Hazel Tang A science writer with data background and an interest in the current affair, culture, and arts; a no-med from an (almost) all-med family. Follow on Twitter.
Although human adopt a mass society outlook, we do not evolve with a natural ability to form swarm intelligence. Unlike birds, bees, and fish which either manoeuvre in a crowd, or make use of the environment around them to conjure tight feedback loops among members, human are not able to create such subtle connections.
Nevertheless, things are changing rapidly with technology, as we are now able to connect with anyone around the world and form virtual collaborations (i.e., crowdsourcing and crowdfunding). As artificial intelligence (AI) becomes prominent, it is likely to augment the digitally established human swarm, by integrating all the knowledge, data and insights that are present. These human-in-the-loop (HITL) AI, as believed by some researchers, could even be more powerful than AI itself.
Human-in-the-loop model
Recently, a group of researchers from Stanford University School of Medicine, Duke University Medical Center and artificial intelligence (AI) company – Unanimous AI, conducted a study to find out whether a human-AI synergy could produce better diagnostic accuracy of pneumonia from chest radiographs, in comparison to diagnoses made by a group of expert radiologists and deep learning algorithms.
What they did was to split 13 expert radiologists into two groups to independently estimate if pneumonia is present on 50 chest radiographs. After which, the radiologists will then be asked to collectively assess these radiographs using a real-time swarm intelligence platform. On top of which, two separate deep learning algorithms were also deployed to participate in the assessments.
According to the result published in npj Digital Medicine, the deep learning algorithm outperformed human radiologists but the human-AI synergy model produces the greatest diagnostic accuracy of all. Practically because it harnessed “the advantages of both (human and AI) while at the same time overcoming their respective limitations”.
Clinical significance
Researchers did point out that since the study made use of only 50 cases, it is hard to achieve the best clinical ground truth. Also, the study only reports if a patient has pneumonia or not but does not underline other pathologies that trigger pneumonia or being clinical treated as one.
Despite so, researchers believe the study has several clinical importance. First of all, physicians could leverage on AI to improve their operational efficiency, especially when there is an increase in the volume of clinical cases and medical record documentation is taking up too much time.
Researchers suggested radiologists could make use of deep learning algorithms to generate quick, automated diagnoses for high-confidence cases as a triage tool, so that more time can be spent on the more complex cases.
Author Bio
A science writer with data background and an interest in the current affair, culture, and arts; a no-med from an (almost) all-med family. Follow on Twitter.
