Discover more from Jakob Nielsen on UX
AI’s Role in Human-AI Symbiosis: Originator or Refiner
Summary: It is cheaper to let AI take the initiative in human–AI collaborations where it generates a wide range of initial ideas that are then winnowed by human judgment. But higher quality can result from the opposite workflow, starting with a human draft and fine-tuning it through AI criticism and variations.
I have noticed that most people use generative AI opposite to my preferred approach. Neither way has a monopoly on excellent results, but understanding both will help you get better results from AI for your projects.
You can unleash the power of cost-effectiveness by letting AI be the creative spark in your projects, only to have human intuition refine it. Yet, don't underestimate the magic going the other way around when a human vision is perfected through AI.
Permalink for this article: https://www.uxtigers.com/post/ai-originator-refiner
In either case, I strongly recommend pursuing human–AI symbiosis, where the two sides work in partnership. The days are long gone when it made any economic sense for a business to employ knowledge workers in splendid isolation without AI assistance. Even at its current GPT-4 stage, AI amplifies productivity and creativity to unprecedented levels. And remember, this is from GPT-4 level AI, not from GPT 5, 6, or 7, as we’ll enjoy during the upcoming decade.
The First Law of AI states: Today’s AI is the worst we’ll ever have. Hence, strategic planning for personal growth or organizational development must integrate the imminent advancements in AI technology.
The future belongs to the symbiants, where humans and AI collaborate rather than compete. (“Symbiant” by Midjourney.)
Soon, symbiants will outperform unaided humans by a factor of two or more. This is the reasoning behind the Second Law of AI: You won't lose your job to AI, but to someone who uses AI better than you do. (See also my article, The AI Revolution Won’t Cause Mass Unemployment.)
But just as humans without AI augmentation will be losers, AI alone won’t cut it. Only a reckless fool relies on AI-generated output without subjecting it to human review. Even better than a single review, create work through an iterative process, alternating contributions from the two partners, AI and human. How to manage this alternation is the topic of this article.
AI as Originator: Initial AI Generation, Followed by Human Judgment
Using AI as the originator in the collaborative process is the most common approach today. In this workflow, we start by describing our goal in a simple prompt, and then we sit back and let AI unfold its creative potential. For example, if we want a children’s bedtime story illustration, we might issue the prompt, “A squirrel sitting on a leafy branch.” If we don’t add further specifications of the desired style, we might get results like this:
Two squirrels in different styles from Midjourney.
Now, the human partner takes over. The right squirrel is much cuter, whereas the left squirrel could be frightening for little kids. So, I’ll pick the right squirrel as the starting point for further iteration and refinement.
In a different scenario, I — the human master of the process — might know that I’m after a symbolic illustration of a squirrel to make a point in an article about the value of squirreling away some user-research nuts (resources) for late in a design project rather than expending my entire research budget in the early phases. In that case, I might feel that the left squirrel is likelier to impress an audience of hardnosed UX managers who would dismiss the cute animal. The artistically rendered animal is not exactly right, but that’s the one I would pick as my starting point for further iterations.
(If you follow my writing — or check my Instagram — you’ll see that I rarely use photorealistic images.)
Similarly, if we’re designing a website, we can start by telling the AI the type of site and what brand associations we’re aiming for. We then ask the AI to originate 20 different designs for this site and pick a few for further refinement.
Or, if we want an article, we feed the AI a proposed title and ask it to give us a few outlines. Or we cut to the chase and ask for a full draft manuscript from the get-go. Then — and I can’t emphasize this enough — we should proceed with an editorial review and change some parts or maybe ask the AI to elaborate on some points while shortening others.
AI as Refiner: Lead with Human Work, Have AI Polish
The opposite approach to human–AI collaboration puts humans even more in the driver’s seat but at the cost of more human up-front effort. For example, if I want to produce an article about the two approaches to human–AI collaboration, I’ll start by writing the draft on my own, based on my thinking about this problem. Then, I’ll ask AI to criticize my draft, suggest improvements, point out things I have overlooked, and write a compelling social media post (complete with emojis) to promote my article. This is precisely what I did for this article.
(In the previous approach, where AI is the Originator, AI does more work, but who cares about the computer’s workload? ideation is free with AI, so we shouldn’t hesitate to ask it for dozens or hundreds of initial ideas.)
This seems to be a less common approach because having AI be the refiner and not the originator demands more human labor. I rarely see this recommended — or even discussed — in articles about generative AI. Of course, it’s inherent in the very name “generative AI” that the expected workflow is to use AI as the originator and reserve human judgment for winnowing (picking the best of the AI’s profusion of ideas) and refinement (further improving that pick).
However, naming aside, having the human serve as the originator while using AI as a refiner works perfectly. I feel that it often produces more insightful results.
AI Origination is Faster, AI Refinement is Better (For Now)
My conclusion is that in today’s world, you usually save time (which is money) from the first approach, where AI serves as the originator. But you get better results (at the price of more human effort, which is expensive) from the second approach, where AI functions as the refiner.
Since the second approach is more expensive, you should reserve it for cases where you’re striving for the highest quality. While we all like to pretend that quality is essential to our business, the truth is that medium quality may often be all the client or customer is willing to pay for. I’m not condemning people who use the first approach. I use it myself quite often.
In the future, when we get the better AI promised by the First Law of AI, I can see the balance between AI as originator vs. refiner change, with more emphasis on origination and broader exploration of the design space. (Any creation is a form of design, including writing an article like this one.)
Alternate Conclusions by ChatGPT
I wrote the above conclusion myself (honest!). I had a few minor refinements from AI. But as an experiment, I also asked ChatGPT for more drastic improvements. See if you like any of these better.
This exercise demonstrates that using AI as the refiner doesn’t mean the end of ideation. My main beef with the “double diamond” design model is that the standard Internet visualization where the two diamonds touch in a single handoff point of contact is wrong. As we proceed through the “converge” stage of design (or the equivalent “refine” stage of writing), creativity ignites, inspired by our intimate immersion in the concreteness of embodying our vision in a design or prose. Facts rule; details inspire.
The Virtue of AI: A Discussion Between Two Minds
Socrates: What do you make of AI as an initiator versus a refiner?
Student: As an initiator, AI saves us time and, thereby, money. As a refiner, it demands more from us but gives back in quality.
Socrates: And which should one opt for?
Student: The latter is costly and should be reserved for endeavors of utmost importance.
Socrates: Ah, so the future may indeed alter this dichotomy?
Student: Precisely, advancements may tip the scale towards origination, enriching the realm of design.
🤖AI Origin vs. Refine🛠️
1/ Quick take: AI starts = fast & cheap
2/ AI refines = better but $$
3/ High quality? Go for the 2nd!
4/ Future AI? Game-changer for origination! #AILaws
AI's the starter, yeah it's faster,
But if you want a masterpiece, be the master.
Cash rules, quality's a chooser,
Pay less; you might just be the loser.
Future's near, AI's getting bolder,
Shiftin' gears from the shoulder to the holder.
WHEREAS, the party of the first part recognizes the utility of Artificial Intelligence (AI) as a preliminary tool for expediting projects at a reduced financial burden; and
WHEREAS, the party of the second part advocates for the utilization of AI in the refinement phase, acknowledging the elevated costs associated with superior quality;
NOW, THEREFORE, it is recommended that the selection between the two approaches be contingent upon the specific quality requirements and financial constraints of the project. Moreover, subsequent advancements in AI technology may necessitate a reevaluation of this advisory.
AI: Quick Starter or Polished Finisher? What's Your Flavor, and What's it Costing You?
Look, if you're in a hurry, let AI kick things off. But if you're after top-notch results, be ready to shell out for some fine-tuning. As AI gets smarter, it might just take over the whole show!
AI-Driven Origination Versus Refinement: A Comparative Analysis
The empirical evidence suggests that AI as an initiator yields time efficiency, thereby optimizing financial resources. Conversely, using AI for refinement tasks requires more human intervention and, consequently, higher costs. Thus, the latter strategy is advisable solely for projects demanding superior quality. Notably, the advent of more advanced AI, in accordance with the First Law of AI, is anticipated to shift the focus towards origination, expanding the scope of design possibilities.
Haiku (My Favorite for Brevity)
AI quick to start,
Yet refinement costs us more.
Future blurs the line.
More on AI UX
This article is part of a more extensive series I’m writing about the user experience of modern AI tools. Suggested reading order:
About the Author
Jakob Nielsen, Ph.D., is a usability pioneer with 40 years experience in UX. He founded the discount usability movement for fast and cheap iterative design, including heuristic evaluation and the 10 usability heuristics. He formulated the eponymous Jakob’s Law of the Internet User Experience. Named “the king of usability” by Internet Magazine, “the guru of Web page usability" by The New York Times, and “the next best thing to a true time machine” by USA Today. Before starting NN/g, Dr. Nielsen was a Sun Microsystems Distinguished Engineer and a Member of Research Staff at Bell Communications Research, the branch of Bell Labs owned by the Regional Bell Operating Companies. He is the author of 8 books, including Designing Web Usability: The Practice of Simplicity, Usability Engineering, and Multimedia and Hypertext: The Internet and Beyond. Dr. Nielsen holds 79 United States patents, mainly on making the Internet easier to use. He received the Lifetime Achievement Award for Human–Computer Interaction Practice from ACM SIGCHI.