Back

An Opportunity to Standardize and Enhance Intelligent Virtual Assistant-Delivered Layperson Cardiopulmonary Resuscitation Instructions

Murk, W.; Goralnick, E.; Brownstein, J. S.; Landman, A. B.

2023-03-14 health informatics
10.1101/2023.03.09.23287050 medRxiv
Show abstract

ImportanceIntelligent virtual assistants (IVAs) are ubiquitous and hold the potential to provide bystander cardiopulmonary resuscitation (CPR) instructions during an emergency. ObjectiveTo evaluate the quality of CPR instructions provided by IVAs. DesignWe evaluated the appropriateness of responses of four IVAs (Amazon Alexa, Apple Siri, Google Assistant, and Microsoft Cortana) to eight CPR-related questions. We also evaluated text-based responses provided by OpenAI ChatGPT, a recently developed artificial intelligence large language model. ResultsOut of 32 responses provided by IVAs, only 19 (59%) were related to CPR, 9 (28%) suggested calling emergency services, and 4 (12%) provided verbal CPR instructions. All responses provided by ChatGPT were related to CPR and suggested calling emergency services. Among responses related to CPR, the answers provided varied significantly in the utility of information provided. Conclusions and RelevanceThese results highlight the need for the technology industry to partner with the medical community to improve and standardize bystander CPR instruction provided by IVAs. Key pointsO_ST_ABSQuestionC_ST_ABSWhat is the quality of cardiopulmonary resuscitation (CPR) instruction provided by intelligent virtual assistants (IVAs) including Amazon Alexa, Apple Siri, Google Assistant, and Microsoft Cortana? FindingsWhen asked questions related to CPR, IVAs frequently did not provide relevant information. Moreover, even when the information provided was relevant, actual CPR instruction was not given in most cases. MeaningThese results highlight the need to standardize bystander CPR instruction provided by IVAs.

Matching journals

The top 4 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.