Summary
Highlights
Ashley concludes by recommending several AI tools: Perplexity for complex questions, Duck AI for privacy, Copilot for task-oriented integration, Elicit for in-depth analysis of papers, Claude for brainstorming and data analysis, and various visualization tools like Research Rabbit and Insightful for mapping research. She notes the availability of a handout with more details on these tools.
Carrie Cisk from CTSI introduces the session on responsible AI use, highlighting its relevance to researchers and investigators. She introduces the speakers, Ashley Pervvis and Tara Roger, librarians from WVU, who will delve into the topic.
Ashley Pervvis begins by providing a brief history of AI, from its origins in the 1950s to the development of machine learning, deep learning, and generative AI today. She explains how AI learns through supervised, unsupervised, and reinforced learning, and details the role of neural networks and large language models in its functionality. The distinction between generative and traditional AI is also clarified, with an example of how ChatGPT operates.
Tara Roger discusses the practical applications of AI, particularly in the medical field. She notes its usefulness in automating tasks like appointment reminders and healthcare coding, creating text for screen readers for accessibility, and aiding in medical research through concept mapping, brainstorming, and summarizing scholarly articles. Tools like NoteGPT and Chat EHR are mentioned as examples.
Tara delves into the ethical and environmental concerns surrounding AI. She highlights data privacy issues, the risk of self-plagiarism or early release of work, and issues with publishers using researchers' work to train AI models without consent. Concerns about biased results, copyright, and the significant environmental impact (water and energy consumption) of AI are also addressed.
This section expands on concerns related to publishing and data integrity. Tara discusses the rise of 'paper mills' and data leakage due to AI tools, and the potential for copyright headaches. She warns against 'ghost citations' or AI hallucinating non-existent sources, the non-reproducible nature of AI results, and its potential to perpetuate bias and misinformation. Recommendations include staying updated on AI trends and carefully reviewing author agreements.
Tara provides tips for responsible AI use, emphasizing understanding the risks and limitations of tools, and checking with mentors or publishers before use. Ashley introduces the 'CLEAR' framework (Concise, Logical, Explicit, Adaptive, Reflect) as a guide for effective and responsible AI interaction, particularly for crafting prompts and evaluating outputs.
Ashley discusses principles for AI use in medicine laid out by the AMA, focusing on mitigating bias, ensuring privacy and security (HIPAA compliance), and avoiding liability. She highlights pitfalls such as limited and expensive datasets, inherited bias in algorithms, and the importance of inclusive design and stakeholder engagement when using AI in healthcare. She stresses that AI should not be used for clinical decisions to prevent provider lawsuits.