of empowering consumers, the commodification of personal data has been debated in legal literature. Nevertheless, recent legislative and judicial development as well as industry practice shows that commodification of personal data is already a practice. Nevertheless, it rejects the doctrine of propertization as counter-productive to overall consumer welfare.
This thesis posits that any regulatory regime for automated consumer credit scoring should depart from the assessment of whether the existing consumer financial law provides a regulatory basis for the phenomenon. Adopting this approach which presumes that law is technology neutral and is able to adapt to emerging technology-based legal challenges is compelling for two reasons.
First, the fundamental basis of consumer financial protection, namely information paradigm, financial literacy and limiting contractual autonomy that are three of the most prevalent regulatory tools are substantially useful tools to regulating automated credit scoring notwithstanding the limitations they suffer. Second, consumer credit law requires financial institutions to comply with certain standards in their decision making such as ensuring accuracy and observing anti-discrimination laws. In the US, these standards are being applied to automated consumer credit scoring. Thus, legal solutions proposed to tackle the challenges of automated consumer credit scoring should not jump to the idiosyncrasies of phenomenon without closely examining the existing consumer financial services law and its responses.
Based on the in-depth analysis of the literature, law, judicial decisions, empirical evidence and industry practice, this thesis proposes six anatomies of the regulation of automated consumer credit scoring. These are (1) technology neutrality as a starting point, (2) separating different areas of automated decision making (3) differentiating between deterministic algorithmic scoring and machine learning-driven scoring, (4), separating different types of consumer credits, (5) creating rules for disclosure of trade secret protected algorithms and (6) and pre-use algorithmic auditing.
Within the framework of adopting technologically neutral regulation that protects the consumer without stifling innovation, the thesis concludes that (a) complete transparency, explainability and auditability in machine learning-driven scoring may not be achieved. Hence, in the field of consumer credits of high socio-economic significance, machine learning-scoring should be banned while it should be permitted in small loan underwritings with lower socio-economic significance.
In the light of the argument that algorithmic transparency is hindered by trade secret law, the thesis argues that rules for disclosure of trade secret protected algorithms should be created. Nevertheless, based on economic analysis of patent regime for algorithms, the thesis rejects the possibility of creating mandatory patent regime for algorithms. Furthermore, compelling businesses to disclose their algorithm in unprotected environment should be rejected due to its adverse effect on innovation.
Within the context of empowering consumers, the commodification of personal data has been debated in legal literature. Nevertheless, recent legislative and judicial development as well as industry practice shows that commodification of personal data is already a practice. Nevertheless, it rejects the doctrine of propertization as counter-productive to overall consumer welfare.
File
| Nome file | Dimensione |
|---|---|
| Regulati...Gikay.pdf | 2.09 Mb |
Contatta l'autore |
|