Fact Checking
6 Ways to Avoid Bias and Hallucinations When Writing with AI
Learn how to prevent bias and avoid hallucinations when writing with AI by applying fact-checking, prompt engineering, and best practices.
For all of its advantages and benefits, AI isn’t perfect. It can (and does) often hallucinate facts and figures. Further, it can also replicate bias depending on the training data it receives (which is why responsible AI approaches are becoming increasingly important).
However, with a keen eye and a better understanding of how AI operates in general, you as the writer or content publisher can take steps to help reduce bias and hallucinations from seeping into your writing when using AI.
One of the most important things to remember when working with AI is that it’s trained on an incredible amount of data, which depending on the model, may include books, facts, figures, and online publications.
If the data contains bias or misinformation, the AI tool could generate biased or incorrect information.
AI can also be convincingly authentic when it cites studies, figures, and experts that aren’t real. By now, we’ve all heard of the law firm that used ChatGPT to cite cases that didn’t exist.
So, considering that AI learns and generates based on its learnings, being aware of the potential for bias and AI hallucinations and then removing them from the copy during the editorial process, is a great first step.
Essentially, prior awareness makes it easier to spot and remove any outdated or incorrect perspectives, so as not to perpetuate the same bias or incorrect information that the AI learned.
Some AI models have the ability to search for information online, but others are still learning and adapting based on old information.
Check the tool you are using to find out if it has a knowledge cut-off date and if so, when that is. For example, OpenAI notes that GPT-4 has a knowledge cut-off date of October 2023.
That means, if an event occurred after the cut-off date, the tool is limited to the recency of information it can provide.
The key takeaway with knowledge cut-off dates? Check facts closely and conduct additional research to make sure you’re including the latest data and information in published content.
What that law firm failed to do is a strong example of why verifying facts is essential.
As a writer or content professional verifying facts before you publish is key.
Using fact-checking tools like Originality.ai’s Automated Fact Checker gives you real-time insight (with sources that you can cross-reference) into whether or not a piece of content is factual.
Taking advantage of fact-checking tools gives your content more credibility while building authority in your niche or industry.
As well as fact-checking the AI output during the editing process, you can also engineer prompts to guide ChatGPT and other AI platforms (however it’s still essential to review the generated text even when tailoring prompts).
Asking the AI to clarify its sources with backlinks can help you get factual information that you can cross-check to verify.
Some prompt examples include:
Further broaden research by prompting the AI to provide different points of view, including arguments both for and against the topic, to get a variety of perspectives.
Even if you’re working with AI, sometimes you need to include a human perspective. It’s best to consult the experts when it comes to in-depth subject matter such as legal, medical, and technical areas.
This makes sure that you’re getting relevant, factual information right from the source and keeping up with industry standards beyond what AI can produce.
It may be a good idea as a general rule, to start now and create a library of reliable and trusted sources, as well as content guidelines for current and future writers.
Define AI guidelines around using generative AI for writing. Then, go a step further and include guidelines on terminology, the tone of your publications, and the writing and formatting style which can help everyone stay on track.
This isn’t to say that every response the AI generates will necessarily have hallucinations or bias in it. At the same time, it’s important to acknowledge that AI still has its limitations.
Be open about the use of AI in your content, incorporate a Site Scanner to highlight instances of AI that may need further review, and use a fact-checking tool to streamline your editorial process. Additionally, you could also encourage readers to bring any inaccuracies or issues to your attention, so that you can make changes accordingly.
In this way, you maintain accountability and transparency in your writing and continue to publish with integrity.
Learn more about fact-checking: