How does one…
 |
| Gemini's view of how the CRISPR process works to edit DNA in vivo. |
… understand the really big, complex, fascinating topics?
If you’ve been following my explorations here on the blog for a while, you know that my obsession has always been sensemaking—the process by which we gather fragments of information and stitch them together into a coherent mental model.
And for decades, the dominant paradigm for online sensemaking was keyword search with fancy notetaking: we’d type a few terms, sift through a list of blue links, read the documents, and synthesize the answers in our own heads, maybe using some tools on the side.
But that paradigm is shifting. We are no longer just navigating to information; we are working with and using agentic AI systems and large language models as interactive partners in the synthesis process itself.
Recently, I’ve been thinking a lot about how we use these tools not just to find quick facts, but to learn deeply complex, highly technical concepts that sit far outside our existing domains of expertise. How do you learn something when you don't even know enough to understand the vocabulary?
To explore this, let’s look at one of the most consequential scientific breakthroughs of the 21st century: the CRISPR-Cas9 gene-editing system. I’ve heard about this for a while and I have a vague understanding of it—but what would be the best way to learn the details?
The Wall of Jargon: The CRISPR Paper
Suppose you decide you want to truly understand how CRISPR works. You might do the academically rigorous thing and pull up the seminal 2012 Science paper by Jennifer Doudna and Emmanuelle Charpentier: "A Programmable Dual-RNA–Guided DNA Endonuclease in Adaptive Bacterial Immunity."
You read the abstract and hit a brick wall. The language is full terms like crRNA, tracrRNA, endonuclease, protospacer adjacent motif (PAM), and target-site recognition.
If you aren't a molecular biologist, your working memory is instantly overloaded. In the old days, you might have opened twenty browser tabs, individually searching for the definition of each term, trying to hold them all in your head while reading the original sentence.
It’s a frustrating, high-friction process that often leads to cognitive fatigue and abandonment. What can we do to make this better?
The Dialogic Method: Breaking Down Complexity with AI
This is where AI changes the game. Instead of treating an LLM like a glorified encyclopedia, we should treat it like a tireless, infinitely patient tutor. The key is to break the complex idea into interactive, iterative components.
Here are three distinct ways to use AI to penetrate a complex paper like the CRISPR publication:
1. The "Translate and Map" Technique: Instead of asking the AI to "Explain CRISPR," which will just yield a generic Wikipedia-style summary, give the AI the specific, dense text that is blocking you.
Prompt: [I am reading a biology paper. I have a strong background in computer science but no background in molecular biology. Please translate this specific paragraph about tracrRNA and crRNA into a systems-architecture analogy. What are the 'software' instructions, and what is the 'hardware' execution?" ]
By anchoring the AI to your existing mental models, it builds a cognitive bridge. Suddenly, the guide RNA isn't abstract biology; it's a search string. The Cas9 protein is a text-editor executing a "find and delete" function.
2. The Socratic Deconstruction When you encounter a complex mechanism, ask the AI to walk you through it step-by-step, but instruct it to pause and wait for your confirmation before moving to the next step.
Prompt: [ Explain the exact sequence of events when Cas9 binds to DNA. Explain only step one, then wait for me to summarize it back to you before you explain step two. Correct any misunderstandings I have.]
This turns passive reading into active, conversational learning. You are forced to interact with the material, ensuring you actually grasp the base mechanics before layering on more complexity.
3. Parameter Adjusting (The "Dial-a-Level" Approach) If an explanation is still too dense, or perhaps too simple, you can adjust the "resolution" of the information.
Prompt: [ You just explained the PAM sequence at a high-school biology level. Now, dial the complexity up to an undergraduate genetics level. Introduce the specific chemical or structural reasons why the Cas9 protein requires this exact sequence to initiate a cut. ]
Epistemic Humility: The Illusion of Knowing
BUT…here’s the most critical part of this new learning method—and the most dangerous trap of the AI era.
When you use an AI to unpack the CRISPR paper using brilliant analogies and perfectly tailored summaries, you will experience a rapid, intoxicating rush of comprehension. Because the AI's prose is so fluent, and because it has successfully connected the concept to your existing knowledge, you will feel a profound sense of mastery.
Do not trust this feeling.
Cognitive psychologists call this the Illusion of Explanatory Depth.
We frequently overestimate our understanding of complex systems. We think we know how a bicycle or a zipper works until we are handed a piece of paper and asked to draw the mechanism from scratch. Then, suddenly, you don’t know anything. AI exacerbates this illusion. When the AI does the heavy lifting of synthesis, it feels like your brain did the work.
To be an effective learner, you must practice epistemic humility.
You must acutely understand the boundaries of your knowledge. Just because you understand the computer-science analogy of CRISPR does not mean you understand the biochemical realities of it. You know the map, but you do not know the territory.
So... How do you establish where your knowledge actually ends?
One way is to force the AI to test your boundaries with you.
The Boundary Test Prompt: Here’s a test prompt for you…
Prompt: [I believe I understand how CRISPR uses guide RNA and the PAM sequence to target and cut specific DNA. I am going to explain it to you in my own words. After I do, I want you to do three things:1. Tell me where my explanation is factually incorrect.2. Tell me what critical nuance I am missing.3. Ask me one challenging question about a scenario where this system might fail or behave unexpectedly, to test if I actually understand the underlying principles.
]
Now... Close those windows you have open with the descriptions of the mechanism.
Then, when you try to articulate the concept to an empty room, you will quickly discover the gaps in your mental model. The places where you stumble, use vague words like "stuff," or hand-wave a mechanism—those are the boundaries of your knowledge.
Stay on this cycle of question-asking and explaining until you’ve reached the saturation point. You’ll know it when you get there.
The Future of Sensemaking
We are moving away from an era of searching for documents and into an era of searching for understanding. The tools at our disposal are miraculous. They allow us to peer into disciplines—like molecular genetics, quantum computing, or macroeconomics—that were previously walled off by decades of specialized jargon.
But using these tools requires a new kind of discipline. We must become expert prompt engineers, not just to get the AI to generate text, but to guide our own cognitive processes. We can use AI to deconstruct the inscrutable, but we must also use it to ruthlessly interrogate our own assumptions. Beware the Illusion of Understanding!
The goal isn't to know everything. The goal is to deeply explore the complex, expand our mental models, and—perhaps most importantly—retain the wisdom to know exactly what we still don't know.
Keep searching. Keep researching.
No comments:
Post a Comment