Human-Centered Considerations When Creating and Using AI Tools
“The use of modern gadgetry cannot supplant the use of proper techniques and principles” - Theodore Schellenberg
The quote in the subtitle is one of my favorites from the father of modern American archivism. Archivism is the recording, organization, and stewardship of historical and anthropological records. Schellenberg was concerned that archivists were using computers to replace critical thinking skills and proper procedures. He wrote a manual to distill the most important processes and considerations for archivists, and it has been in use for decades.
Schellenberg knew that proper work and procedures had to be human-oriented in order to be effective. Humans (especially at the time of his writing) had more contextual and background information than computers. A similar ideology is necessary for all types of work with technology.
Last week, I wrote a blog post about the Human Intelligence Movement and examined their central idea: that human ideas, priorities, and work should inform how students are trained to work with AI tools.
This week, I updated the College of Southern Idaho AI Literacy Course, an Open Educational Resource, twice in one day. I also updated it a few more times before, but both of these times were because advancements in the AI world made emphasizing user responsibility even more important than in the past.
Human-Centered Concepts For Creating AI Tools (and Structured Prompts)
The image below features the main considerations on which we should focus when creating custom AI tools or using general AI tools with structured prompts.
These concepts help the user (or creator) ensure that they “bring the human” to human-machine interactions like I discussed in my last post. If we are conscientious and deliberate about answering each of the questions below, we create a human-AI experience that will be manipulated by the human.
Content Strategy – what are you going to design the tool to create?
Task Goals – what tasks do you want it to fulfill?
Communication Preferences – how do you want it to communicate with users?
Data Strategy – what data do you want/not want users to give?
Interface Plan – what options or appearances will most attract your users?
Fine-Tuning Plan – What licenses or types or origins will be the source of your training data? Will you use internal data or documents? Which documents will you use? How will you get relevant permissions?
Prompt Design – What specifically will your system prompt be? What headings will you use? How will you instruct your tool to interact with the data in documents?
Training and Videos Related to These Concepts
The Human Intelligence Movement will undoubtedly release some formal videos or trainings related to these concepts. In the meantime, I and others have created our own videos and trainings.
This video is part of a series of Practical AI Tutorials related to the Professional Productivity with ChatGPT and AI Bootcamp I did with Steve Hargadon in July. The series (3 one-hour sessions and 10+ hours of tutorials) is available for $149/person. More materials will becoming from me and Steve in the future.
I also do workshops for the Online Learning Consortium, including this high-level introduction to AI in digital education. This week-long course is available for non-members of the consortium for $290.
What If I Don’t Have Hundreds of Dollars Earmarked for Prof Dev?
There are also free options, like this blog and the This Week in AI podcast. Additionally, you can follow a host of people on LinkedIn. Denise Turley, the Vice President of the United States Chamber of Commerce and a professor at the University of the Southwest, also has her own YouTube channel that offers AI training. View her two videos below that cover business-related applications of AI tools. In both videos, she takes the custom GPT route over the general AI structured prompt route.
Check out the rest of Dr. Turley’s videos on her YouTube channel, and feel free to go to any of other links and make them feel appreciated.
Other possible sources of inspiration include Dr. Anastasia Betts, who examines the importance of human knowledge, and Ioannis Anapliotis, who considers the ways that artificial intelligence can develop and demonstrate human ideas. Matthew Karabinos examines hos to create safe interaction environments for AI-student collaboration. Those four, and others like them, provide an excellent introduction into human-centered AI use and creation.
Human-Centered AI Use and General AI Literacy
The concepts and applications in this article relate to my AI Literacy Framework, which speaks about humans (as “users”) seven times. It also stresses that we must be active participants in human-AI interactions, rather than being passive consumers.
Workflow and Custom AI Tool Examples
I know that this has been a text-light post so far, so here are some text-filled workflows and Custom AI tools that I created. Keep the considerations from above in mind as you go through these projects.
Creating Videos with GenAI Tools
1. Create text prompts for image and moving picture generators, and scripts for the video audio, using prompt engineering techniques with a text generator such as ChatGPT, Bard, or Claude. ChatGPT is the most recommended tool for this step.
2. Convert the video script to an audio format using an audio generative AI tool such as text-to-speech.online, Bark, TTSMaker, or another open or free service.
3. Use the text prompts you created in step 1 for image generators with Dall-E, Getimg.ai, ideogram, or designer.microsoft.com.
4. Use text prompts for video generators you created in step 1 with tools such as typpo.app, teach-o-matic, or Chromox.
5. Combine all of these using your favorite audio and video creating software, including free offerings like those from Canva and CapCut.
Creating Images with Captions with GenAI Tools
1. Create text prompts for image generators using prompt engineering techniques with a text generator such as ChatGPT, Bard, or Claude. ChatGPT is the most recommended tool for this step.
2. Use the text prompts you created in step 1 for image generators with Dall-E, Getimg.ai, ideogram, or designer.microsoft.com.
3. If you do not want to create the captions yourself, you can put the images into a text generator, such as Claude or ChatGPT, and ask it to come up with a good caption for the image. Make sure the caption is descriptive and follow accessibility guidelines.
4. For optimum accessibility and equity, make sure to use text generators to create alt-text for the image. Or, write it yourself. Microsoft, Harvard, W3C, and other resources have excellent resources on writing high-quality alt-text.
Creating Cover Images with GenAI Tools
1. Create text prompts for image generators using prompt engineering techniques with a text generator such as ChatGPT, Bard, or Claude. ChatGPT is the most recommended tool for this step. Be sure that you tell the text generator to consider optimum cover or banner image traits.
2. Use the text prompts you created in step 1 for image generators with Dall-E, Getimg.ai, ideogram, or designer.microsoft.com. Be sure to include the parameters of the cover image suggestions for the particular site.
3. For optimum accessibility and equity, make sure to use text generators to create alt-text for the image. Or, write it yourself. Microsoft, Harvard, W3C, and other resources have excellent resources on writing high-quality alt-text.
Creating a Podcast Episode with GenAI Tools
1. Create scripts for the podcast audio using prompt engineering techniques with a text generator such as ChatGPT, Bard, or Claude. ChatGPT is the most recommended tool for this step. Also create image prompts for podcast section markers if you want to.
2. Convert the script to an audio format using an audio generative AI tool such as text-to-speech.online, Bark, TTSMaker, or another open or free service.
3. For podcast sections, use the text prompts you created in step 1 for image generators with Dall-E, Getimg.ai, ideogram, or designer.microsoft.com.
4. If you want video to accompany the podcast, you have multiple options. For AI-generated video such as a digital narrator you can use Adobe Express Animate-from Audio, Chromox, Teach-O-Matic, or Typpo. If you want to include moving images from your screen or something like that, you can play the audio, record the screen without sound according to the speech of the audio, and then save that moving image.
5. Another option is to include still screencaptures of your content and then have the audio speak while the image is displayed throughout a particular section. If you want to bring note to certain areas of the screen, you can use a highlighter to create different versions of the screenshot and then put those versions according to the audio.
6. If you want to create music for the podcast, you can use open source music creators such as Magenta. However, I recommend using open access music productions available on Openverse, Freegal, or another open source music source.
7. Combine all of these using your favorite audio and video creating software, including free offerings like those from Canva and CapCut.
Custom GPTs
All of my custom GPTs have been developed with the human-centered concepts above in mind. These include my Library Science Tutor, SavoyGPT (a Gilbert and Sullivan impersonator), an Instructional Design Assistant, a Search Query Optimizer, and a Library Programming Assistant. Another excellent tool is Dr. Jeanne Beatrix Law’s First-Year Comp Assistant.