Another tip in a series provided by the Offices of Information Security, Information Systems & Computing and Audit, Compliance & Privacy

AI is becoming more integrated into the applications we use in our day-to-day lives. AI can, and does, make things easier for us. However, we need to use AI with a sense of discernment and a bit of skepticism.

AI sometimes has something called “hallucinations,” where the response from the AI outputs things that are not factually correct.

For example, AI has fabricated legal cases that professionals have mistakenly included in legal documents. It has also falsely given several companies incorrect business or financial information. People basing their actions on these incorrect, AI-provided “facts” could cause embarrassment and legal and economic harm.

The AI companies can help mitigate the possibility of hallucinations by refining their training data and using other internal processes to try to prevent hallucinations.

You can mitigate the possibility of AI hallucinations by refining your prompts and manually confirming information after engaging with AI to validate for accuracy.

To do this, when creating your prompts, be clear and specific about the questions you ask so that the AI returns as precise responses as possible. You should also provide context and relevant details while writing your prompts.

As a part of your prompt, you can also ask the AI to provide sources. When prompted, most AI tools cite their sources and provide links, which you can use to check for response accuracy. You should also consider independently verifying AI responses.

University of Pennsylvania AI Guidance: https://isc.upenn.edu/security/AI-guidance.

--

For additional tips, see the One Step Ahead link on the Information Security website: https://isc.upenn.edu/security/news-alerts%23One-Step-Ahead.