GenAI limitations

There are multiple ethical concerns, spanning a wide range of issues, that impact creating, sharing or just reading genAI text outputs. Some of those ethical concerns are explored further on this page (e.g. bias, fairness, copyright and legal, misuse and manipulation). As a student, you’re expected to develop critical thinking skills. Which means you need to interrogate how the genAI tool was made, who owns it, how are others using it, and how you use it in an ethical framing. Overall, it’s about being a global citizen who engages ethically and productively with technologies.  

In relation to text-based genAI outputs, we want you to be consider ethical questions like:

• Are you adding more noise to the mis-, dis- and mal-information sphere? Find out more in the Misinformation Guide  
• Is it your information to upload to a tool?
• Is the information private or sensitive and do you want that added to a large language model dataset?
• Do you know where your information is going?
• By using text-generation are you undermining your creative writing or learning?
• Are you comfortable using a text-generation tool that may have stolen the work of writers in developing its training data?
• Be aware that not everyone has access to the genAI text-based tools and what does that means for your use?  
• Have you thought about the energy consumption that large AI models like Bing Chat, ChatGPT or Google Bard use?
• Who is maintaining the text-based genAI tool and what are their working conditions?

Watch this video for a leading AI expert’s opinion on ethical genAI use.

Be aware that genAI text-tools have a range of ways they can be misused and manipulated. In relation to text-based genAI outputs, some examples about what misuse and manipulation might look like are:

• Impersonating the written voice of individuals or organisations which are used in scams or phishing attacks 
• Impersonating the written voice of individuals or organisations to discredit them
• Creating biased narratives (aka propaganda) in text-based outputs to persuade or convince people of a particular ideology or way of thinking 
• Using fabricated information that looks authoritative in an essay or report
• Take credit for creation without acknowledging the use of genAI in their work.

Watch this video that explores what happens when genAI text generates false information that isn’t checked. To appreciate the vast depth of the misinformation crisis, explore the library’s Misinformation Guide.

The quality of genAI text can vary and is affected by a lack of clear instructions and ambiguity. Achieving consistent high-quality results across diverse applications therefore remains a challenge.

When I asked Bing's Copilot (available free to you as a student or staff member of Deakin) to produce a short, engaging poem about AI and Australia. This is what it produced:

Binary whispers echo 'neath gum tree shade,

AI dances with Dreamtime, a digital serenade.

In the land of kangaroos and Southern Cross above, Wisdom blooms anew

—a pixelated love.

Looking at the above poem it is easy to see how genAI tools struggle to comprehend or reproduce content that involves sarcasm, tone, or cultural nuances. Which is why prompting is so important as the quality of your prompt impacts the quality of the output.

To get better at prompting, practice using our prompt engineering module.

Be conscious of in-built bias due to underlying patterns in the information the genAI tool is trained on. If the training data used to train generative text models is biased, the generated text will also exhibit biases. This could result in the reinforcement of stereotypes, discrimination, or other forms of bias present in the training data.

Using genAI text verbatim means you’re creating plagiarised content, forgeries, or content that infringes on copyrights and intellectual property rights and raises legal issues as genAI tools can’t create original work.

Remember you need to acknowledge any work that has generative AI content as per the Deakin Guidelines that can be found here.