Generative Artificial Intelligence and Academic Research

Generative artificial intelligence produces output--text, images, sound, video, etc.--based on patterns extrapolated through machine learning from large collections of training data. (For definitions of terms related to AI, see this glossary provided by the Center for Integrative Research in Computing and Learning Sciences.)

A search engine like Google is designed to search within a collection of information to find answers to specific queries. Generative AI tools like ChatGPT are chatbots, not search engines. When prompted, the chatbot uses statistical models built on patterns from its training data to generate a dynamic response that supplies content associated with its predictions about the meaning of the words in the prompt. In a response from ChatGPT, for example, the result is a string of text consisting of the next most likely word based on the previous words.

Because generative AI tools respond so quickly and fluently, it's easy to assume that the information provided is authoritative and accurate. However, it's important to remember that every response is just pattern matching, with no true understanding behind it. The accuracy of the response depends on having enough training data to mathematically calculate an appropriate answer.

Generative artificial intelligence is not actually intelligent. It does not think, just mimics thought. 

Generative AI tools like ChatGPT can be applied in various ways to assist with academic research, as discussed on the "AI Chatbots and Research" page of this guide. However, it's important to keep in mind potential issues.

• False information: If AI is trained on data harvested online, it can learn and spread propaganda and misinformation. Also, generative AI tools sometimes simply make up content included in responses, which is referred to as "hallucinating." When used for academic work, a response may seem plausible and even include citations, but the sources may turn out to be fabricated. So it's always important to fact check the information provided when using generative AI for research and academic writing.
• Bias: Although people may think of AI systems as being objective, their training data and the algorithms they use are created by human beings, who are prone to bias. Generative AI tools can learn and replicate biases, resulting in output that contains stereotypes and other harmful content. 
• Privacy: AI tools may capture and reuse personal data from their users. Also, anything created and uploaded by a user may become part of the training data, depending on the policy of the system.
• Ethical concerns: Many generative AI tools use training data harvested from the internet without the consent of the creators of the content, creating a situation where human creators are forced to compete with systems that have been trained to mimic their own work. The use of generative AI also has significant environmental costs. Training and running AI systems requires huge amounts of electricity and fresh water to cool the servers. This impacts the electrical grid and the availability of water for human consumption, a significant problem in areas already experiencing water scarcity.