AI that can be explained Explainable AI (XAI) refers to
approaches or design decisions used in automated systems such that artificial
intelligence and machine learning produce outputs with a logic that humans can
understand and explain.
The extensive usage of algorithmically assisted
decision-making in social situations has raised considerable concerns about the
possibility of accidental prejudice and bias being encoded in the choice.
The fact that a human operator is not involved in automated
decision-making does not rule out the possibility of human bias being embedded
in the outcomes produced by machine computation.
Artificial intelligence's already limited accountability is
exacerbated by the lack of due process and human logic.
The consequences of algorithmically driven processes are
often so complicated that even their engineering designers are unable to
understand or predict them.
The black box of AI is a term that has been used to describe
this situation.
To address these flaws, the General Data Protection
Regulation (GDPR) of the European Union contains a set of regulations that
provide data subjects the right to an explanation.
Article 22, which deals with automated individual
decision-making, and Articles 13, 14, and 15, which deal with transparency
rights in relation to automated decision-making and profiling, are the ones to
look out for.
When a decision based purely on automated processing has "legal implications" or "similarly substantial" effects on a person, Article 22 of the GDPR reserves a "right not to be subject to a decision based entirely on automated processing" (GDPR 2016).
It also provides three exceptions to this right, notably
when it is required for a contract, when a member state of the European Union
has approved a legislation establishing an exemption, or when a person has
expressly accepted to algorithmic decision-making.
Even if an exemption to Article 22 applies, the data subject
has the right to "request human involvement on the controller's side, to
voice his or her point of view, and to challenge the decision" (GDPR
2016).
Articles 13 through 15 of the GDPR provide a number of
notification rights when personal data is obtained (Article 13) or from third
parties (Article 14), as well as the ability to access such data at any time
(Article 15), including "meaningful information about the logic
involved" (GDPR 2016).
Recital 71 protects the data subject's right to
"receive an explanation of the conclusion taken following such evaluation
and to contest the decision" where an automated decision is made that has
legal consequences or has a comparable impact on the person (GDPR 2016).
The question of whether a mathematically interpretable model
is sufficient to account for an automated judgment and provide transparency in
automated decision-making is gaining traction.
Ex-ante/ex-post auditing is an alternative technique that
focuses on the processes around machine learning models rather than the models
themselves, which may be incomprehensible and counterintuitive.
You may also want to read more about Artificial Intelligence here.
See also:
Algorithmic Bias and Error; Deep Learning.
Further Reading:
Brkan, Maja. 2019. “Do Algorithms Rule the World? Algorithmic Decision-Making in the
Framework of the GDPR and Beyond.” International Journal of Law and Information Technology 27, no. 2 (Summer): 91–121.
GDPR. 2016. European Union. https://gdpr.eu/.
Goodman, Bryce, and Seth Flaxman. 2017. “European Union Regulations on Algorithmic Decision-Making and a ‘Right to Explanation.’” AI Magazine 38, no. 3 (Fall): 50–57.
Kaminski, Margot E. 2019. “The Right to Explanation, Explained.” Berkeley Technology Law Journal 34, no. 1: 189–218.
Karanasiou, Argyro P., and Dimitris A. Pinotsis. 2017. “A Study into the Layers of Automated Decision-Making: Emergent Normative and Legal Aspects of Deep Learning.” International Review of Law, Computers & Technology 31, no. 2: 170–87.
Selbst, Andrew D., and Solon Barocas. 2018. “The Intuitive Appeal of Explainable Machines.” Fordham Law Review 87, no. 3: 1085–1139.
