“We will kill all the birds,” said the Frenchwoman excitedly on Skype.
It was only later that her American colleagues realized what she meant to say: “We will kill two birds with one stone.” The idiom was hilariously lost in translation.
Languages are complex. Consider that in English, a “bank” is an institution where you keep your money, but in Dutch, the word means “cough.” The misuse or misinterpretation of a word or gesture could result in an amusing anecdote, or worse, be offensive. But in the clinical care setting, it could mean life or death.
The clinical care setting also has its own language of complexities and nuances. Individual hospitals may even have their own shorthand and slang. Instead of “Ampicillin and Gentamicin” one might say “Amgen.” A clinician might say aloud to a colleague, “I need to go look at the results from the chest X- Ray” but might use the abbreviation “CXR,” when writing or typing. If emerging Artificial Intelligence technology is going to be truly useful to clinicians, it will need to understand these nuances of clinical care language across many locations and use cases.
What is Natural Language Processing?
On the frontier of healthcare technology is the development of algorithms that can learn the language of clinicians, how they speak and what they’re looking for. One goal is to build platforms that “understand” what a clinician means and can support the clinician in the clinical care setting.
Human interaction with computers has evolved over time. Natural Language Processing (NLP) is the ability for a computer to accept input, spoken or typed, in a conversational manner – even using slang or shorthand. You may have first-hand experience of this with Alexa and Siri.
But for NLP to work across healthcare, it will need to evolve to another level of sophistication. The technology will need to synthesize a digital form of “empathy” for the clinician-user’s intent. A comparative question could be: Will clinicians be able to interact with computers as they do with each other? The best way to learn a language is to be immersed in it. How can we immerse the computer in the clinical language and teach it to intuitively understand it?
Learning to “Understand” Humans
GE Healthcare and Roche Diagnostics have been working with a team of nurses to capture more than 100,000 vocal utterances from clinicians – impressions of how they might ask for something in the clinical care setting. These impressions are the first step for NLP algorithms that may help build a conversational bot to interact and respond to clinicians’ enquiries. The bot is in essence “adapting” to human intent through the input of thousands of clinicians. The goal is to teach the bot how to decipher the clinical language and glean that for which the clinician is asking.
Envisioning the future, the bot would become like a fellow collaborator who’s “with you” in the noise and chaos of the ward floor, like a fellow colleague who truly “gets you” and your clinical intent. These impressions and algorithms are the first step toward making this a reality.
Why is this important? On any given shift today, a clinician may be tending to fifteen patients that are extremely sick, with multiple medications prescribed for each. Patients are having lab work done throughout the day and there are numerous tasks to attend to every shift.
What if instead of a deep dive into the Electronic Medical Record (EMR), with numerous clicks to find data, one could simply ask the bot (cognitive agent) questions via NLP: “What is my patient’s current diagnosis?” “Who is next of kin?” etc. Could such a bot help the clinician-user spend more time looking after their patients and less time searching through EMR menu screens? Could it help reduce the mundane aspects of their work and liberate time to do what they were trained to do?
Technology That Gets Smarter Over Time.
It is the early days of development of such technology. As NLP algorithms get exposed to hundreds of thousands, if not millions, of clinician impression, data scientists hope to continue to improve its ability to understand and interact with the user.
One day, a voice-activated conversational bot* could become a valuable, intuitive member of the clinical care team. If it is done right, it could not only liberate the clinician to care for patient but also augment the clinical intelligence of caregivers.
*Technology in development that represents ongoing research and development efforts. These technologies are not products and may never become products. Not for sale.