Navigating Ethical AI: Lessons from COVID-19 AI Challenges

Abstract

“Europe: Toward a Policy Framework for Trustworthy AI” chapter in the Oxford Handbook of Ethics and AI How would the recommendations apply to some of the challenges described in the following short stories? Trustworthy AI must express consideration of three significant components throughout its operational cycle. The system must be lawful in that it complies with all applicable laws and regulations. Trustworthy AI should also be ethical, thus ensuring adherence to ethical values and principles. Finally, the system must be robust from the technical and social perspective to express consideration of the occurrence of unintentional harm, which even AIs with good intentions can cause. Gauging the trustworthiness of an AI system from such an approach, one can argue that ethical consideration and adherence to any AI system can enhance success in consideration of the other two components as understanding what is written and sticking to it defines adherence t all other laws/rules therein. Considering “Artificial intelligence for COVID-19: savior or saboteur” article, using an untested AI algorithm upon emergency authorization proves a threat. It can result in unreliable data concerning the effects and extent of transmission of the current pandemic Among Americans. One of the significant recommendations of AI HLEG is to adopt a risk-based policy framework which translates accountability and liability elements of ethics guidelines into a legal regime that holds AI practitioners responsible for any risks associated with the deployment of an AI system to the society. The risk of untested AI systems in reporting COVID-19 statistics amongst Americans may imply that even the data produced is unreliable or confusing, impacting society severely. To this effect, the American AI system experts must express consideration of the society through the deployment of the untested AI from servicing the Covid-19 concerns or be ready to face the consequences of any risks therein. In the second article, "Of 300-plus imaging-based AI models for COVID-19 diagnosis, zero...