One of the key tenets of crisis management is identifying and verifying the facts of the issue at hand before communicating. Use of AI tools is increasingly helpful as we develop strategies and messaging. There are a wide variety of tools available, both free and paid, that can support human centered crisis planning, response and measurement. None are without their pitfalls, though. According to a new book by Karen E. Sutherland, Artificial Intelligence for Strategic Communication, an important and perhaps not well understood issue is the frequent inaccuracy of AI generated information. According to Sutherland and other experts, AI “hallucinates”; in other words, it fabricates information. The fabrications can be very convincing and presented in a well-written style that can be hard to identify as false.
People want to believe that information on the internet, or AI generated content is always objective and accurate. But we know that it is not. The AI cannot differentiate between what is true or not, biased or not. If the datasets used to train the tool have inaccurate information or built-in bias (and they all do), these are amplified in the AI output. In other words, garbage in, garbage out. And we should assume that there is garbage in the output.
We should not, therefore, use AI generated output without fact-checking it. And we also need to use fact-checking methods that do not further amplify inaccurate content. In other words, don’t rely on AI to fact check itself. The AI we are using today is still in its infancy, and it requires oversight and validation by humans who are subject matter experts in the content being developed.
What to do next? Learn. Educate yourself on the strengths and weaknesses of the tools you are using. Try different tools to find the one that best meets your particular needs. Learn how well-written prompts can lessen inaccuracy and bias. Implement a fact-checking protocol to assure that content is clean before deploying it.
