Chair(s)

Introduction

Generative artificial intelligence (AI) chatbots are computer-based programmes based on deep neural networks that can simulate human-human conversation via text or voice to engage in a dialogue with users. With ChatGPT, there has been an explosion of new large language models (LLMs) on the scene. Each LLM is distinct and performance can vary. However, all LLMs suffer the same limitations, as they are not trained to be factual but to sound credible, posing concerns particularly in high-stakes environments like health care. “Hallucinations” are a known weakness of generative AI chatbots, leading to nonsensical outputs. Lack of explainability and trustworthiness are major concerns for LLMs.   

ChatGPT lacks the capacity for self-correction, leading to a lack of accountability. Inaccurate or misleading responses of models can perpetuate misinformation and hinder vital learning and improvement processes for AI systems. 

Deep learning systems have billions of parameters, enabling them to learn complex, non-linear relationships about the world. It also means that we cannot intuitively understand how and what they are learning. They are the ultimate “black box” – once unleashed, their outputs are difficult to analyse, even by their developers. This makes it difficult to trust these models to behave consistently and safely. The question we must seek to answer is how we can leverage the usefulness of AI while guarding against its flaws and misinformation. ChatGPT and other LLMs are already at large and have seen unprecedented adoption over the past year. Some businesses that began by “experimenting” with ChatGPT now use it, often unchecked, to save on labour costs. Yet, it remains a technology without regulatory safeguards, highlighting the need for human oversight.

How is this technology likely to impact the world of health care and pharmacy, and whether it represents an opportunity or a threat, is a question worth exploring.

Programme

Learning objectives

Take home messages