DeepScribe, a company focused on leveraging artificial intelligence to serve as a medical scribe, contracts with 200 humans to listen in on medical visit recordings and fix the AI's errors, according to a report in The Wall Street Journal.
DeepScribe's medical scribe platform is used by nearly 1,000 physicians and healthcare providers to record, transcribe and then create physicians' notes based on conversations with patients during medical visits. But the algorithms often have "hallucinations" and provide false information about patients, including errors in listing medications and adding in "disjointed, nonsensical sentences," according to the report.
Each note is reviewed by one of the human contractors for errors, including inaccurate medication information and ICD-10 code generation. The reviewers have minimal ICD-10 training and rely on the physician and medical staff to confirm codes in their reports.
Human workers reviewing the notes increased the AI's accuracy by 15 percentage points to 95 percent, according to the report. But their use also raises ethical questions since the recordings include patients' names and intimate details about their lives.
DeepScribe has raised $37.3 million in venture capital funding since it was founded in 2017 and is on track to hit $6.5 million revenue this year. In 2021, the company was valued at $180 million.