Also: This open textbook (or wiki channel) is intended to help librarians and other information professionals learn about AI. It is not, in itself, meant to be seen as promotion of AI. If anything, the goal is harms mitigation or harms reduction.
Introduction
Machine learning (ML) is a subset of artificial intelligence (AI) that enables computers to learn from data and improve their performance on specific tasks without being explicitly programmed for each task. ML uses algorithms trained on data from one or more sources to identify patterns, generate predictions, classify information, and adapt to new data.
ML supports a wide range of applications, including image classification, natural language processing, data analysis, recommendation systems, and outcome prediction, tasks that may involve capabilities traditionally associated with human intelligence. At its core, ML involves developing computational models that learn relationships within data and use those learned patterns to make predictions or decisions. Performance can improve as models are trained on additional or better-quality data, although the quality of the results depends on factors such as the training data, model design, and evaluation methods.
Types
Machine learning encompasses several major approaches (or, types), including supervised learning, unsupervised learning, semi-supervised learning, and reinforcement learning. large language models (LLMs) such as GPT and Bidirectional Encoder Representations from Transformers (BERT) are built using machine learning techniques, particularly deep learning. DL is a branch of ML that uses artificial neural networks with multiple layers ("deep" networks) to learn patterns from large amounts of data. Deep research is a process of conducting extensive research, sometimes assisted by AI systems built using deep learning.
"...Most medical ML models rest on fragile evidence from noisy labels, poor thresholds, unstable metrics, and weak validation; evidence-based medical AI demands rigorous ground truthing, calibration, uncertainty reporting, and external validation; clinical utility assessment and post-deployment monitoring are essential but largely neglected in current ML pipelines; a call to researchers, reviewers, users, and vendors to debate and raise methodological standards in medical AI."
Many (if not all) of the AI-powered search tools such as Elicit.com use retrieval augmented generation (RAG) / deep research techniques to deliver results. RAG refers to a technique combining the strengths of retrieval-based and generative AI models. In RAG, an AI system first retrieves information from a large dataset or knowledge base and then uses this retrieved data to generate a response or output. Essentially, the RAG model augments the generation process with additional context or information pulled from relevant sources.
Presentation
Note: This video is meant to be for informational purposes only. Any claims of the video should be tested for accuracy and verified.
Biased outputs in machine learning
There is substantial evidence that machine learning models and AI tools can exhibit bias.
Amazon developed an AI system to screen job applicants’ résumés, but the tool was found to be gender biased against women (Dastin, 2022). Because the model was trained on historical hiring data from a predominantly male industry, it learned to penalize résumés containing terms such as “women.” In another case, an Amazon facial recognition system incorrectly matched 28 members of the U.S. Congress with individuals who had been arrested for crimes. Similarly, Google’s natural language processing models have been shown to label sentences referring to religious and ethnic minorities as “negative,” reflecting biases embedded in sentiment analysis training data.
Comparable biases have been identified in neural networks designed to recognize skin lesions when training datasets included only 5–10 percent images of Black skin. Barros et al. (2023) reported that model accuracy dropped by nearly half when evaluated on images of Black skin, increasing the risk of misdiagnosis and adverse health outcomes for Black patients. Given that Black patients have an estimated five-year skin cancer survival rate of approximately 70 percent, compared with 94 percent for white patients, the consequences of deploying such biased algorithms in clinical settings can be substantial.
Librarian view
"...The idea that we should outsource academic authorship to LLMs rests on the assumption that writing is (only) a mechanical, predictable or reductive process which, with the right prompts, can be replicated with ease." — Masters, 2025.
Bottom line: For health sciences librarians, AI tools may offer useful support when working with health professionals; however, many of the underlying processes raise serious concerns for those committed to scientific accuracy, transparency, and methodological rigour in evidence reviews. Information about AI tools is evolving rapidly, so readers should verify details using current sources or consult a librarian.
Note: librarians are trained to distinguish between searching for sources and searching for answers. Many AI systems prioritize the latter while obscuring the former, and transparency remains a significant limitation. Transparency is a critical part of doing knowledge synthesis (KS).
Note: Please use your critical reading skills while reading entries. No warranties, implied or actual, are granted for any health or medical search or AI information obtained while using these pages. Check with your librarian for more contextual, accurate information.
