I see a lot of extreme hot takes on AI. It’s either completely revolutionary or it’s total garbage. There aren’t as many people voicing a moderate take on AI. Personally, I think AI is useful in some situations, but it isn’t changing everything about my entire role.
The extreme views
These aren’t exact quotes, but rather paraphrases that are representative of statements I’ve read or heard. Have you seen comments like these?
On one end of the spectrum is the hype about how revolutionary it is.
- “AI will revolutionize everything about how we live and work in the next 2 years!”
- “AI can create whole courses in minutes with no skill required!”
- “The whole software industry will be gutted within 2 years. No one will buy software anymore since they can vibe code their own!”
- “Anyone in L&D who isn’t completely changing everything about how they approach their work will be left behind.”
- “Nobody will need to create courses at all. Everything can be generated by AI on the fly and personalized to the user.”
On the other end of the spectrum is the view that everything about AI is garbage, that AI is just a passing fad.
- “I’ll never use any AI in my work!”
- “If I see someone share an AI-generated image, I immediately block them.”
- “Why are we even talking about AI? The whole thing is a grift that will be gone in a year or two like every other shiny thing.”
- “Everything AI produces is garbage.”
- “I don’t know any instructional designers who are actually using AI in their work in any substantive way.”
- “Remember when people said we’d be doing elearning on our smart watches? AI is the same kind of dumb fad.”
In the instructional design subreddit, one user used to go so far as to regularly comment asking the mods to ban everyone who mentioned or asked questions about AI. They were so opposed to AI that they didn’t even think we should discuss it.
A moderate take on AI
What I don’t see as much of is the views in the middle. Of course, there are some–I follow several rational folks who share more balanced opinions. But those views don’t get as much attention as the extreme views. Based on my own experience, I have a moderate take on AI. Although both the revolutionary and garbage exist, I find myself working in the middle ground. There are genuinely practical uses for AI in improving efficiency and expanding capabilities right now. I can envision larger changes on the horizon too, and I think we need to be looking ahead at the long-term possibilities.
Amara’s law
“We tend to overestimate the effect of a technology in the short run and underestimate the effect in the long run.”
–Roy Amara, futurist
I find myself frequently revisiting Amara’s law in reflecting on AI. Keeping this framework in mind helps me evaluate claims about AI.
Overestimating the short-term effects
I see a lot of hype about how AI will revolutionize everything in the next few years, making the mistake of the first part of that statement. Part of the issue is that no matter how capable the technology is that people, organizations, and behaviors don’t change as quickly.
“Even if AI development stopped today, we would have years of change ahead of us integrating these systems into our world.”
–Ethan Mollick, professor and AI researcher
I find myself more in agreement with Ethan Mollick. The cultural, organizational, and societal changes related to AI are going to take much longer than what I think most predictions state. The initial changes with AI are going to be small and incremental, regardless of how revolutionary the technology might be. Organizations move slower than the rapid pace of technological change. The technology is continuing to improve as well, and it’s going to take longer for the cultural changes to happen.
Underestimating the long-term effects
I think the opinions that everything AI is garbage and a passing fad that will vanish in another year or two miss the second half of Amara’s law. To be clear: a lot of what people produce with AI is garbage. We’ve probably all seen ridiculous AI-generated images with too many fingers, creepy AI avatars in the uncanny valley, and AI-generated videos with broken physics. You’ve read text that was painfully obviously generated with AI.
But the technology is improving. AI images are already at the point where they can be indistinguishable from real photos, and AI videos are improving rapidly. AI voices are significantly better than they were even a year ago, and that’s already significantly affecting the voice over industry.
It’s impossible to really predict the long-term effects of AI. Personally, I’m often skeptical of new technology trends. I keep up on what’s happening so I know what’s possible, but I don’t usually jump in as one of the earliest adopters of technology. Having seen lots of trends fizzle out and not live up to the hype, AI feels different to me. Partly, that’s because we can use AI in practical ways right now–it’s not just something that could hypothetically be useful in the future.
This feels like a much larger shift for the long term as well. In the same way that the internet changed many jobs and created new ones, I think AI could be similar. The internet is just how we get things done now–communicating, shopping, working. Being online is integrated our lives in hundreds of small ways that differ from thirty years ago. Thirty years from now, AI could be similarly integrated in numerous small ways in our jobs, education, and communication.
People are justified in raising ethical and practical concerns with AI too. We need to think about how we keep a “human in the loop” for most uses of AI in order to mitigate the potential harms. The energy use and environmental effects will require innovation to address. I’m really hoping to see more improvements in energy efficiency, not just innovations in producing more energy. Intellectual property and information security are other big topics. AI can enable scams and frauds, some of which are pretty scary. Those are all long-term effects that often get glossed over in discussions of AI.
Standing in the middle now
So where does that leave me with my moderate take on AI? I’m somewhere in the middle. While I don’t think people need to panic if they’re not using AI, I do think it’s smart to spend time experimenting with the tools to figure out how they can help. Even if right now you focus just on the things where you can improve your efficiency by a small amount, that’s probably worthwhile. Starting with the kinds of tasks I shared with my useful prompts for instructional designers is one entry point.
You have to spend time using the tools to get a real sense of what’s possible. Ethan Mollick has stated several times that it takes about 10 hours working with an LLM like ChatGPT or Claude to be truly familiar with the capabilities. While the tools are improving all the time, you’re more likely to find LLMs useful if you’re working with one of the newer and better models. If you’re on a free LLM and every bit of text sounds like the same AI text you read everywhere else, just switching models may give you better results.
While I do use Claude for plenty of tasks, one of the bigger shifts in my own work is in media generation. I have worked with some amazing voice actors, but it’s getting harder to justify the cost for a lot of my clients. AI voices aren’t as good, but they’re good enough for a lot of applications.
For images, I’m probably using more AI-generated images than stock images these days. It’s not just that I can generate the same images as stock images, just faster and cheaper. It’s that I can generate unique, custom images, tailored to specific audiences and purposes. If I need a series of images in the same style to create a unified look, I don’t have to hope I can find a good set in an existing library somewhere. I can generate consistent characters in more natural poses than was ever possible with cutout character images. AI image generation has limits and issues, but so do stock images.
I’m already at the point where I can do more with AI images than I can with stock images, and that’s a genuine improvement in the quality of what I produce. It’s not revolutionary, but it’s not garbage either. It’s why I find myself with a moderate view.
I also wrote a reflection on how I thought AI will affect instructional design in October 2023. Rereading that post today, almost a year and a half later, I feel like my previous predictions have been reasonably accurate so far.
Where are you in the continuum?
Where are you in the continuum from “AI is revolutionary” to “AI is garbage”? Are you somewhere in the middle like me? Are you leaning more to one end of the spectrum or the other? Let me know in the comments.
More on how I’m using AI now
These are some of my “working aloud” posts on how I use AI, focused on the practical applications without hype or fearmongering.
- 7 Useful Prompts for Instructional Designers
- Create Consistent Character Images in Midjourney
- Experiments with Midjourney Character Images
- AI Images with Consistent Style in Midjourney
- Using the Midjourney Editor to Change Character Expressions
- Experimenting with Character Generation in ConsistentCharacter.ai
- AI Image to Video Experiments with Hedra
- How to Use AI to Write Scenarios
- Draft Choices and Alternate Paths with ChatGPT
- Prototype Branching Scenario Written by ChatGPT
Christy, I can’t thank you enough for this post. SO much AI-related hype bombards my email inbox it’s hard not to feel that, despite my best efforts to keep up, I’m falling behind. I appreciate your common sense approach to this ever-shape-shifting animal we know as AI.
You’re welcome! I’m glad this helps you cut through the hype. There is a lot of FOMO with AI.
A lot of what is happening now with AI is true early adopter stuff. The tech is changing so quickly that the specific skills like prompting are going to be very fluid. Pay attention and experiment when you can. Take breaks when it gets to be too much.
I would say I’m in the upper middle. I’m excited about AI, I am using it, and I am sacrificing my time to learn and get good at it because I know it will only get better. I agree that we need clarity on intellectual property questions- this has been my biggest hang-up in using it in my own content.
Thanks for your insights!
Upper middle is a good place to be too. You can be skeptical of the most egregious hype while still looking forward to new possibilities as the technology improves.
It’s funny–I actually wrote this post last week. After I scheduled it, I had a call about a potential new project that has me really excited. That would include training AI on some proprietary data (and gathering and licensing data to use for this purpose). If that project actually happens, I will probably shift to that upper middle range myself. The possibilities for small, local AI models trained on specific data are intriguing, and this gets around some of the problems with intellectual property.
I’ve been an AI promoter for literally decades. Yet, I’m probably one of your ‘garbage’ voices. It’s not that generative AI isn’t useful, and you’re using it in plausible ways, but I take Cathy’s point: IP infringement. Where did those images you’re asking for come from? The voices were trained on what? What text was used to train the language models?
I’ll suggest it’s stolen IP, to the point where OpenAI is petitioning the government for the right to violate copyright as a competitive edge internationally. That’s not how it should work. Yes, we can wait ’til the courts sort it out, but I’d rather be proactive.
Lots of legitimate ways to use AI, for sure, and I’m a fan, but generative AI has, largely, been built in dubious ways. And don’t get me started on the environmental cost ;).
Christy, you know I’m a fan of your work (I’m reading this, yeah?), but when it comes to generative AI, I have to point out the flaws as well as the upsides, and copyright is a big problem to me.
I’d label you as an “AI skeptic” rather than at the very far end of the scale. You aren’t refusing to even look at AI, which I see too much of.
For voices, the IP question is pretty straightforward. ElevenLabs paid voice actors to train the model. They have licensing agreements, and the voice actors retain the rights to their voices.
For text and images, it’s not as straightforward. I have great respect for you, Clark, but this is an area where you and I disagree. I think it actually is fair use.
The purpose and character of the use is transformative, which weighs in favor of it being fair use. AI tools like LLMs and image generators take bits from many different sources and remix them into something new.
I also think it doesn’t affect the potential market enough for that factor of fair use. I know my blog has been used to train AI, just like yours. But I don’t see that I’m losing a dime of potential income due to it. Nobody is using AI instead of hiring me to create training.
The works used for training are generally published content, another factor in favor of fair use. There is an argument that they shouldn’t be training on unpublished content, and that may be something the courts rule against.
So, that’s 3 factors in favor of fair use. The final one is the strongest argument against fair use, which is that complete works are used. But fair use is never a question of a single factor. It’s always a question of looking at all of the factors together.
Whether the courts rule that way or not is of course an open question. And of course, you’re allowed to disagree. Two smart people are never going to agree on everything. 😊
Interesting. I can’t see it as ‘fair use’. 1) They’ve taken everything to train, not just ‘bits’. The fact that it’s outputting bits isn’t the issue, it’s what they’ve taken to generate it. 2) It’s not whether it’s keeping me from getting work, it’s whether they’re taking my work and not recognizing that in any way. They’re charging for output generated on the work of many, with no recompense. 3) My blog is published CC: BY SA NC. Thus, to take it and charge isn’t fair use. And, yes, we’ll see what the courts say. Thanks for the reply.
How much they use and how they use it are two separate factors for fair use, but it’s understandable to conflate them.
Have you set up your blog to block the AI bots? You can change your settings to prevent them from crawling your site. That won’t remove any data that has already been pulled, but you could at least prevent them from using your data moving forward.
I figure if the IP attorneys can’t agree whether it’s fair use or not, it’s OK if we disagree too. I do think that people should have the option to protect their data, and you should adjust your blog settings to stop it.
BTW, if you’re not blocking AI bots yet, here’s an article explaining how to do so: https://www.20i.com/blog/how-to-prevent-ai-from-scraping-your-website/
I am also in the moderate camp. I am finding good uses for AI right now, but I’ve also been disappointed in other areas. I haven’t been impressed with AI’s writing, but we’re still using the free tools. You’re not the first person I’ve heard say that the paid tools are way better.
AI voices are getting better and better. Professional voice talent is a huge expense for us (some of our large products have 30-40 HOURS of narrated content.) We’re slowly adding AI voice, launching products, and watching for negative comments or slowing sales. So far, we’re not seeing any complaints or lower sales.
We’re using AI image generation from stock sites and in Adobe’s products. Our organization is still wary about IP and copyright infringement. It isn’t just corporate culture that may slow down AI adoption. The legal ramifications of AI usage is still up in the air. LIke many places, my organization is still working on our AI use guidelines.
In another 6 months, the free LLMs may be comparable to the current paid models for writing. But you may want to pay $20 for one month of Poe to try some prompts in multiple different LLMs to compare your results. That might help you determine if the difference is worth paying for, or if you’re better using it for ideation and other tasks now.
If I had projects with 30-40 hours of narrated content, I would 100% be looking at AI voices too. Even a small amount of negative comments or slowing sales might be a reasonable tradeoff.
Adobe Firefly is the best in class for protecting intellectual property–they only use licensed images to train their model. I think it’s reasonable for companies to be cautious about the legal questions, and you’re right that it’s more than just culture that needs to change. We need clarity on intellectual property questions before most of the the widespread adoption will be possible.