Hallucinations
Generative AI and Google do not work the same way. Google uses your prompt to search for pre-existing information that includes the words you used in your prompt. Generative AI takes your prompt, analyzes it, pulls information from a bunch of different sources about your topic, and synthesizes that information together in order to give you a new response, not found elsewhere on the internet. While this is a really cool function, it does mean that AI has a tendency to get stuff wrong when it synthesizes information together. In fact, current iterations of generative AI will rarely, if ever, tell you it doesn’t know the anwer to something. It will make something up in order to answer your question, even if the answer is wrong.
When AI gets something wrong by providing misinformation or fake information, it is called a hallucination.
Bias
Since AI is trained on data produced by humans, the same biases that appear in human work appear in AI generated work as well. Time and time again, studies have shown that some AI tools provide more left leaning information while other provide right leaning responses. Also, common stereotypes of people groups are often re-inforced by AI responses. Just like with any news article, AI must be evaluated for these biases.
Also, keep in my that AI wants to make the user happy and answer the users prompt no matter what, so if the prompt you input into AI is biased, you will only get biased information from AI. For example, look at the questions below. Notice the different way in which they are worded will prompt AI to produce biased information:
- Negative Bias: Why is the Vietnam War the worst foreign war the U.S. has ever gotten involved in?
- Positive Bias: Why is the Vietnam War a less controversial war than the War on Terror?
- Neutral: What are the reasons why U.S. citizens were both for and against involvement in the Vietnam War?
Additional settings for Safari Browser.
