Building Trustworthy AI. Main questions, possible solutions and a case study as example
Korhonen, Satu (2022)
Korhonen, Satu
2022
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:amk-202202112379
https://urn.fi/URN:NBN:fi:amk-202202112379
Tiivistelmä
Recent years has seen a surge of ethical guidelines from companies and institutions concerning artificial intelligence and regulation is fast approaching. The purpose here was to take a closer look at the research done in trustworthy AI utilizing one framework, that of the EU, and develop a set of questions that would aid in developing trustworthy AI solutions with the goal to realize the potential benefits of AI while safeguarding individuals and the society against the potential issues involved. A second goal was to utilize this set of questions in developing a proof-of-concept phase execution of trustworthy AI for Aveti Learning as well as evaluate its trustworthiness and identify directions for further development. The data had problems especially in completeness, but a K-Means cluster algo-rithm followed by a Random Forest classifier was developed to allow for Aveti’s mentors to find stu-dents in need of help. The Random Forest algorithm was deployed as a REST API app utilizing Flask. Also, security features such as a rate limiter was implemented. A failsafe method was created in case of environmental difficulties and can be incorporated into the learning platform. The questions created and adopted served to focus the development on all aspects of trustworthiness and seem to be a useful tool in creating more trustworthy AI solutions.