A Review on Human–AI Interaction in Machine Learning and Insights for Medical Applications

Abstract

Objective: To provide a human–Artificial Intelligence (AI) interaction review for Machine Learning (ML) applications to inform how to best combine both human domain expertise and computational power of ML methods. The review focuses on the medical field, as the medical ML application literature highlights a special necessity of medical experts collaborating with ML approaches. Methods: A scoping literature review is performed on Scopus and Google Scholar using the terms “human in the loop”, “human in the loop machine learning”, and “interactive machine learning”. Peer-reviewed papers published from 2015 to 2020 are included in our review. Results: We design four questions to investigate and describe human–AI interaction in ML applications. These questions are “Why should humans be in the loop?”, “Where does human–AI interaction occur in the ML processes?”, “Who are the humans in the loop?”, and “How do humans interact with ML in Human-In-the-Loop ML (HILML)?”. To answer the first question, we describe three main reasons regarding the importance of human involvement in ML applications. To address the second question, human–AI interaction is investigated in three main algorithmic stages: 1. data producing and pre-processing; 2. ML modelling; and 3. ML evaluation and refinement. The importance of the expertise level of the humans in human–AI interaction is described to answer the third question. The number of human interactions in HILML is grouped into three categories to address the fourth question. We conclude the paper by offering a discussion on open opportunities for future research in HILML.

Potential benefits and limitations of machine learning in the field of eating disorders: current research and future directions

“…Identifying eating disorder risk based on digital data, such as that on social media, and providing help seeking information may also be distressing for those who were unaware of their vulnerability. Thus, researchers are advocating for the importance of collaboration between humans and ML algorithms in healthcare settings to make the most accurate and appropriate recommendations [ 79 , 80 ]. Human involvement is needed because ML methods are not always transparent (i.e., it is sometimes unclear what predictions were based on), ML predictions can be inaccurate, and ML methods may not capture all of the intricacies of each specific situation [ 79 ].…”

“…Thus, researchers are advocating for the importance of collaboration between humans and ML algorithms in healthcare settings to make the most accurate and appropriate recommendations [ 79 , 80 ]. Human involvement is needed because ML methods are not always transparent (i.e., it is sometimes unclear what predictions were based on), ML predictions can be inaccurate, and ML methods may not capture all of the intricacies of each specific situation [ 79 ]. Further, human involvement is important for therapeutic alliance which can improve treatment outcomes and can reduce higher dropout rates associated with completely remote care [ 81 , 82 ].…”

“…Thus, ML may be a useful tool to aid clinicians in the identification and treatment of eating disorders. Human-in-the-loop ML allows people to change the results of ML based on their skills and expertise, which can improve the power of ML to deal with complicated real-world tasks [ 79 ]. For example, ML models could be guided by researchers, clinicians, and those with lived experience to make recommendations based on available data, and then clinicians could decide whether to enact those recommendations based on their own clinical experience and knowledge.…”

A Cognitive Framework for Delegation Between Error-Prone AI and Human Agents

“…With HCAI scenarios, the focus is with the human user or collaborator. This can take the form of humans using algorithm output to aid in decision-making [14], [15], shared autonomy [16], sensory augmentation [7], autonomous driving [17]- [19], human-robotic collaboration [20], etc. With HCAI focusing on different mechanisms supporting the combination of human and artificial intelligence, or the augmentation of human reasoning, it is important to consider potential errors, of either AI, or human, or both.…”

“…In the base case, systems are designed to utilize input data to make a decision or classification for human use. This can range from medical diagnostics [14], [15] to social cue interpretation [7]. In this case, humans utilize the output of the system to guide their actions.…”