Gen-AI

Evaluating Gen-AI output

Reviewing outputs produced by Gen-AI involves critical reading and thinking skills. Additionally, Gen-AI tools and technology may not provide you with a comprehensive range of information, as AI’s ability to access scholarly information is currently limited by paywalls.

Personal value

Gen-AI outputs are based on probability and the technology cannot think critically. This means it cannot verify its own outputs, even if it includes links, names and sources.

Therefore, while Gen-AI can produce convincing, professional-sounding content, you are still considered the expert.

If you have been permitted to use Gen-AI as part of your assessments, you should view Gen-AI results with this thought in mind. AI output should be modified to fit and reflect your work and ideas rather than changing your work to fit the AI’s ideas, style and work.

Whether you are using AI in your day-to-day life, for study, or for assessment (when permitted), you need to fact-check any important information to ensure it’s correct. Gen-AI technology is not perfect; it reflects the bias and prejudice of the data it was trained on and could contain misinformation. It is vital that you critically assess both the output of Gen-AI-created content.

Checking outputs

Questions to ask yourself:

• Can I find the original information source and verify the information provided? AI-generated references can be false or made-up (known as hallucinations), so always double-check that the information is real and accurate.
• Can the information presented be supported by other sources found elsewhere online? Do other reliable sources state the same thing?
• Is the information consistent throughout the text result? Gen-AI outputs are built from a combination of different sources, so it’s important to check for inconsistencies or contradictions.
• Are there any obvious gaps or biases?
• How up-to-date is the information provided? Some text-based AI generators work from information available up to a certain date.
• When working with data, does the result equal the expected number? Have all pieces of information been included and/or categorised correctly?

For more information on reviewing information, visit the critical thinking guide.

Tool investigation

Currently, most Gen-AI technology cannot access live or up-to-date information, nor can it search the internet the same way we can. Its knowledge is limited to what is present in its training data, and the responses it provides are bound by the rules put in by the company that owns it.

Further, most tools have not been trained using information that is behind paywalls, subscription-only, or in academic journals requiring special permission or user log-ins. This means that the information used to train Gen-AI models may not be peer-reviewed, scholarly, or include all relevant, recent academic research.

To discover the limitations of an AI tool, you can query LLMs to determine their last updates, their ability to source information, and the information in their training data set.

Simple questions to ask:

• What data were you trained on?
• When is the cut-off date for your knowledge?
• Are you able to access the internet or live news?
• What bias could be present in the answer you have provided above?