Austrian Research and Technology Report 2020

180
Austrian Research and Technology Report 2020
ulation (GDPR) gives a person the right not to be 
subject to a decision based solely on automated pro-
cessing. 
In its Ethics Guidelines, the EU Commission’s Ex-
pert Group on Artificial Intelligence defines four eth-
ical principles for trustworthy AI. Alongside the com-
prehensive set of indivisible rights set out in interna-
tional human rights law, the EU Treaties and the 
Charter of Fundamental Rights of the European 
Union, these form the basis for overcoming the chal-
lenges described (see also Fig. 3-8): 
1. Respect for human autonomy
2. Prevention of harm
3. Fairness
4. Explicability
The principle of respect for human autonomy, for in-
stance, requires humans interacting with AI systems 
to be able to keep full and effective self-determina-
tion over themselves. AI systems should not unjustifi-
ably subordinate, coerce, deceive, manipulate, condi-
tion or herd humans. Instead, they should be de-
signed to augment, complement and empower human 
cognitive, social and cultural skills. In accordance 
with the second principle, AI systems should neither 
cause nor exacerbate harm or otherwise adversely 
affect human beings. This entails the protection of 
human dignity as well as mental and physical integri-
ty. Particular attention must be paid to situations 
where AI systems can cause or exacerbate adverse 
impacts due to asymmetries of power or information, 
Fig. 3-8: Framework for trustworthy AI
Trustworthy AI
Legitimate AI
Ethical AI
Robust AI
Foundations of trustworthy AI
Ensuring compliance with ethical 
principles based on fundamental rights
Recognising and releasing 
tension between them
• 
Respect for human autonomy 
• 
Prevention of harm
• 
Fairness 
• 
Explicability
Ensuring implementation of 
core requirements
Continuous assessment and 
consideration of the core requirements 
during the entire life cycle of the AI system
• 
Primacy of human agency and 
human oversight 
• 
Technical robustness and safety
• 
Data protection and
•
Transparency
•
Diversity, non-discrimination and fairness
• 
Social and environmental well-being
• 
Accountability
Ensuring operational 
implementation of the core requirements
Adjustment to the specific AI application
Framework for trustworthy AI
Technical
methods
Non-technical 
methods
Assessment of trustworthy AI
Trustworthy AI assessment
Realisation of trustworthy AI
7 core requirements
4 ethical principles
CHAPTER I
CHAPTER II
CHAPTER III
data quality management
Source: High-Level Expert Group on Artificial Intelligence (2019).

---