Data and algorithm ethics: privacy, fairness, and explainability

• Pages: 34-51

Chapter 2

Data and

algorithm ethics:

privacy, fairness,

and explainability

35Ethics of Connected and Automated Vehicles

2.1 INTRODUCTION

CAV operations require the collection and use of great volumes and varied

combinations of static and dynamic data relating to the vehicle, its users,

and the surrounding environments. Through algorithms and machine learning,

these data are used for CAV operations on dif‌ferent time scales, ranging from

second-by-second real-time path planning and decision-making, to longer-

term operational parameters concerning choice of routes and operating

zones, up to longest-term user prof‌iling and R&D investments.

Consequently, data subjects need to be

both protected and empowered, while vital

data resources need to be safeguarded

and made accessible to specif‌ic actors.

This can only occur aer due consideration

of ethical principles of human dignity and

personal autonomy. In this context, these

fundamental principles are tied to specif‌ic

principles concerning privacy, fairness, and

explainability.

First, the notion of privacy encompasses

each individual’s authority to determine a

private sphere for personal conduct and

self-development, including privacy of

communications and the ability to control

the terms and conditions of personal

information sharing. Privacy is not only

an ethical imperative but an enforceable

fundamental right in the EU. Standardly,

respect for privacy requires a valid legal

basis (pursuant to Article 6 GDPR) for any

collection, processing, use or exchange of

personal data.

Second, fairness and explainability are

binding data protection principles that are

enshrined in secondary EU law (e.g., the

GDPR, the Law Enforcement Directive, and

the data protection instruments that apply

to the EU institutions). Fairness requires

that personal data collection, processing,

uses, and outcomes do not discriminate

negatively against any individual or

group of data subjects. This entails that

data-driven CAV operations should be as

inclusive as possible, and that equal access

and opportunities need to be safeguarded

for all parties, particularly for potentially

vulnerable persons and groups.

Finally, in line with previous reports27,

explainability (Explainable AI) requires that

the objectives, mechanisms, decisions and

actions pursued by data- and AI-driven

CAV operations should be accessible,

comprehensible, transparent and traceable

to users and data subjects, in a way

that goes beyond a strictly technical

understanding for experts.