What does AI have to do with citation equality? One of the biggest problems with some AI tools, Large Language Models (LLMs) in particular, is that they don't tell us what documents were used to train the technology. If the corpus of documents contained mostly white male voices, the tool may produce racist material and present it as fact. Many tools, ChatGPT included, don't cite their sources, either, which can make validating information even more difficult. There's no real way to find a balanced voice.
The resources gathered below explore this problem more deeply. Unfortunately, because both citation justice and AI tools for academia are so new, there hasn't been a lot of research conducted. But some of the resources will help you start thinking about some of the implications.
What impact do LLMs and other AI tools have on Citation Justices? Let's start with a couple of questions to help you think about these impacts.
Equity in AI is hydra of a problem. Multiple heads of bias tend to get introduced to AI tools, from the writing of the code, to the training documents, to the prompts you create. Bias can be hard to identify and eliminate even when you can identify. The resources listed below help you start thinking about bias in AI, and in LLMs in particular.
"Eliminating Algorithmic Bias Is Just the Beginning of Equitable AI", Harvard Business Review
This short article is a well rounded article to the various ways that bias can be introduced to AI. As you read through it, think about where citation comes in. How can the documents chosen to train LLMs, the questions we ask, and the lack of citations increase bias?
Eliminating Algorithmic Bias Is just the Beginning of Equitable AI
"How AI Image Generators Make Bias Worse", The London Interdisciplinary School
The video below uses AI image generators as an examples of how bias is introduced into AI and how difficult it is to combat. The video should start you at 2:24, but feel free to rewind and watch from the beginning. As you watch, replace image generators with large language models. Are the bias introductions similar? Were you surprised by any of the issues discussed? What do you consider a fair representative sample?
There are few (if any) tools that beat reading, summarizing, writing, and citing yourself. But there are some tools that can help make the process a little easier to evaluate for equity. Check out the tools below. Do you think they are easier to use for citation justice? How could they be improved?
"The Assumptions You Bring into Conversation with an AI Bot Influence What It Says", Scientific American
The prompts we write for AI tools often carry implicit bias with them. Implicit biases are the negative feelings or perceptions we harbor unconsciously against a group of people. When our prompts are biased, this can often illicit a more biased response. This Scientific American article describes this process. Can you think of ways to mitigate implicit bias in prompt construction?
The Assumptions You Bring into Conversation with an AI Bot Influence What It Says
