Artificial Intelligence (AI) and Scholarly Communications

What is Artificial Intelligence?

Artificial Intelligence (AI) describes a class of computer programs that can approximate human reasoning, performing tasks like creating text or images, analyzing data, and organizing information. It is important to understand that while AI applications appear to have the ability to reason, create, and problem-solve like a person, these programs are actually trained to recognize patterns in large amounts of training data and to report on those patterns or, in the case of generative AI tools, to craft text, images, or other media based on the materials they have been trained on. The most common example of this is what is known as a Large Language Model (LLM), which is an AI system trained on a vast textual dataset to recognize patterns in and mimic human language, such as summarizing content, answering questions, translating from one language to another, or crafting new texts. 

In the context of scholarly communication, this definition of Artificial Intelligence is meant to highlight several things: 

• AI tools are based on pattern recognition and predictive “reasoning” but are thus  limited to the parameters of their training data and construction. It is important to remember the limitations and fallibility of these tools, especially when working in academic research and publishing; 

• All AI applications are necessarily reliant on vast sets of data derived from previously published, human-authored content in order to be trained and to operate usefully, and that this opens a variety of legal and ethical questions with regard to the use of copyrighted materials

• Understanding what data that AI tools have been trained on affects our understanding of these tools’ individual limitations, biases, and value to us as researchers. 

AI tools may provide a range of valuable aids in research and publishing, including search and analysis of vast quantities of previously published material and the generation of visualizations and illustrations, translations, or accessibility components previously omitted due to limitations in expertise or cost, and fostering creativity. AI tools may also enable scholars and publishers to experiment with new formats for digital scholarship by providing support with design and coding. This guide is intended to highlight opportunities incorporating these tools into published research in a thoughtful and informed way.

What should I know about AI Tools?

"AI" is a term that is applied to a wide variety of tools, and it is important to understand some of the categories of artificial intelligence-driven applications as a basis for understanding how they might shape your own work and/or be received by your community of practice.

• Generative AI refers to tools that create new, unique content (texts, video, images, code) using predictive algorithms. These tools, many of which fit into the category of Large Language Models (LLMs), are trained on large datasets to recognize patterns that allow them to generate new content based on common and similar aspects of the works they have been powered by. A generative AI tool might be able to write a poem based on the format, type of subject matter, length, and other features that it have been identified to it as poetry within its training data. 


• AI-assisted search engines or “agents” can retrieve, summarize, and extract parts of records in response to a search query. These tools can be used to search through very large sets of information, such as conducting an initial literature review within a large database, or to summarize the contents of a long article. Many of these tools also provide the ability to use natural language queries - to ask a question narratively, as if you were talking to another person, rather than using advanced search features or boolean operators, which are traditional methods for searching academic databases and indexes.
  
  A known issue of these tools is the possibility that they will present or create factually incorrect results or "hallucinations," since they overlap the same predictive algorithms that power LLMs and other generative tools. 


• Automation refers to rules based software, and includes tools like spelling and grammar checkers.