Research using artificial intelligence has grown during the pandemic, sparking concerns over methods
Researchers and clinicians are using artificial intelligence technologies to manage and understand the coronavirus pandemic, but the European Commission’s Joint Research Centre evidence service has raised concerns about rushing to trial such algorithms in a crisis.
A JRC analysis of the PubMed publications database found an uptick in the number of papers referring to AI between January and June. But the JRC researchers warned that “many studies seem to be affected by problems related to [the] difficulty in accessing the relevant data, small samples, biases in the data, or inconsistent labelling of the training datasets”.
The researchers, working as part of the Commission’s AI Watch project, found that data-driven AI has mainly been applied to four broad types of medical task: individual diagnosis, outbreak prediction, social care and resource management.
They warned that there is a “lack of standards for evaluation and international coordination” of medical R&D using AI, and issues around appropriate data selection and curation for training systems. “Given the state of the art, we cannot yet trust an algorithm on its own to support decisions influencing human lives,” they said.
The analysis report, published on 27 July, is the second from AI Watch, a group of JRC staff assigned to monitor and report on developments in the field for the Commission and other policymakers. Its language echoes many political talking points on the Commission agenda, including the concept of technological sovereignty and a push for faster digitisation of EU societies.