Main question | What information do the employees of GroupM need to know/learn to effectively use generative AI in their daily workflow? |
Subquestions |
|
Method used | Literature study, survey, day in the life |
Date | 02-04-2023 |
Link to document | Research 5: Knowledge required for the training program |
It is important to first have a clear understanding of the target audience's knowledge before creating tailored content to their needs. For this reason together with Anouk, I created a survey to test the target audience's knowledge on generative AI. The aim of this survey is to find out how they currently write text prompts, what they want to learn, their problems/frustration and if they are aware of the legal and ethical dangers when using generative AI.
People had difficulty with coming up with correct text prompts, mentioned they want to learn how to write good text prompts, learn more use cases for generative AI for specific work situations and in some cases were afraid of the legal and ethical concerns around generative AI. When writing text prompts themselves out of 11 people almost everyone used in some cases private information in a prompt.
In general the information that people need and want is:
The best generative AI tools out there right now are ChatGPT and Midjourney. For this reason I will look at these two AI tools on how to write effective text prompts for them.
For text generation to create effective text prompts it is important to check your prompt if it is clear & concise, no ambiguity and uses natural language, has a specific target it needs to accomplish, and the context contains detailed information about the situation. Optional to add certain guidelines on how the output should look like. If you find yourself in difficult situations and you don’t know how to write a good prompt, some prompt patterns can help you formulate one or get inspiration from people before you. Last but not least it is important to be creative, unique and experiment when writing prompts.(White, 2023)
Image generation works differently than text generation. For this reason prompting is in some ways similar but also different. For image prompting it is very important that you yourself have a clear vision of how you want your image to look like, it comes in handy if you are knowledgeable in creating art, photography, etc. The general set-up for an effective image prompt is: clear and detailed description of the context/scene it takes place, the eventual output or display, the style (lighting, art style, theme, etc), and composition (aspect ratio, camera angle, resolution, etc). For general prompting it is important to take inspiration from the community: What prompts do they use?, and experiment yourself. (Kadir Akin, 2023)
As mentioned in the survey they wanted to learn more about how they could potentially use generative AI in their own workflow. I held a few days in the lives and interviews of employees to confirm the creative cycle in my previous research and get more in depth information about the tasks of the creative departments. Essentially generative AI could be used in each phase but in the creative phase it can help with storyboarding, making a coherent story to convince the client, writing copy, inspirations for mockups, change format of image by adding new parts, and more. There are so many possible solutions of generative AI. The best way to show these possibilities is done via real-life scenarios where they could apply generative AI.
The last thing mentioned in the survey is the ethical and legal concerns when using generative AI. WPP (holding company of GroupM), legal and technology team made a presentation on how to responsibly use generative AI. They mentioned when using this technology it was important to look at your input if there is no sensitive data. For the output they mentioned to check the limitations on the tools itself (Biases, generic, misleading content), Ownership of the output, I.P. infringements and ethics. They created a set of principles that everyone should uphold when using generative AI.
The knowledge missing from the employees is: writing good text prompts to get the desired results, specific use cases with generative AI in their workflow and ethical and legal considerations when using generative AI. These three topics are the missing knowledge for employees to effectively start using generative AI in their workflow. In this research I have for every topic the information summarized. The last thing to do is create a fitting training program that addresses these topics.