Skills Education

Background to the development of the AITAS

When generative AI (GenAI) tool ChatGPT was released to mainstream public in late 2022, Michelle was an early adopter of the technology and started to immediately explore the use cases and implications of ChatGPT (and subsequent other tools released), especially with regards to teaching and learning environments.

Into 2023, as Michelle continued her university studies she wove GenAI into each of her assessments, commenting on the ethical and practical implications of its use. Her university lecturer at the time recognised Michelle's keen interest in the topic and suggested Michelle create a rubric to overlay and guide use of generative AI in educational environments. But that fell to the wayside as other priorities came to the fore for Michelle. Michelle continued to monitor the advancement of GenAI tools, making her own regular posts about various use cases and outcomes of trial applications. 

One of the most fascinating issues for Michelle was the gradual and forecast degradation of factual content; as the AI developers scraped the internet for training purposes, the pool of information on which the large language models were being trained became more flawed as AI-generated content (based on biased and flawed initial information) became part of the training data pool - giving rise to even more hallucinations that may have been for some, quite believable.

It wasn't until late 2024 that the idea of a rubric again surfaced onto Michelle's radar. 

Each year, Michelle is contracted to perform QA functions on presenter materials for a major industry conference. The purpose of the contract is to ensure accuracy and authenticity of content (including proper referencing of source materials) submitted for delivery at the conference.

When performing that role in 2024, Michelle was struck by the rise of session material that was both obvious and highly likely to have been generated with or by AI. Yet the rate of disclosure about this was low to zero.

One session in particular triggered the first thoughts of the AI Transparency Advice Scale. If it listed 10 citations, on checking, only one source was real (although still incorrect). The implications of this really started to hit home.

As a parallel, Michelle started to note a shift in social media feeds with content changing in style and tone compared to regular 'voice' of the poster, as well as a shift in volume. The ease with which GenAI can generate content has meant a proliferation of information. Yet, in many cases - especially with regards to compliance related matters - the information presented was incorrect; either partially or wholly. 

While there are many benefits to use of GenAI and the efficiencies it brings, and despite requiring students to adhere to policies that dictate allowable levels of AI involvement in assessment, the disclosure by businesses and educators on the levels of their own use of AI remains murky.

The AI Transparency Advice Scale was devised to be a simple indicator easily applied to content to inform the reader about the information they are consuming. Especially in the context of examples provided above, the utility of the Scale would be beneficial in supporting informed user action. 

Naturally, application of the AITA Scale is dependent on truthful disclosure of the business/individual developing the content. And other limitations may include relative subjectivity on determining levels within the Scale. 

However, if applied, the AITA Scale will give a clear gauge on the degree of AI use in how content has been produced. We believe this supports transparency in business and educational communications, and fosters development of critical literacies as audiences are prompted to recognise the origins of what they are seeing, hearing and reading.

- Updated 3.4.25