A friend emailed me asking about the latest AI-powered tools that we instructional designers might use in our work. 2022 witnessed an explosion in the quantity and quality of AI-powered media creation tools, and I wanted to summarize some of the main types and leading products in this space that we can use in our day-to-day instructional design practice.
Uses for AI Content Creation Tools
Text to Image Generation
One of the most eye-catching new uses for ChatGPT is tools like DALL-E2 and Midjourney, where you can write a text-based prompt and receive back incredibly detailed original artworks in a variety of styles.
Uses in ID:
Clearly, these tools can potentially generate license-free artworks that can illustrate novel concepts that you may not be able to source from other free image search options. If you need an image of “Barack Obama riding a polar bear during the French Revolution”, this is your best option outside Photoshop.
AI based Photo to Art
AI is being used to automagically improve existing images in tools like Prisma or its newest product, Lensa. (I even got a series of Lensa selfies done this break 😆). In this type of tool, there is little-to-no human input in the creative process — the AI ingests existing images and makes its own “improvements” independently.
These improvements can be subtle, like a conventional “auto enhance” feature that optimizes images for color and clarity, or it can radically redesign an image into different art styles as in the examples below:
Uses in ID:
You may be able to use these types of image tools to provide a memorable and coherent look and feel to a series of images without fancy Photoshop skills or graphic design chops.
Intelligent Media Manipulation in Conventional Tools
This year conventional multimedia tools from Adobe, DaVinci, and Affinity (among many others) all announced clever features that leverage the power of AI to dramatically improve the power and usability of these tools.
Adobe’s Sensei integrates with their existing Creative Cloud products to improve tedious or inexact tasks like removing backgrounds from images and video to cleaning up poorly-recorded audio narration.
DaVinci Resolve (my favorite video editing suite) has a paid tier that features AI-powered features like intelligent pixel selection, enabling you to highlight objects within a moving video frame and apply effects to them.
Uses in ID:
It’s a growing trend that most of the design tools we already use and depend on will start to see AI-powered enhancements integrated into them. Keep an eye on product announcements from your favorite media creation tech vendors to learn more about the new features coming soon.
Another eye-catching feature of ChatGPT is its ability to quickly write essays, poems, blog posts… generating convincing original works in almost any style imaginable.
Many academics are blasting this as “the end of the college essay”, claiming that students will just generate essays rather than writing them themselves. Truly, we are entering a world where all kinds of writing will be significantly easier to produce, for good and for ill.
Tools like Rytr.me can indeed generate fairly coherent texts, but they often will exhibit good use of the mechanics of language while failing spectacularly on accuracy or deep subject knowledge. Someone on my socials aptly compared AI writing to “mansplaining” — speaking eloquently about things it knows nothing about…
Another more valuable use of ChatGPT is to provide an “intelligent spell/grammar check” in tools like QuillBot, where you are still the main creative force, but it helps you format your ideas appropriately for your audience. We use this for copy editing on our team since some of our IDs speak English as a second language. We can pass text through a tool like this as a “first pass” for catching non-standard uses of English that would trip up our learners.
Uses in ID:
I probably will not use something like Rytr to generate my documentation for me or to write my blog posts, but I have found it useful during the planning phase to come up with an outline that I can build upon or improve. You are probably already using a rudimentary version of this in the enhanced sentence autocompletion features in Android, iOS, Gmail, and MS Outlook.
However, intelligent AI-powered spelling and grammar checks are a huge improvement over the mediocre tools we have in MS Word for correcting our work, and can be integrated in all kinds of different tools to better support diverse ways of working.
Intelligent Transcription and Text-based Media Editing
Our team has been making increasing use of Descript, a tool that
- analyzes audio/video files
- generates text transcripts via AI
- lets you edit the video and audio by editing text.
This is incredibly powerful in practice, and it enables us to quickly
- generate accurate transcripts from any media (proofreading still required)
- edit media faster than traditional timeline-based editing
- edit media without traditional editing skills
This is a case in which we are able to offer more services within the same time and resource constraints using a nice AI-powered tool.
Costly API = Costly Services
All of these tools are based upon huge data sets that have been trained at great expense over years. Every tool that we’ve discussed here calls on an API that costs money to access, so every application where we see AI-powered tools will always come with a significant cost (to somebody).
You use Midjourney > they pay to use ChatGPT > Who pays them?
Likely, many of these tools will follow the “drug dealer model” where you use the tool a little for free at first, get hooked, and pay more later once you’re dependent. Always ask yourself which of these tools you would actually pay money for, and how much.
This is also to say that most of these tools will come as paid subscriptions, not one-off license payments. Consider this when replacing a tool you already own with a tool you will be paying for continuously into the future.
Questionable Authorship/ Ownership
These models were trained by letting an AI browse through millions of web pages, images, and other creative works so it could learn how to create its own. Already, the artists and creators whose work contributed to that training are gearing up for a fight over their intellectual property, which they say was improperly used.
By using these tools, you are also using media whose intellectual property is not clearly legally defined. Before putting out a significant body of AI-generated art, you should have a contingency plan if you later find out you didn’t have the right to share that work and need to go back and replace it all.
As mentioned above, ChatGPT and other large language models may sound convincing, but they do not know what they are talking about. Under no circumstances should we look at these tools as replacement for human subject matter experts, but should rather find ways that they can augment and enhance our human-powered content development work.
AI reflects the unfortunate features of the humans who trained it
AI has shown an unfortunate tendency to reflect and reinforce the existing power inequalities present in human society. After all, it’s trained by reading millions of web pages written by humans, so it internalizes the same kinds of biases present in human thinking and treats those as the template for its original creations. You may remember Microsoft’s racist chatbot…
The challenges of AI go beyond this sadly humorous example, though, into the subtle ways that marginalized voices and identities are quietly excluded from AI training data sets. Whenever using AI-powered tools, it is incumbent upon you (the human) to think critically about the ways it includes (and excludes) certain types of people and adjust accordingly.
Liked this post? Follow this blog to get more.