AI Research & Literature Searching

In many contexts (but not all), “AI” refers to large language models (LLMs). These tools use a variety of text analysis methods to generate text based on a body of inputs. 

Because AI tools are so varied, and the term “AI” is applied to a wide range of algorithms and computing methods, it is important to evaluate each tool you use in order to understand how an output has been generated.

There are several ways to classify different types of AI, including classifications based on ability, functionality, and technology.

Here are a few ways to classify tools based on the technologies they use:

• Machine Learning: a core branch of AI that allows systems to learn from data without the need for specific programming each time. By processing the data, ML models can recognize patterns, predict outcomes, and improve their accuracy over time.
• Deep Learning: a specialized subset of machine learning that mimics the structure of the human brain using artificial neural networks. These networks enable machines to recognize complex patterns and make sophisticated decisions.
• Natural Language Processing: a branch of AI dedicated to enabling machines to understand, interpret, and respond to human language. By combining linguistics with machine learning, NLP allows computers to process text and speech, facilitating communication between humans and machines.
  • NLP uses both machine learning and deep learning.
• Computer Vision: AI systems interpret and analyze visual information from the world around them. Using DL techniques, computer vision allows machines to identify objects, recognize faces, and understand spatial relationships.

Source: Types of AI: Explore Key Categories and Uses (Syracuse University)

Each AI tool employs different methods to generate its results. There are a few basic questions you can ask to get started in understanding what an AI tool does:

• What computation and analysis methods does this model use? There are myriad computational and text analysis methods that can contribute to large language model (LLM) outputs.
  • Many AI models are proprietary, so users cannot determine how text outputs are generated. 
  • Some platforms do offer some insight into their methods – whether it be deep learning and natural language processing, or metadata analysis. For example, Open Knowledge Maps provides documentation on their processing pipeline. 
  • It’s good to do some research on these methods before you start using an AI tool so that you can better understand its outputs and how reliable they are.
• What inputs trained this model? Different AI models are trained on different inputs. 
  • Models like OpenAI’s ChatGPT are trained on the open web, meaning their results are influenced by websites with varying levels of reliability. This includes massive amounts of data from sources like WikiPedia, Reddit, and even some books.
  • Other models, sometimes called small language models (SLMs) are trained on a more constrained set of inputs. These tools can reduce computational loads and the narrower inputs can produce more precise outputs. It is still important to ALWAYS fact check language model outputs.
  • Some models combine methods. For example, the Oxford Academic AI Discovery Assistant is trained on Oxford Academic’s works, but with support from ChatGPT models.

Large Language Models (LLMs) sometimes provide citations for papers that don't exist, so you should ALWAYS confirm whether a source is real or not if you use AI to find literature. 

Here's a few strategies to confirm whether a source is real:

• Search Crossref: searching the DOI in Crossref is a quick and easy way to see whether a paper exists. The Digital Object Identifier (DOI) is a persistent, standardized identifier, so it can be a helpful reality check and retrieves a reliable URL. 
• Search Google: you can also search Google for the paper using the title and author(s). This might result in finding some papers that are not exactly the same as what's cited, but similar.

If you find the paper but encounter a paywall using either of these methods, try finding it in the Macalester library catalog or submit an ILL request.