Skip to Main Content

AI for Research

Links and information for a wide range of generative AI resources.

GAI Hallucinations

Generative AI resources are known to produce convincing yet fake citations. The citations may include names of real researchers/authors and authentic research publications. These are known as AI hallucinations. Generative AI is not built to be truthful; AI tools are built to generate a response from learned data sets. If the user asks for citations, generative AI will create citations based on the prompts given and the LLM datasets used to train the AI. In other words, AI gives you what you asked for without consideration of the veracity of the information.

Ethical Risks

GAI are not free of bias and raise considerable ethical questions. Consider the following when using generative AI:

Copyright and Data Provenance

Who owns the data from which responses were generated? Generative AI tools are making use of massive datasets from multiple sources. Can the output be validated and not infringe on IP and copyright? Is the information being used without consent? Who controls the data being used to train the AI and its output?

Data Privacy and Disclosure of Sensitive Data

GAI are trained on large language models that may include personally identifiable information. The unintentional release of the data can be in violation of data privacy laws. Additionally, as a user take care in not entering personal and private information that may be reviewed by humans and possibly released through searches.

Correlation vs Causality

Correlation suggests a relationship; causality suggests one factor causes another. Generative AI works by grouping facts together based on probability or by correlating the data for a particular outcome. The user needs to consider why the system gave the answer it did and be able to interpret the results.