The human cost of AI in healthcare | James McCormack | TEDxLakeheadU

The Daily Life and Challenges of Doctors

• The average doctor's day involves numerous tasks such as communicating with nurses, nurse practitioners, residents, medical students, and dietitians, as well as checking emails, talking with patients, and writing notations, all of which must be done on top of living a normal life with family and kids 10s.
• A technologist might offer to take over some of the doctor's work, such as writing and other tasks, to allow the doctor to focus on high-value things, and would also offer the opportunity for the doctor to review everything that is done 1m42s.
• When artificial intelligence takes over some of the same tasks, it is often referred to as revolutionary and world-changing, which raises questions about why this is the case when a human technologist would not receive the same praise 2m6s.

The Role of Technology in Healthcare

• Healthcare technology exists at the intersection of health, software, and artificial intelligence, with health encompassing not just physicians but also nurses, pharmaceuticals, and other fields 3m30s.
• In 2021, only about 2% of clinicians in Canada used artificial intelligence or machine learning in their practice, and it was not yet ubiquitous, with some using it for simple tasks such as image classification and risk stratification 4m30s.
• The desire to work in healthcare technology was inspired by personal experiences, including a mother who is a nurse and a father who is a self-taught programmer, as well as an uncle who worked in artificial intelligence in the 1990s 5m40s.

Personal Motivation and Early Experiences in Healthcare Technology

• The goal of using technology to better the world was exemplified by the development of a system to help a visually impaired grandfather read, which was built by family members and eventually led to over a million downloads 7m10s.
• To accomplish the goal of using technology to help people, it was necessary to go back to school and work alongside clinicians, including a first mentor, Dr. Ronald Keren, who taught valuable lessons 8m30s.

Challenges in Clinician Adoption of Technology

• The opportunity to attend an innovation boot camp at Harvard and MIT led to a focus on building technology that clinicians want to use, which is a challenging task due to factors such as clinicians' limited tech-savviness and dislike of change 10s.
• Clinicians often have a general bad experience with software, which is usually chosen for them, and they spend a significant amount of time working on computers, leading to an administrative burden, as seen in a study where 200,000 US doctors spent 5.8 hours on computers for every 8 hours of patient time 2m6s.
• The administrative burden leads to burnout, which is a major concern, and building technology to help clinicians can make a significant difference in their lives and the lives of their patients 2m6s.

The Impact of AI on Administrative Burden in Healthcare

• The introduction of large language models in November 2022 has significantly changed the landscape, enabling the solution of problems such as administrative burden and documentation, and allowing clinicians to spend more time with patients 5m42s.
• The use of artificial intelligence, such as scribe technology, can help reduce the administrative burden by listening to conversations between doctors and patients, recording and summarizing discussions, and assisting with tasks such as referral letters and forms 8m10s.
• The adoption of artificial intelligence in healthcare is increasing, with a significant uptake in the US and Canada, where 20% of clinicians are now using AI, and an additional 40% are willing to use it 9m40s.

The Shift from Authoring to Reviewing in AI-Driven Workflows

• The shift from authoring to reviewing is occurring with the use of Artificial Intelligence (AI) in various fields, including healthcare, where doctors are no longer authoring everything, but rather reviewing what AI is doing 10s.
• The old model of using software in healthcare was simple, with doctors using it back and forth, but the new model is complex, with AI systems continuously working in tandem with themselves, taking over a lot of the work process 42s.
• An example of this shift is the 10-minute visit, where transcription software continuously records 10 seconds at a time, and the AI system selects what it thinks is most relevant to go forward, removing the rest, and then rearranging and adding words to create a note on behalf of the clinician 2m6s.

AI's Expanding Role in Research and Decision-Making

• AI systems are also doing research on behalf of doctors, looking at the body of research, selecting the most relevant papers, and writing a summary, making decisions and picking what gets added, removed, or left out 2m6s.
• The failure rates of transcription, which is about 5%, and the hallucinations that occur with AI models, are not being accounted for, and this is not just a healthcare issue, but also affects other fields such as law and programming 2m6s.

The Broader Implications of AI Automation

• The automation of tasks is leading to a complete restructuring of what is known as knowledge work, and this is a renaissance that cannot be stopped, with systems becoming more complex and taking on more tasks on behalf of humans 2m6s.
• The environmental impacts, effects on cognition, error rates, and trust issues are all concerns related to the use of AI, and job loss is a significant issue, as systems take on more tasks and become more efficient 2m6s.

The Responsibility and Limitations of AI Use

• Ultimately, AI is here to stay, and it is up to individuals to control what they put into these systems and what they take out of them, and to use them in a way that complements their own abilities and intuition 2m6s.
• The importance of reviewing generated content is emphasized, as the system is responsible for authoring it, not the individual, and it is crucial to review the content before sharing it with others or adding it to a medical record 0s.
• This review process requires additional work and effort from the individual, highlighting that the use of these systems actually involves more human involvement, not less 0s.
• The need for extra work and review is stressed to ensure the accuracy and reliability of the generated content, underscoring the importance of human oversight in the process 0s.