THE CHALLENGE
UNESDA, an association that represents interest in the European soft drinks industry, expressed needing help optimising the workflow of their comms department, specifically, improving their internal social media workload. One of the ways they wanted to improve this aspect is through the use of AI tools. However, to achieve their goal, they needed to know:
- What sort of AI tools would be the best fit for them.
- How does one work with AI tools to get the needed results.
- How to set up a framework for working with the tools.
WHAT DID WE DO
To help UNESDA achieve their goal, we have organised an internal, two-day training session for their comms department, during which we went over various topics regarding AI.
On day 1, we went over how one defies AI and how it is important to change one’s mindset when working with AI, in regards to copy.
During day 2, we went over AI tools one can use for generating images, but also about various dilemmas surrounding the ethical use of AI. From minor mistakes and biases to deep fakes and the uncanny valley phenomenon.
WHAT DID THEY LEARN
DAY 1
To work with AI tools and get the desired results, we have to change the mindset and approach: it’s not creating, it’s delegating tasks.
4 Basic Modes of Interaction
As such, there are different modes one can interact with AI. You can treat AI as:
- An Intern, where you ask it to “Do something for me.”
- A Thought Partner, with whom you consult on an action plan.
- A Coach/Critic, who you ask for an opinion on what can be improved on.
- A Teacher, that you ask to teach or explain a concept to you.
Experience has shown, that most people use the Intern mode.
The RACEF Framework
Lastly, we have explained the concept of the RACEF framework. RACEF is short for role, action, context, examples, and format. It is one of the frameworks that one can use to setup/train ChatGPT and other AI tools, as close to the desired result as possible. Here is how it works:
- You introduce yourself to the tool by your ROLE (can be factual or fictional).
- You state an ACTION you need AI to do (example: “Write a copy about biodiversity.”).
- You explain the CONTEXT of what you want AI to do (Give them information about the topic.)
- You provide AI EXAMPLES of how you want the end-result to look (examples of previous copies)
- You explain the FORMAT of how you want the result to look (The copy should be 2 paragraphs, with emojis)
DAY 2
Besides writing copy, another task for AI tools is image and video generation and editing. But with these sorts of tools, one has to be wary of things going wrong.
Ethics of AI Creation
First off, users have to use AI in an ethical manner. We can use AI to erase regional differences or remove people that do not want to be featured or get copyrighted.
If we give the AI tools power to make decisions, we may end up with an end result that is not be desirable.
We all know that some AI-generated images may feature noticeable mistakes like extra limbs or wonky eyes. We’ve all seen the examples, and we know to expect them.
But AI tools rely on information that can be found online, which can include certain biases. For example, you can request a picture of a CEO, and it will generate an image of a man in a suit. Or you can ask for an image of Sudanese Barbie, and it will feature a doll with a weapon. Other times, it may even go in the opposite direction of overregulation by generating an image of a Black Viking. Hence, it’s best to keep an open eye on these images.
Deepfakes
Another problem that we may face regarding ethics of AI-generated images is the problem of deepfakes. A deepfake is an artificial image or video produced using “deep” learning, a unique form of machine learning. While they are not always easily recognisable, there are ways to identify them:
- AI tends to create hyper-realistic, overly polished faces.
- As we mentioned above, biased training data delivers biased outputs.
- Look for details that are not quite fitting or wrong.
- Look for contextual inconsistencies.
The Uncanny Valley
Speaking of hyper-realistic, overly polished faces, AI-generated images are also known for being subjected to the Uncanny Valley phenomenon. Defined as “an unnerving sense when androids, humanoid robots, and audio/visual simulations mimic humans yet aren’t convincing,” this phenomenon can have a less than desired effect when trying to create images as close to realism as possible.
Instead of creating hyper-realistic entities, one needs to focus on visually appealing, quality images with a distinct style.
SUMMARY
AI tools can be useful, but if one is not equipped/trained in how to use them in an ethical way, then you will never get the result you need.
Hence, it is crucial to have compact training in order to know how to interact with them and what makes them tick.
By doing so, one can master AI tools with swiftness and ease.