Kelvin Smith Library
It’s important to recognize that AI tools and the information systems and Large Language Models (LLMs) they’re built on can bring up different ethical considerations. As outlined in the AI literacy framework, considering the safety and privacy measures, identifying ethical issues, and recognizing errors, inconsistencies, and biases in AI-generated content is an essential part of using these tools effectively.
Hallucinations are falsehoods, logical inconsistencies, or incoherent statements that can arise in the process of AI response generation. They are, at present, an unavoidable phenomenon of generative AI tools. Understanding that hallucinations can occur, AI users should check any sources, citations, or other information (for example, the name of an expert in a particular field) before including those citations in research projects.
AI is built on human-generated content, which inherently contains explicit and implicit biases. When these biases are baked into the information systems underpinning generative AI tools, they can introduce bias in AI outputs. Recognizing that bias in technology is real is a crucial step in evaluating AI outputs.
As with any online platform, tool, or application, be discerning about what types of information you provide or input. In addition to data leaks or breaches, some proprietary Generative AI tools use the information you provide to help train the tool. CWRU's The Daily has some additional tips:
Artificial intelligence tools collect and store data from users as part of their learning process. Any information entered into an AI tool becomes part of its training data, which may then be shared with others.
Among the types of data that should not be used in AI tools are:
Please visit our AI Citation and Academic Integrity Page for more information.
Please visit our Copyright Page for more information on ethical considerations for copyright and AI.
Like many other types of software, LLMs have both proprietary and open source tools for users to choose from.
Proprietary tools like Copilot, ChatGPT, and Claude offer ease of use and a low barrier to entry. Subscriptions provide instant access to working models and can include extensive support. Proprietary tools could be a good fit for simple work, or projects that cannot afford significant IT resources necessary to run an open source LLM locally.
Open source tools can be much more labor intensive to get working and to maintain, but that effort comes with several significant benefits. First, the model’s workings will be completely transparent. The materials it is trained on, the code written to interact with it, the nature of errors and other aspects of the LLM will all be available for you to see. This is not the case with proprietary tools that keep these elements hidden to protect their business model.
Second, open source tools allow you to conduct research that can then be repeated and verified by others. Using open source tools allows you to control the versions and settings of your LLM. You can share those details with others who can then try to repeat your research. Using proprietary tools for research means your research will be impossible to verify, as model versions and settings will change, frequently without any notice, and older versions will no longer available for use.
You can contact the Digital Scholarship Department of KSL at freedmancenter@case.edu to discuss which kind of LLM would best meet your research.