It can be challenging to identify material generated by AI tools. AI detection tools are often inaccurate, and generated images and videos are increasingly realistic and difficult to discern.  

Evaluating content using the strategies below, keeping in mind the tendencies of generative AI tools to hallucinate and perpetuate bias, will help to ensure that the content you are seeing is reliable and accurate. 

Hallucination

Hallucination is the term given to the tendency for generative AI tools to make up answers that are false and incorrect. This is because these tools are pattern-recognizing machines; they provide output that is the most likely response to your input. Like a weather forecast, these tools make predictions that are not always accurate. There is no guarantee of factuality and no commitment to truth. 

Bias

Biases are another important consideration to think about when using generative AI tools. These tools create biased information for several reasons: 

• People may embed their biases when they create them (e.g. by privileging certain algorithmic patterns over others). 

• There can be biases in the datasets used to train them. 

• Generative AI may create its own biases from how it interprets the data it has been trained on. 

This section on Bias has been adopted and adapted from the University of Alberta Library’s Ethical Considerations page in their Using Generative AI guide. This material licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.  

The SIFT Method was developed by Mike Caufield from the University of Washington to assess information and can be applied to information generated by AI tools. 

Stop 

• Take a moment to consider the information or output provided by the AI tool in more depth. 

Investigate the source 

• Investigate the AI tool you are using with the ROBOT test to identify any possible weaknesses, shortcomings, or bias in the tool. 

• Check the currency of the data that the AI tool is trained on.

Find better coverage 

• Locate additional resources on the same topic as the information generated by the AI tool and compare what non-AI-generated scholarly material states to the AI-generated content.

Trace claims, quotes, and media to the original context 

• Ask generative AI tools to list their sources when generating content, and confirm that: 

• Those sources are real 

• The information in those sources match the generative AI tool’s output claims

SIFT text and graphics adapted from “SIFT (The Four Moves)” by Mike Caulfield, licensed under CC BY 4.0.