“Empathetic chatbots will revolutionise AI in Medicine”

Since the launch of ChatGPT, the use of AI in medicine has exploded.  Doctors are not only using AI for administrative purposes, but AI is now used to interpret tests e.g. ECGs, mammograms and brain scans. AI shows far greater accuracy than doctors and patients are now seeing the benefits. Brainomix an  AI platform used to interpret brain scans, has  halved diagnostic time for strokes and has massively improved outcomes by tripling the likelihood of stroke victims living independently post stroke. 

Despite these impressive developments, healthcare has not seen a reduction in workforce due to AI. In fact, AI technology may be serving to “re-humanise” healthcare by allowing doctors to spend more time with their patients. A recent study by MIT of 900+ healthcare professionals demonstrated that 72% of respondents were keen to implement AI to improve efficiency. 

Perhaps the most unexpected innovation in medical AI technology, is the development of empathic machines. One study published by JAMA Intern Med collated the answers doctors and ChatGPT offered to 195 medical questions posted on social media. The results showed that 78% of ChatGPT’s responses were of good or very good quality, while physicians only managed this on 22% of responses. More interestingly, 45% of ChatGPT’s responses were considered empathetic or very empathetic, significantly higher than the physician’s scores of only 4.6%. The AI responses were on average 211 words, therefore much more detailed than physicians average of only 52 words.  

In another recent research paper, a Large Language Model (LLM) AI system optimised for diagnostic dialogue in the medical field, outperformed doctors on 28 out of 32 evaluation axes relating to conversational quality, diagnosis, and management plans. The AI demonstrated greater empathy and more accuracy.  

Since the advent of ChatGPT there has also been a lot of anecdotal reports from doctors using AI to enhance their communications skills. Dr Tamayo-Sarver, an ER physician at the Good Samaritan Hospital, San Jose California, has used AI to aid communication with patients and families in the ER and is a strong advocate for its use to improve doctor-patient communication. Dr Tamayo-Sarver first realised the benefits of using ChatGPT while treating a 96-year-old who presented with breathlessness due to pulmonary oedema (fluid build-up in the lungs). This patient’s family were insistent that she should be given IV fluids and couldn’t understand why fluid should be restricted in this scenario. After several failed attempts to explain, Dr Tamayo-Sarver decided to ask ChatGPT for its explanation. After reading the chatbots empathic and easily understandable reply to the family, he noticed that the families “agitated expressions immediately melted into calm agreeability”. He now uses ChatGPT routinely to help empathically explain specific medical scenarios in an accurate way, that’s understandable for patients. This also has the benefit of being available in written form for the patient so they can revisit the explanation.  

Empathic response chatbots are being introduced into the NHS to assist with the overwhelming number of mental health service referrals.  Limbic Access is an AI assistant for conversational referrals and supports clinical decisions in behavioural health services. A new addition to this platform is Limbic Care, a clinical AI companion that provides conversational support and guided cognitive behavioural therapy. This reinforces therapeutic interventions and assist’s clinicians with their burgeoning number of patients.  

Pioneering interfaces like Empathic voice interface (Hume AI) mean that the AI can now recognise and respond appropriately to nuanced vocal modulations, guiding language and speech generation and will provide an enhanced patient experience. It has been trained on millions of human interactions and is fast responding with high EQ and very eloquent human-like responses. I tried the Hume demo and had a very insightful conversation about AI ethics while the bot analysed my mood via the intonations in my voice. This interface named EVI detects 25 tones and expressions in the human voice to respond appropriately and achieve a fluent, empathic and natural feeling conversation.  Hume have also recently published results of a study of 423,193 facial expressions from 5,833 participants and found that there are 28 distinct dimensions of facial expression, with 21 showing cultural universality. This will ultimately help EVI visually interpret emotion, making it even better at social interaction. This will have huge applications in patient communication.  

Currently AI is acting as an adjunct to medical care, releasing doctors from clerical and administrative tasks to spend more time with their patients. However, AI has already demonstrated the capacity to machine learn. When this was applied by DeepMind technologies to create AlphaFold, a deep learning AI capable of predicting the 3D structure of proteins, it  became the first AI to receive the Lasker Award for Contribution to Medicine. We have also seen the creation of compassionate response AI (ChatGPT) and AI that can read our emotional cues (EVI). DeepMind Technologies has also recently announced the introduction of a revolutionary AI training method: JEST (Joint Example Selection and Trust). This novel way of training AI is 13 times faster and 10 times more energy efficient than previous technologies.   

With the introduction of these new AI technologies and such rapid advancements, we are currently only seeing the tip of the iceberg in terms of potential applications as medicine enters a new era.  

​​​