A potentially significant limitation of AI is the bias that can be embedded in the products it generates. Large language models (LLMs) are fed immense amounts of data and text available on the internet and trained to simply predict the most likely sequence of words in response to a given prompt. Therefore, the LLM will reflect and perpetuate the biases inherent in the inputted internet information. An additional source of bias lies in the fact that some generative AI (GAI) tools utilize reinforcement learning with human feedback (RLHF), with the caveat that the human testers used to provide this feedback are themselves non-neutral. Accordingly, GAI tools, like ChatGPT, are documented to have provided output that is socio-politically biased, occasionally even containing sexist, racist, or otherwise offensive information.

Selected Readings 

• De Vynck, G. (2023, August 16). ChatGPT leans liberal, research shows. The Washington Post.
• Drahl, C. (2023, October 6). AI was asked to create images of Black African docs treating white kids. How'd it go?. NPR.
• Metz, C. (2021, March 15). Who is making sure the A.I. machines aren’t racist? New York Times [Digital Edition].

Generative AI (GAI) tools have introduced new challenges in academic integrity, particularly related to plagiarism.

Plagiarism is typically defined as presenting someone else's work or ideas as one's own. While a generative AI tool might not qualify as a "someone," using text generated from an AI tool without citing is still considered plagiarism. University at Buffalo instructors have the academic freedom to determine what tools students can and cannot use in pursuit of meeting course learning objectives. See Artificial Intelligence Guidance from the UB's Office of Academic Integrity. Individual policies for using and crediting GAI tools may vary from class to class. GAI tools, such as ChatGPT, have been known to generate false citations, and even if the citations represent actual papers, the cited content in ChatGPT might still be inaccurate.

Related Recommendations

• If GAI tools are permitted to be used for topic development, in the early stages of research, you might not need to cite them at all, but it's still important to check with your instructor first.
• If you are providing commentary or analysis on the text generated by a chatbot and are either paraphrasing its results or quoting it directly, a citation is always required. You can find more information on citing GAI tools on this guide's Citing Generative AI page.
• If you are a researcher planning to publish in a journal, it is best to review that journal's policies on the permitted use of GAI tools. (See 'Selected Readings' below for a couple of examples of journal policies.)
• It's important to always look up citations and check to make sure they are accurate. If you're citing information from a GAI source, try to cite the original source, rather than the GAI.

Selected Readings

• "Artificial intelligence: Editorial policies." (2023). Nature. 
• Hoover, A. (2023, August 17). Use of AI Is Seeping Into Academic Journals—and It’s Proving Difficult to Detect. Wired.
• Liang, W., Yuksekgonul, M., Mao, Y., Wu, E., & Zou, J. (2023). GPT detectors are biased against non-native English writers. Patterns.
• "Science journals: Editorial policies - authorship." (2023) Science.
• Staiman, A. (2023, September 14). Publishers, Don’t Use AI Detection Tools! The Scholarly Kitchen.

There are currently also multiple privacy concerns associated with the use of generative AI (GAI) tools. The most prominent issues revolve around the possibility of a breach of personal/sensitive data and re-identification. More specifically, most AI-powered language models, including ChatGPT, require users to input large amounts of data to be trained and generate new information products effectively. This translates into personal or sensitive user-submitted data becoming an integral part of the collection of material used to further train the AI without the explicit consent of the user. Moreover, certain GAI policies even permit AI developers to profit off of this personal/sensitive information by selling it to third parties. Even in cases when clear identifying personal information is not entered by AI user, the utilization of the system carries a risk of re-identification as the submitted dataset may contain patterns allowing for the generated information to be linked back to the individual or entity.  

Related Recommendations

• Avoid sharing any personal or sensitive information via the AI-powered tools. 
• Always review the privacy policy of the generative AI tools before utilizing them. Be cautious about policies that permit for the inputted data to be freely distributed to third-party vendors and/or other users. 

Selected Readings

• Gupta, M., Akiri, C., Aryal, K., Parker, E., & Praharaj, L. (2023). From ChatGPT to ThreatGPT: Impact of generative AI in cybersecurity and privacy. IEEE Access, 11, 80218–80245. https://doi.org/10.1109/ACCESS.2023.3300381
• Hunter, T. (2023, April 28). Why you shouldn’t tell ChatGPT your secrets. The Washington Post.
• Schneier, B., & Sanders, N. (2023, July 20). Can you trust AI? Here’s why you shouldn’t. The Conversation U.S.
• Shafiq, M  (2023, September 25). Understanding the ethics of data generative AI: Privacy concerns and solutions. LinkedIn.

When discussing the ethics of AI, it is important to consider the impact on the environment and human labor. In their article for Ars Technica, Sasha Luccioni uses the following graphic to visualize the numerous costs of generative AI (GAI).

Selected Readings

Luccioni, S. (2023, April 12). The mounting human and environmental costs of Generative AI. Ars Technica. https://arstechnica.com/gadgets/2023/04/generative-ai-is-cool-but-lets-not-forget-its-human-and-environmental-costs/