Blog Post

Experimenting and being transparent with AI

By: Marco Campana
June 7, 2024

On its own this is a super useful post from CAST (Centre for the Acceleration of Social Technology) (from the UK, of course) with a great tool you can use to quickly think through any use of Generative AI - the AI Experimentation Canvas.

"It’s a simple canvas that helps us think through any use of GenAI. As the test or practice becomes bigger, so we might use a more detailed version to document and think through the approach. We’ve got a Miro version that is easy to edit, and printable versions if Post-its and Sharpies are your style."

Super interesting and useful.

But the second section is what I found really interesting, and something we can all start implementing today - the AI Transparency statement:

"For now, that means adding an AI Transparency statement to the end of our writing and presentations — originated by Kester Brewin — to be clear about our use. This helps in conversations with partners but also ensures we’re being mindful of our use. The simple table below is the current structure we use — in this case, completed for this article."

The long rationale, from Brewin's article:

"My work – for a research charity exploring the impacts of AI on the UK labour market – means that I read daily about the profound implications of this technological revolution on almost every occupation. In the creative industries, the impact is already enormous.

This was why, having finished the book, I decided that my friends were right: I did need to face the inevitable question head-on and offer full disclosure. I needed an AI transparency statement, to be printed at the start of my book.

I searched the internet, thinking that I’d be able to find a template. Finding nothing, I had to come up with one myself.

I decided on four dimensions that needed covering.

First, has any text been generated using AI?

Second, has any text been improved using AI? This might include an AI system like Grammarly offering suggestions to reorder sentences or words to increase a clarity score.

Third, has any text been suggested using AI? This might include asking ChatGPT for an outline, or having the next paragraph drafted based on previous text.

Fourth, has the text been corrected using AI and – if so – have suggestions for spelling and grammar been accepted or rejected based on human discretion?"

Here's the simplified table used in the CAST article:

It makes sense for all of use creating content to use this and let our audiences know where and how AI has been used in our content, services, etc. Like other content you share, think about what creates trust. It could be providing your sources. Providing the last update date. Being clear about your bias and mission, which impacts your perspective and content. Is what you're sharing accurate and true on the day you share it? Things you might already acknowledge using the simple CRAAP test:

  • Currency: Is the source up to date?
  • Relevance: Is the source relevant to your research?
  • Authority: Where is the source published? Who is the author? Are they considered reputable and trustworthy in their field?
  • Accuracy: Is the source supported by evidence? Are the claims cited correctly?
  • Purpose: What was the motive behind publishing this source?

This is an extension of that trust building behaviour.

Modifying for our purposes

Brewin was writing a book, so his focus was on the use of AI and text. But ours is broader. So I would modify the questions we ask ourselves to:

First, has anything in this piece of content been generated using AI? This could be text, images, videos, etc.

Second, has anything in this piece of content been improved using AI? This might include an AI system like Grammarly offering suggestions to reorder sentences or words to increase a clarity score. It might be uploading an image to a GenAI tool and asking for different edits or new versions in different image or art styles.

Third, has anything been suggested using AI? This might include asking ChatGPT for an outline, or having the next paragraph drafted based on previous text. It could also include images or other elements suggested by a GenAI tool, such as images suggested in a Perplexity.ai "Generate image" prompt in the sidebar.

Fourth, has anything been corrected using AI and – if so – have suggestions for spelling and grammar been accepted or rejected based on human discretion?

How I'm using it

So I made a simple version for myself, which I'll start using on my site. So, for today's article here's my transparency statement:

Leave a Reply

Your email address will not be published. Required fields are marked *

arrow-circle-upmagnifier