We have created this guide to define generative AI tools and help you navigate some of the concerns and opportunities with using these tools. While there are numerous generative AI tools available with more being developed regularly, we will focus primarily on ChatGPT in our discussions and examples.
If you have any questions at any point, it's a good idea to contact your instructor directly or ask a librarian.
The UK ADVANCE team provides recommendations and guidelines on the use of AI for the UK community. See the most recent updates on their website:
Special thanks to Makayla Wells (MSLS, '24) for her work in initially creating this guide.
What is Generative AI? According to IBM, "Generative AI refers to deep-learning models that can generate high-quality text, images, and other content based on the data they were trained on." These models can be used for a wide variety of creative or academic purposes, from generating poems to proposing new molecules.
What is ChatGPT? ChatGPT (GPT stands for Generative Pre-trained Transformer) is an advanced Natural Language Processing (NLP) tool that uses machine learning and a Large Language Model (LLM), that is, a large training data set of text examples, to generate human-like text responses to prompts given by a user. ChatGPT was introduced in November of 2022 and has continued to develop since then with multiple iterations. Please note that the subscription version is more effective for most purposes than the free version that is available on OpenAI's website. Other AI tools have been introduced to generate text, images, audio, and videos, which have improved in quality in recent months and years.
How does ChatGPT work? In part, it uses text prediction. Based on training with a very large data set of written text, ChatGPT responds with 'likely' words that are appropriate for the prompt given by a user, which resemble human language. During training, the system goes through a couple stages of development. First, there is an unsupervised pre-training stage in which the training data being given to the model is unlabeled, and learning occurs ‘naturally’. After this pre-training, algorithms are refined further by the AI’s creators in a supervised process to better exhibit natural language in outputs. Please see this article from OpenAI for a more thorough description of the training process.
Given how this AI tool works, users should be cautioned that ChatGPT is not considering all relevant sources to produce the best academic answer to an inquiry, but rather is generating appropriate text based on its training data set and the influence of AI engineers. Users may find outputs to sound authoritative due to the advanced natural language reproduction and an assumption that a chat bot would be unbiased, but outputs are only as good as the training data and do not rely on a human-like deliberative process.
Other Popular Generative AI Tools
Users may get the impression that generative AI tools are searching for Internet data in real time, but this may not be the case. When ChatGPT was originally introduced, its training data only went up to the year 2021. However, Microsoft Bing now incorporates ChatGPT into the search engine tool Copilot, allowing users to find information and links from current websites. Users should always question what training data is used for a generative AI tool and from what time period.
Additionally, users should be aware that using a generative AI tool connected to the Internet likely means that prompts and user data are being collected for training purposes. Users should avoid sharing sensitive information or anything that they do not want to be collected by the company. They may consult a company's privacy policy for more information. ChatGPT provides some information on data collection here.
Companies are increasingly incorporating generative AI tools into their business models. According to McKinsey & Company, a multinational management consulting firm, "McKinsey research indicates that gen AI applications stand to add up to $4.4 trillion to the global economy—annually" (April 2, 2024). Developing generative AI is very resource intensive. Only the largest and well-funded companies are able to invest into the development of their own model, but others may use a developed generative AI product like ChatGPT to assist with tasks like written communications. Models like ChatGPT can be fine-tuned by the user to be a better fit for various tasks. Other generative AI products may assist with occupation-specific tasks such as creating higher resolution versions of medical imaging.
Generative AI goes far beyond the production of text. There are many AI image generator services now available. DALL-E was also developed through OpenAI and is a popular service for generating images. Increasingly, AI programs can produce videos too. Often, there are telltale signs that media is produced by AI (such as too many fingers on a representation of a person), but this technology is rapidly advancing. Students and researchers alike should be aware that AI-produced media is increasingly prevalent on the internet and may not be advertised as such.