Artificial Intelligence, A Student Guide

Generative AI tools like ChatGPT are able to produce a lot of different content, from quick answers to a question to creating cover letters, poems, short stories, outlines, essays, and reports. However, it often contains errors, false claims, or plausible sounding, but completely incorrect or nonsensical answers, so be sure to take the time to verify and check the content created to catch these problems.

Generative AI can also be used to create fake images or videos so well that they are increasingly difficult to detect, so be careful which images and videos you trust, as they may have been created to spread disinformation.

Fact checking is very important when using AI!

Generative AI relies on the information it finds on the internet to create new output. As information is often biased, the newly generated content may contain a similar kind of bias.

Examples of potential bias include gender-bias, racial bias, cultural bias, political bias, religious bias, and so on. Closely scrutinize AI-generated content to check for inherent biases.

AI content may be selective as it depends on the algorithm which it uses to create the responses, and although it accesses a huge amount of information found on the internet, it may not be able to access subscription-based information that is secured behind firewalls. Content may also lack depth, be vague rather than specific, and it may be full of clichés, repetitions, and even contradictions.

AI tools may not always use the most current information in the content they create. In some disciplines, it is crucial to have the most recent and updated information available. Think, for example, about the recent pandemic. Research was going at a very fast pace and it was important to have not only the most comprehensive and most reliable data available, but also the most recent. Technology is another area that is constantly changing, and information that is valid one year, may not be valid the next. There are many other examples, and it is important that you check the publication dates for any sources of information that are used in AI-generated texts.

Note: As of April 2024, ChatGPT's (GPT-3.5) information comes from January 2022 and it does not have the ability to pull current information from the Internet.

Generative AI tools don't always include citations to the sources of information. It is also known to create citations which are incorrect and to simply make up citations to non-existent sources (sometimes referred to as AI Hallucination). It may provide citations by an author that usually writes about your topic, or even identify a relevant well-known journal, but the title, pages numbers, dates, and sometimes authors are completely fictional.

Not crediting sources of information used and creating fake citations are both cases of plagiarism, and breaches of Academic Honesty. Be sure to check OneSearch from our library and/or Google Scholar to verify whether the sources are correct or even exist.

Generative AI tools rely on what they can find in their vast knowledge repository to create new work, and a new work may infringe on copyright if it uses copyrighted work for the new creation.

For example, there have been several lawsuits against tech companies that use images found on the internet to program their AI tools. One such lawsuit by Getty Images which accuses Stable Diffusion of using millions of pictures from Getty's library to train its AI tool. They are claiming damages of US $1.8 trillion.

There is much debate about the ownership of copyright to a product that was created by AI. Is it the person who wrote the code for the AI tool, the person who came up with the prompt, or is it the AI-tool itself?  Currently U.S. copyright law has a "human authorship requirement," however works containing AI-generated material may also contain sufficient human authorship to support a copyright claim.

Recent research has raised the concern that as more text is published that has been created by generative AI, this AI-generated content will enter the training datasets for new generations of AI, which may decrease the quality of the data since errors in early generations of AI compound themselves over time. A study by Shumailov et al. (2024) found that the inclusion of AI-generated content in training datasets led to model collapse, which is "a degenerative process whereby, over time, models forget the true underlying data distribution, even in the absence of a shift in the distribution over time" (p. 2).