Table 4.

Challenges and tradeoffs of artificial intelligence in terms of COVID-19.

Challenge	Description / illustrative example	References
Logistical maintenance	Compulsory public health lockdowns can impede communication between machine learning data specialists, engineers, and healthcare personnel to ensure proper usage and maintenance of the AI algorithm.	(Debnath et al., 2020)
Existing AI technique limitations	The performance of deep learning is restricted by the provided human data, inherent computation of all case-related probabilities, and the volume of training data.	(Hussain et al., 2020)
Invasion of individual privacy	Public health COVID-19 responses and guidelines mandated the use of private user data (i.e. contact tracing)	(Naudé, 2020)
Training data shortage	Scarce training data can compromise the performance and efficiency of COVID-19 centered AI models.
Heterogeneous data cohorts	The disparity in COVID-19 variables can influence the results of predictive AI models. Such variables include the incubation period, levels of dyspnea, and oxygen saturation.	(Adly et al., 2020; Debnath et al., 2020)
Chatbots	Chatbots are conversation simulators created to offer an alternative platform of human-AI communication. These systems require long term training and maintenance to minimize the incidence of deficient clinical diagnosis and false response inputs.	(Madurai Elavarasan and Pugazhendhi, 2020)