If you are confused about how to cooperate with AI without harming our own cognitive potential, this essay is for you.
For the very first time, it feels like I am not the protagonist of my own task.
November 30, 2022. I missed ChatGPT’s launch. I was too busy with the company blog website, translating the design from Figma to React components. This is a very sophisticated task, and it demands high concentration for a long period of time.
My working routine is coupled with a Pomodoro clock, which means I have 45 minutes of concentration time. I set up the clock and immediately get inside the zone. When I finish the clock and return to the world, I am with some sort of progression, and clock by clock, the results begin to show themselves. I use the same system for all my writing, 45 minutes clock, full concentration.
That might be the reason why I can enjoy this boring and detail oriented process which is disliked by many. For me, it’s a very rewarding experience and I usually work in a batch until midnight when everything goes quieter. Same as today, when ChatGPT launched, I finished the first version of the blog website. Sitting in front of the bright monitor, shining, flicking, the blog is alive, and I am proud of myself.
Fast forward three years. Last night I was building a new analytics dashboard. I grant AI access to the Figma file, and ask it to generate the react components, then wire up with our backend layer within a single Pomodoro session. This task previously will cost me more than three Pomodoro sessions.
Once I finish the task, I sit in front of the desk and look at the result. I can’t feel the same fulfillment I had before, it’s just too fast and too easy, there is no struggle in it.
I felt emptiness. The speed of AI disrupts my rewarding system.
And I begin to wonder, is it because I am better at my job so I can finish it at such a fast pace, or is it due to AI being far better than before. How can I distinguish between my personal growth and external technology advancement? The fine line between these two is diluted.
The usage and the advancement of AI also pose a great question which is hard to answer:
How can I work with AI in a way that won’t dilute our skill and cause us to be replaced like a spare part.
Ira Glass, host of “ This American Life”, has a famous quote about creative work:
Nobody tells this to people who are beginners, I wish someone told me. All of us who do creative work, we get into it because we have good taste. But there is this gap. For the first couple years you make stuff, it’s just not that good. It’s trying to be good, it has potential, but it’s not. But your taste, the thing that got you into the game, is still killer. And your taste is why your work disappoints you.
A lot of people never get past this phase, they quit. Most people I know who do interesting, creative work went through years of this. ... Put yourself on a deadline so that every week you will finish one story. It is only by going through a volume of work that you will close that gap, and your work will be as good as your ambitions.
In his words, the progression of creative work can be simplified as:
Your taste inspires you to create
You found there is a gap between your creation and taste
Through endless practice, you have the chance to close the gap
The use of AI breaks this struggle and uses it as the selling point. It might work for routine work, but it poses a huge risk to our creative work. It simplifies the implementation, with the mediocre prompt you can already get good results. And with dozens of adjustments, you might be able to match a temporary ideal state of your idea.
There is a high chance that creators stop here, not pushing forward. They mistreat the generation as their own creation. The result might be good enough for their purpose, but they miss the most important journey, staying with the discomfort within our creative heart, struggling to put together a version where the implementation matches our taste.
A 2024 study titled ‘Your Brain on ChatGPT’1 tested 54 participants writing essays with LLMs, search engine or brain-only and they found out that: People who use LLM to write the essay have less ownership of their creation.
They can’t even remember what they wrote.
And judged by real humans, they tend to be considered writing poorer than the brain-only and search-engine group.
The brain-only group not just performs better, they have more reflection and improve better in the Brain-To-AI session (Where participants discuss their essay with AI), they found more insight in the interaction between their essay and the AI response.
This also matches my observation: my essay performed better when I write the first draft with my brain only, without any generation, and then use the AI to criticize and edit it together. I can grant many different perspectives without involving my friends as proofreaders.
In our time, we might not be able to fully comprehend the consequences of using AI in our creative work. But living in the center of the storm, it’s also our privilege to feel the reward and pain with our body before any conclusion is made. After 18 months of exploration, millions of tokens used, 15 newsletters published, I now have two pillars, three methodologies as a tool to explore “How to work with AI.”
Those pillars are the thinking process that I am currently practicing; they are just exploration and not rules.
Taste is the reason why our creation feels different then others. It is also the source of new innovation, inspiration and thinking. And it’s the reason why we can’t be easily replaced.
Jiro Ono2, the world’s greatest sushi chef, can tell the difference between a well message octopus and the single digit temperature difference of rice without metering it. He asked one of his apprentices to practice tamago for four months before approving it. The technique is not the most important thing here, the taste to tell the difference is.
Ansel Adams3 took 40000 photos in his life and considered only 200-300 photos to be truly great, that is a 99.5% “failure rate.” But those 39,700 “failed” photos weren’t wasted. Each one taught his eye what “almost right” looks like, so he could recognize “perfect” when it appeared.
Taste is what enabled their greatness. And taste develops through struggle, through the gap between vision and execution, repeated thousands of times.
So when adapting any kind of AI tool into our working process, the first thing is to try to ask, “Will this stop me from improving my taste?”
Think about cooking. A trained chef can use MSG strategically, they know what is the role of this pinch of MSG, it amplifies what’s already there. And the reason why they can do it is due to years of tasting, adjusting and struggling, they know what MSG do to their palate.
But the reliance on MSG for a home cook will make them lose the natural feeling toward any kind of umami, and it will stop them from improving their skill to create a diverse and rich flavor using natural products.
AI is MSG for thinking. It delivers “good enough” instantly. But you never develop the palate to distinguish good from truly excellent. You never learn what you’re amplifying.
There is a litmus test to examine whether you are using an AI tool in a way that will sabotage your thinking, taste development, and creativity. I called it the Copy and Paste Litmus test, asking these questions to yourself:
Do you copy and paste the generated AI content into your own creation?
Do you abandon your ownership and argue with people about a topic with the AI-generated content as the proof? Like saying “I think it’s true since AI answered this way.”
With the face of any struggle, what is your first action? Is it asking AI for help or trying to think of a solution by yourself first?
The idea behind these questions is whether you use AI content blindly without a second thought, which not only removes your opportunity to improve but also stops you from struggling with yourself like we mentioned earlier, considering it is the core development of one’s taste.
We need to be extra cautious around this kind of adaptation if “Originality” is a merit we care about.
The brain is a very sophisticated organ, and if we look closely, it’s also a lazy one. The thing we practice every day will become our inner strength. But the things that we didn’t practice will soon be forgotten.
The ability to think and work deeply is also this kind of skill. And they are more scarce than we think, especially in this era. It’s hard to stay focused and practice those tiny, boring techniques day after day.
You might ask what is the difference from other cognitive risks we face in this era like social media, the bloom of short content, and search engines if we extend the time horizon to even older periods.
Social media and short content fragment your attention, with it you can still think, but you are distracted from deep work. Search engines outsource memory, but you still know where to find the information and you still need to think it through. But AI outsources judgement and creation, the core acts of thinking. You are not just distracted; you’re bypassed.
When use the AI tool, don’t forget to ask ourselves, whether this will stop us from improving our cognitive ability, if it amplifies or replaces it?
To utilize these two pillars, I developed three methodologies to further explore the ability to work together with AI.
Usually, my writing can be simplified to three parts:
Research -> Write -> Rewrite/Criticize
Research includes reading and generation. I didn’t have this structure at the beginning of writing this newsletter, but after the sheer amount of the generation, discussion and practice, I found out instead of asking AI to directly generate the content, they are better at generating the reference points. And we become the consumer of those reference points.
Take my workflow for example, first I will lay down a short brief of what I am going to write (200-300 words). And then I will ask AI to ask me as many questions as possible. They are very good at asking good questions, and I found lots of my insight during answering them.
After this process, I will have a knowledge base filled with questions and answers which can serve as the foundation of further exploration both for AI and me.
And then I will first draft the outline and ask AI to criticize it. After several rounds of adjustments, I will settle on a version and proceed with writing the first draft.
When I get into the writing zone, there is no more involvement of AI. I always write the first version of a draft and the first time rewriting brain-only. There is no workaround, just me, keyboard, and my words.
In this way, I can make sure that no matter how much I dive into the generation world with AI, I will always have a fine line between my own creation and generation. All my work starts from me, not from the AI. This is a very crucial first step, it not only gives me the opportunity to polish the skill, the taste, and the thinking process, it also serves as a self-reflection and reconfirmation:
I am the owner of my work, and I take full responsibility for it.
After the first draft and the first rewrite, I will enter the multiple rewrite phase, which involves the use of the AI. But I didn’t ask AI to directly clean up my essay; instead, I asked it to review and criticize my essay with all the information it has. Including the Q&A section, the outlines I drew, and the thinking process I added on my drawing board.
AI is very good at this; they can think through the mist clearly, and I can even ask AI to play a role like the famous writer I admire, using their tone and perspective to criticize my work. This gives me more perspectives, which can only be obtained if you have many draft readers.
But this is also a double-edged sword since the ability to criticize one’s creation is also the foundation of one’s taste. That is the reason why, alongside this method, I also developed a technique called the blind critique test.
I still remember a teacher saying in one of the historical classes in college. He is very knowledgeable about the ancient dynasty in China: Qin. Especially the first emperor of China, Qin Shi Huang.
Qin Shi Huang is a very powerful figure in Chinese history, and he went through many dramatic events in his life. My teacher always said, the good way to read history is like playing a detective game. You are the detector, when you read a crucial point of the history, and have all the knowledge at that moment, you should close the book and ask “What if you are this character, what will you do?”
Why, I remembered one student asked the teacher why we need to operate like this. What is the benefit here? The teacher stared directly at that student and said, “So you are the participant right now, and you can finally feel the emotion and really understand why the figure does this kind of action at that moment. Until then, you are just an observer, and who only observes history, doesn’t really learn.”
That is the key to this step too. Right before the AI generates its first criticism, you should first criticize your own work, with a step further, asking yourself, what will AI criticize this? Like what my teacher said, who only rely on AI to criticize their work without stepping in earlier to criticize themself, miss the opportunity to really learn from it.
The show is better than the say, so I decided to share all the details of how I work with AI with you. Below are multiple documents, unsanitized, no cleaning up, directly exposed to the public domain for you to examine. There are two very important pieces which you might find interesting: the Q&A section and the criticism section. You can also find the corresponding prompt in those documents.
Take Q&A for example. I asked AI to interrogate me about taste, cognition, and methodology. And I answered them brain-only.
Q: Your litmus test is “do you copy-paste from LLM.” But what about someone who reads AI output, closes it, and then rewrites it in their own words from memory? Are they a pipe or a medium? Where’s the line between “inspired by” and “outsourced to”
A: I will argue that, if the person remember it and rewrite it with their own word, that is already a mental burden, in a positive way. In this way, they already becomes a medium if they say the line in their own words, their own interpretation.
Can you use AI for 10 years and still develop original thinking?
This is the question that keeps haunting me through this journey. We are living in an era where originality becomes a must-have, and working with AI becomes an unstable factor in the equation.
What if it greatly reduces my chance of developing my own idea? What if it reduces my cognitive ability? What if it makes me lazy? All these questions remain unanswered for me. And I think they will still be there years later.
But I will still work with AI, that is the tipping point of our generation, and I want to take part in it. Observe and really understand what this technology stands for. Maybe in the far future, generative AI will just be a road sign to abandoned cities. But at least at this point. It inspired me more than anything else.
I will keep writing and use this 10-year journey as the testing stone for myself to really understand the meaning of this technology. And how it changes my writing and thinking.
Thanks Lucy and Shaka for reading the draft of this essay.