Through the process of training, generative AI models such as OpenAI’s GPT-3.5 – referred to as large language models (LLMs) when trained on text data – embed knowledge and facts from the source data that equips them with properties, which means they can be used as the foundation upon which other models can be developed, such as chatGPT.
“In the case of chatGPT, it leverages the knowledge embedded within GPT-3.5, and has been optimized – using human feedback – to support plausible dialogue and conversations,” said Saurabh Johri, PhD, chief scientific officer at Babylon, an AI-powered telemedicine technology and services company.
We interviewed Johri to find out what generative AI can potentially do in healthcare, and specifically in telehealth.
Q. You describe generative AI as a transformational force in healthcare. Please elaborate.
A. Few industries are as data-rich, text-heavy and in critical demand for automation as healthcare. Beyond these attributes, there is sharp information asymmetry that exists both for patients and for clinicians; patients want to be better informed about their health, and clinical teams yearn for more timely, easily accessible insights on their patients and populations to better inform their delivery of healthcare.
In this context, generative AI models, particularly LLMs such as GPT-3.5 (and others such as Google’s PaLM) and derived technologies such as chatGPT have the potential to transform healthcare. On the patient side, generative AI can generate rich, accurate medical advice and information (from multiple independent sources) to better inform and educate patients on their condition or symptoms.
LLMs such as GPT-3.5 can also adapt the information and advice – through a process of “style transfer” – for ease of comprehension, for example, by removing medical jargon, simplifying the language for different reading abilities or translating the content into other languages.
On the clinician side, generative AI has the potential to reduce the administrative burden on clinicians, for example, by automating tasks such as writing referral letters, clinical coding and the summarization of clinical consultations.
Beyond these use cases, generative models such as GPT-3 are opening up the possibility of providing healthcare professionals with timely, easily accessible insights on their patients and populations through simple conversational interfaces.
Q. Your company uses generative AI as part of your telemedicine services. Please explain how this form of AI helps caregivers working via telemedicine.
A. We are deploying AI technology – including generative AI models – across our technology platform to support our members and healthcare professionals delivering telemedicine consultations.
We have developed and deployed proprietary generative AI models to understand better the evolving risk profile of members/patients on our platform to ensure that our clinical teams can prioritize the highest-need members first.
We have also developed generative AI models optimized for telemedicine consultations to automatically summarize patient-clinician consultations in near-real time, reducing the administrative burden placed on clinicians and supporting more focused consultations with their patients.
Our experience in developing AI technology for telemedicine and healthcare ensures that we are continuing to innovate with AI technology to support our care teams with timely access to insights and evidence-based clinical next-best actions to support the highest quality of care for their patients.
To this end, we are developing solutions to provide our clinical teams with access to predictive insights and care recommendations that are delivered through intuitive, conversational experiences that are supported by generative AI.
Q. Some say 2023 will be the beginning of “the AI model” in healthcare. Why this year? What has happened to make this year the year for this kind of technology?
A. The general release of chatGPT at the end of 2022 provided the public with access to the power of rapidly developing AI technologies in the field of generative AI, which until then had been restricted to major academic and commercial AI research labs.
Besides the obvious capabilities of the technology, its rapid adoption can be attributed to its accessibility. This is made possible by the technology’s conversational interface, which has allowed professionals from various fields, including healthcare, to discover the potential of AI without having to write a single line of code.
Faced with the day-to-day realities of an over-burdened healthcare system, clinicians are being asked to do more with less. Driven by necessity, a growing community of healthcare innovators are experimenting with chatGPT and an emerging ecosystem of tools, to assess where and how this technology can support automation in clinical workflows.
Despite the impressive capabilities of technologies like chatGPT, they are still susceptible to errors, referred to as “hallucinations,” whereby unintended outputs are produced that are factually inaccurate and inconsistent with the input text.
These errors can be overcome by adapting or fine-tuning generative AI models with rich clinical data, and verifying the outputs of these systems with robust sources of clinical knowledge – work that we are undertaking at Babylon to deliver technologies that support increasingly powerful predictive capabilities but also consumer-centric conversational capabilities.
Now that the technology is here, it is up to industry leaders and innovators to adopt it.
Follow Bill’s HIT coverage on LinkedIn: Bill Siwicki
Email the writer: email@example.com
Healthcare IT News is a HIMSS Media publication.