The future of AI in medicine | Conor Judge | TEDxGalway
TLDRConor Judge reflects on his journey from a young actor with a stammer to a medical consultant and lecturer, highlighting the imbalance in healthcare delivery due to excessive data collection. He introduces multimodal AI as a solution, which processes diverse data types to enhance medical decision-making. Judge discusses various AI applications in healthcare, emphasizing the importance of trust, explainability, and clinical trials for safe implementation. He envisions a future where multimodal AI democratizes specialized care, urging a balance between compassion and technology.
Takeaways
- 😀 Conor Judge, a medical consultant and lecturer, shares his experience and insights on the future of AI in medicine.
- 🕵️♂️ The speaker compares his past role as a detective in a play to his current role as a doctor, both involving data collection and mystery-solving.
- 📊 The imbalance in healthcare delivery, with 70% of time spent on data collection and only 30% on decision-making and patient communication, is a global issue.
- 💻 Technology, particularly electronic health records, has increased the administrative workload for doctors, reducing the time they can spend with patients.
- 🤖 The potential of responsible AI use in healthcare is introduced as a solution to some of the current challenges in the medical field.
- 🧠 Multimodal AI is defined as AI that processes various types of data, similar to how a doctor uses different senses to assess a patient.
- 🏥 Examples of single-model AI in healthcare include systems for analyzing chest X-rays, diagnosing eye diseases, and predicting conditions like Parkinson's from retinal images.
- 📚 Large language models, like Med-PaM, have demonstrated the ability to answer medical questions and even pass medical licensing exams.
- 🔗 Multimodal AI, such as Med-PaM M, can take multiple types of inputs and perform various medical tasks, improving healthcare efficiency.
- 🔒 For multimodal AI to be implemented safely, trust, explainability, and randomized clinical trials are essential.
- 🤝 The importance of maintaining human involvement in healthcare, ensuring AI is used in conjunction with trained professionals, is emphasized.
- 🌐 The future vision includes multimodal AI making healthcare more efficient, personalized, and accessible, especially in remote and low-income areas.
Q & A
What was Conor Judge's role when he first stood on the stage at the town hall theater 26 years ago?
-Conor Judge was a 12-year-old boy participating in a drama competition for schools, playing the role of a detective in a play written by his best friend, trying to solve a mystery about a fictional hotel called Hotel El chipo, Nono.
How does Conor Judge describe his current professional situation after 26 years?
-Conor Judge is now working as a medical consultant in a hospital for half of his time and as a senior lecturer in applied clinical data analytics at a university for the other half.
What is the '70-30' imbalance in healthcare that Conor Judge refers to?
-The '70-30' imbalance refers to the time doctors spend on collecting information about patients (70%) versus the time spent making decisions and communicating with patients (30%), which has been exacerbated by the introduction of electronic health records.
What is multimodal AI, and how does it differ from single model AI?
-Multimodal AI is an artificial intelligence that takes in data in many different forms, such as text, images, and numbers, similar to how humans process information. Single model AI, on the other hand, processes only one type of data, like images or text, and is less versatile.
Can you provide an example of single model AI in healthcare mentioned by Conor Judge?
-One example is Chest Link by OxyPit, a medical AI triage system that can autonomously report on chest X-rays by looking for 75 abnormalities and is the first system to receive regulatory approval for such use.
How does the AI model developed by researchers at University College London contribute to eye health?
-The AI model is trained on 1.6 million pictures of the retina and can diagnose eye diseases, predict outcomes of conditions like macular degeneration, and even predict Parkinson's disease years before symptoms develop.
What is Med-PaM, and what milestone did it achieve?
-Med-PaM is a medical large language model released by Google. It is the first AI model to pass a US medical licensing exam, initially scoring 67% and later improving to an expert level score of 86% with Med-PaM 2.
What is the significance of multimodal AI in the context of Conor Judge's talk?
-Multimodal AI has the potential to improve healthcare by taking in various types of data and performing multiple medical tasks, making it a promising tool for more efficient, personalized, and accessible health services.
What are the three key elements needed to implement multimodal AI safely in healthcare according to Conor Judge?
-The three key elements are trust, explainability, and randomized clinical trials. Trust is essential to overcome anxiety about AI reliance, explainability helps understand AI decisions, and randomized clinical trials ensure AI models are tested and validated like medicines.
How does Conor Judge envision the future integration of AI and human interaction in healthcare?
-He envisions a world where AI, particularly multimodal AI, is used to enhance healthcare efficiency and personalization, allowing doctors more time to spend with patients, fostering better understanding and improving health outcomes.
What is the 'eyeball test' mentioned by Conor Judge, and why is it significant?
-The 'eyeball test' refers to the instinctive assessment of a patient's condition by simply looking at them, which has been shown to be more accurate than some sophisticated models. It signifies the importance of human intuition and context in healthcare, which should be considered alongside AI insights.
Outlines
🎭 Reflecting on a Journey from Stage to Healthcare
The speaker begins by reminiscing about a drama competition at the town hall theater 26 years ago, where they played a detective with a stammer in a play about a fictional hotel robbery. Fast forward to the present, they now work as a medical consultant and a senior lecturer, still 'solving mysteries' but in a healthcare context. They highlight the imbalance in healthcare delivery, where doctors spend 70% of their time on data collection and only 30% on decision-making and patient communication. The speaker introduces the idea of using multimodal AI to address these challenges, contrasting it with the single-model AIs that are more commonly known.
🤖 Exploring the Potential of Single-Model AI in Healthcare
The speaker discusses three cutting-edge examples of single-model AI in healthcare. The first is 'Chest Link' by OxyPit, an autonomous AI system for triaging chest X-rays, identifying abnormalities and assisting radiologists. The second is an AI model developed at University College London that can diagnose eye diseases and even predict Parkinson's disease from retinal images. Lastly, they mention 'Med-PaM', a medical large language model by Google that has passed a US medical licensing exam, showcasing the capabilities of AI in answering medical questions.
🔮 Embracing Multimodal AI for a Future of Personalized Healthcare
The speaker advocates for the implementation of multimodal AI in healthcare, emphasizing the need for trust, explainability, and randomized clinical trials. They discuss the importance of patient trust and the public's anxiety regarding AI in healthcare. Explainable AI is highlighted as a way to understand AI decisions, and the necessity of clinical trials for AI models is compared to testing medicines. The speaker envisions a future where multimodal AI, including patient images or videos, can make healthcare more efficient and personalized, especially in remote areas with limited access to specialized care. They conclude by stressing the importance of compassion and the human-AI relationship in enhancing patient care.
Mindmap
Keywords
💡AI in medicine
💡Multimodal AI
💡Medical triage
💡Machine learning
💡Natural language processing (NLP)
💡Task sharing
💡ECG
💡Explainability
💡Randomized clinical trials
💡Compassion
💡Healthcare efficiency
Highlights
Conor Judge's return to the stage after 26 years, reflecting on his past as a young actor with a stammer.
The comparison between the time spent on data collection versus decision-making in healthcare.
The imbalance in healthcare delivery exacerbated by the introduction of electronic health records.
The potential of multimodal AI in medicine to address current challenges in data handling and patient care.
Definition and explanation of multimodal AI, contrasting it with single-model AI.
The introduction of OxyPIT's chest X-ray triage system and its regulatory approval for autonomous use.
The development of an AI model by University College London for diagnosing eye diseases and predicting outcomes.
AI's ability to predict Parkinson's disease from retinal images, years before symptoms appear.
The importance of using AI in conjunction with healthcare professionals for patient care.
Google's Med-PaM, a medical large language model that passed the US medical licensing exam.
The release of OpenAI's multimodal version of Chat-GPT and its capabilities in analyzing medical images and scenarios.
Google's Med-PaM M, a multimodal AI that takes various inputs for performing medical tasks.
The need for trust, explainability, and randomized clinical trials in implementing multimodal AI safely.
Survey results indicating public anxiety about healthcare workers relying on AI for treatment.
The concept of explainable AI and its importance in understanding AI decisions in medicine.
The necessity of randomized clinical trials for AI models to ensure their effectiveness and safety.
The role of the 'eyeball test' in patient assessment and its accuracy compared to AI models.
The vision of integrating patient images or videos into multimodal models for more personalized healthcare.
The potential of multimodal medical AI to make healthcare more efficient, personalized, and accessible globally.
The emphasis on prioritizing compassion and understanding in the integration of AI with human healthcare professionals.