Work

ART of AI governance : Pro-ethical conditions driving ethical AI governance

Year: 2021
Type: dissertation
Abstract: In recent years, artificial intelligence has transformed world from agriculture to health care. Alongside even greater opportunities, challenges also rise. Despite its young age, the field of AI ethics has revealed many problems, such as biases and opaqueness of AI systems. Guidelines for ethical AI exist, but a global consensus is yet to be achieved. A gap between ethics and practice is starting to shift focus from abstract ethical constructs towards organizations and practice. Starting point of this thesis is that bottom-up ethics work falls under AI governance, which is another area of active research. AI governance models range from procurement to systematic scale, but organizational models are scarce. Because of its many sides, AI governance in the context of organizations that develop AI also need further defining. This thesis addresses these gaps by focusing on the link between AI governance and AI ethics in organizations developing AI. We develop an AI governance model on the basis of literature utilizing proto-ethical constructs of accountability, transparency and responsibility (ART). This model is then enriched with data gathered from semi-structured interviews of 10 Finnish AI professionals. The purpose is to explain relationships between vendors and customers and highlight possibilities and challenges of implementing practical ethics into AI development process. The findings reveal that ethics currently manifest in AI development, especially in the banking and financial sector. Organizational AI governance offers many avenues to implement, enforce and oversee ethics not just in the development process, but in the organization as a whole. The most remarkable finding is that, while vendor organizations display responsibility in their work, a threat to ethical AI often rises from the customer's side. The interviewees noted that customers do not seem to be interested in paying for ethics, and because of customer's ultimate decision-making power, this can hinder vendors' from giving advice on the ethics of AI. This finding requires further investigation, for which this thesis offers a good starting ground.
Institution: University of Turku
Language: English
Open Access status: green
OpenAlex ID: W3194738552
Cites: 75
Cited by:
Related to: 10