Artificial Intelligence (AI) Guide

Potential Issues of Using Generative AI

• Reliability – Natural language processing tools, language models, or chatbots are trained on a collection of writing to generate human like text in response to a prompt, not necessarily to generate truthful or reliable answers. They are also restricted by their training corpus.
• Transparency – In the same way that natural language processing tools, language models, or chatbots are not necessarily generating truthful for reliable answers, they are not specifically created to locate and credit the sources for the responses they generate.
• Bias – Since natural language processing tools, language models, or chatbots are trained on a large body of writing created by humans with human biases, the responses generated could reflect and further enforce those same biases.
• Privacy – Like any online platform, AI tools accept the input of private information from users who may not be aware of how the platform intends to use that information.
• Denial of Individual Autonomy, Recourse, and Rights – There is difficulty in assigning accountability/responsibility to the responsible parties of algorithmically generated outcomes. (Leslie, 2019)
• Unreliable, Unsafe, or Poor-Quality Outcomes – Careless data management, negligent design and production processes, and questionable deployment practices can lead to the application and distribution of AI systems that result in unreliable, unsafe, or poor-quality outcomes. These outcomes can negatively impact the individual’s or public’s wellbeing, undermine public trust in the responsible use of AI, etc. (Leslie, 2019)
• Hallucinations – The generation of outputs may sound plausible but are either factually incorrect or unrelated to the given context. These outputs often emerge from the AI model's inherent biases, lack of real-world understanding, or training data limitations. In other words, the AI system "hallucinates" information that it has not been explicitly trained on, leading to unreliable or misleading responses.
• Equity – Many generative AI tools – both text- and image-based – were trained on large collections of material taken from the internet without prior approval or involvement from the human authors who created the content. These same human authors will be forced to compete for their livelihood with an automated system that can work faster.
  • Many AI tools also perpetuate the dominance of English in online spaces, threatening Indigenous languages.
  • Access to powerful, quality AI tools may require payment and thus divide those who cannot afford to pay for the tools from the advantages enjoyed by those who can pay.

Potential Benefits of Using Generative AI

• Automating Repetitive Daily Work – Multiple companies are exploring how the integration of generative AI tools into office programs and applications can improve workflows and communication.
• Multilingual Assistance – Generative AI tools may help students and professionals who are required to write in English but do not feel comfortable in this language.
• Improving Instruction - Instructors have been utilizing natural language processing tools (e.g., ChatGPT) to brainstorm new assignments, improve assessments, develop differentiated instruction, and engage students.

The Artificial Intelligence (AI) Guide provides an introduction to this evolving field for faculty, fellows, residents, postdocs, students, and staff. Due to the rapid advancement of this emerging technology, information in the Guide may become outdated at times. 

For information on Artificial Intelligence (AI) Data Security and Privacy, see Artificial Intelligence (AI) Data Security and Privacy - Information Resources (utsouthwestern.net), VPN/On Campus access only.  NOTE:  this Guide supplements but does not supersede information provided by UT Southwestern or University of Texas policies and guidelines.