Image generated using openart.ai

There are several myths and misconceptions around generative AI that are worth considering. We address some key ones here, especially ones that pertain to its responsible and productive use in education.

Generative AI learns from my inputs

Generative AI tools can better understand us if we have an extended conversation with it. For example, in a conversation we may prompt it to take on the persona of an expert Socratic tutor, only answering questions by asking more questions. It is then going to keep this earlier prompt in mind when responding to later prompts in the same conversation. Its ability to do this relies on its ability to look back through that same conversation in what is called a ‘context window’. The context window varies in size between different AI models - the older GPT-4 has a context window of around 6,000 words, while the newer AI models have context windows upwards of 90,000 words or more.

Your conversations do not automatically or immediately get used to train the AI models that you are using currently. When AI models are created by AI companies and labs, they need to be 'trained' on large amounts of text and other information, which is how they 'learn'. Some AI companies may take and use your inputs to train future AI models - this is why it is important to be aware of your privacy rights and also opt out of data collection where possible. AI tools made available by the University do not train AI on your inputs.

AI outputs are detectable

Software that purports to detect the use of AI-generated text is prone to false positives and false negatives. Research has suggested these tools are not reliable and may be biased against non-native English writers. In addition, generative AI software such as ChatGPT does not ‘know’ whether it generated a piece of text – even if it may produce a convincing response when asked. The use by students who are knowledgeable enough about prompting, follow 'adversarial' AI-masking approaches freely available on social media, or have access to detection-bypassing software is very likely to be undetectable by software or by humans. AI detection relies on the user not knowing how to use it properly and can be circumvented by AI tools designed to do so or by tactics advertised to students available widely on social media.

Using generative AI destroys the environment

Humans have a deeply complicated relationship with technology and modern conveniences. The systems that you are using right now to access this page consume energy and water that are needed to power and cool datacentres across the world. Half a cup of coffee takes 140 litres of water to produce, as well as having numerous other environmental impacts. Some early estimates suggest that a conversation with ChatGPT consumes 500 mL of water, although AI models are getting more efficient all the time. AI models are becoming increasingly efficient. It's estimated that a Zoom call or streaming video consumes significantly more energy and water compared to a conversation with generative AI.

Therefore, while it is crucial that we are aware of the real environmental impact of generative AI usage, we must also be cognisant that our engagement with modern life also has a severe impact on the environment - and work together towards a more sustainable future in all that we do.

Its references are all fake

This may have been true of very early AI tools like the initial release of ChatGPT in late 2022, but it is no longer the case. There are many AI tools that can engage with and synthesise real sources, including those that have access to real-time scholarly literature and the internet.

I can make my assessment AI-proof

Unfortunately, unless it is a secured, supervised assessment, this is not possible. Generative AI tools are getting better all the time, and legacy approaches to 'AI proofing' assessments (such as requiring a personal reflection, current references, in-class sources, multimedia submissions, analysing class data, etc.) can all be completed using AI.