A Dermatologist’s Beginner’s Guide: Using Ambient Artificial Intelligence in Your Practice

Abstract

Automatic Speech Recognition (ASR) is the fundamental technology that enables the conversion of spoken language into written text. The strengths of ASR include interpreting voices, identifying different speakers in a conversation, and following dialogue. With the support of human trainers, ASR systems can further improve and optimize their performance based on feedback. However, their effectiveness can be limited by factors such as noisy environments, poor microphone placement, and variations in dialects and accents. See Exhibit 1 for definitions of key AI terms.

Large Language Models (LLMs) aim to bridge these gaps by using Natural Language Processing (NLP) to fill in missing elements of a conversation. As a type of generative AI, an LLM predicts the next word in a sequence based on patterns found in its training dataset (Figure 1). The goal of an LLM is to mimic human language by identifying which words are likely to follow one another, given the context provided in the prompt and its prior training. LLMs can generate original content and optimize their outputs based on human input. However, because they function by predicting word sequences, they do not inherently understand whether their outputs are true or false. As a result, LLMs can produce inaccurate or fabricated responses, commonly referred to as “hallucinations” in their responses to prompts.

Real-time ambient AI scribes, which leverage machine learning to process conversations, show promising potential to reduce the documentation burden, enhance the quality of doctor–patient interactions, and support clinicians in their daily work. When combined, LLMs and ASRs can offset each other’s limitations, resulting in an ambient AI scribe that is more effective and practical for use in clinical settings (Figure 2).