---

Dear Reader,

In the early 2000s, I worked as an early humanoid version of ChatGPT. Well, sort of.

It was still the comparatively early days of the internet, and a friend and I had launched a publication called “KnowledgeNews.”

Our goal: Deliver the history, science, and culture readers needed to understand the day’s headlines. When the news was about Iran, we delivered a 1000-word summary of 100 years of Iranian history. When the news was about censorship debates in America, we covered the creation of the First Amendment. When the news was about nanotechnology (remember that?) we delivered an analysis that made tiny tech more imaginable.

My job: Find sufficiently reliable sources of information quickly, summarize what they said effectively, and provide a pleasing response to our readers’ questions—before the news cycle shifted too much.

I was, in effect, a human ChatGPT.

My sources were often online. I got very good at using Google (before googling was a verb) and verifying with Yahoo! and Microsoft’s Encarta online. Wikipedia was still in its infancy, so I used old-fashioned print encyclopedias. I also used my trusty public library.

These often provided the raw information I needed. For additional insights, I used telephone and email outreach. When we wrote about the polar ice caps melting, I found a phone number for a researcher in Nova Scotia and called him out of the blue.

“Hello,” I said. “I’m a writer for an online newsletter in the United States, and I want to understand how ice caps work.”

He spent over an hour talking to me, freely explaining his life’s work to a novice who was curious and wanted to understand what an actual ice expert knew.

I had a similar experience when I wrote about a volcanic eruption in Hawaii. A scientist in Kauai offered to let me stay at his home if I wanted to come see a volcanic island for myself.

It was a wonderful job. Literally. I scrambled to learn things quickly and often found myself filled with wonder. Digging as fast as I could into ice caps, I found layers of complexity that extended back to the Ice Age and down to the bottom of the periodic table. I quickly realized that 100 years in Iran was a tiny fraction of Persian history, and all I could touch in my timeline was the barest of basic facts.

To capture and quickly channel the knowledge it would take to truly understand such topics was a practically impossible task. The knowledge was much too big, and the aperture was much too small. The best I could hope to do was to help people find a way in, some story or framework or schema that helped them see a little bit more.

Every article I shared had a link at the bottom that said “Want to learn more? Click here.” And each of the links I shared was carefully chosen to widen the aperture I’d tried to open a little further, to launch my reader on a journey of discovery akin to my own.

Part of me was always surprised that people found our little service valuable, but they did. Twenty years before the birth of Substack, we had an email newsletter with 25,000 paying members.

“We tell people we deliver the knowledge behind the news,” I said to my partner at one point. “But what we actually do is use the news to try to get people to think.”

“Good,” he replied. “Thinking needs a marketing department.”

That’s more true now than ever, especially since encyclopedic knowledge has been repackaged as AI.

KNOWLEDGE IS NOT WHAT YOU THINK

As an early form of artificial intelligence, my powers were admittedly limited. I was (and am) built exclusively on the messy processing architecture human evolution left behind, and I have to eat, drink, breathe, and sleep before I can even start to think.

Once I do, my thinking habits include inherited biases and learned ignorance as well as areas of comparative expertise. I studied Shakespeare in grad school, taught Rhetoric to college freshmen, and helped start tech and marketing companies. That taught me nothing about nanotechnology or Iranian history.

Benchmarked against the breadth and speed of today’s AIs, I was a bad human joke. But when it came to delivering on KnowledgeNews’s overt promise—delivering needed snippets of knowledge in real time—my personal limitations weren’t the main challenge.

Here’s what I learned while dealing daily with the problem of delivering knowledge in a timely and pleasing package.

1. There isn’t one “knowledge,” and there never was.

Knowledge moves constantly. Ideas get added and subtracted by the day, the hour, the minute. The sources are always changing. More importantly, they always disagree about important points. Not just because of politics or particular biases, but because they’re the products of fallible humans using language, stories, and numbers to try to make sense of the world.

Experts disagree about how to catalogue species, how to conceptualize quarks, which historical events were most important, how language takes root in human brains and shapes subsequent thoughts, and on and on. I have an encyclopedia on my shelf that belonged to my great-grandfather. Most of it is now misleading if not dead wrong.

If you think we’re standing at the end of knowledge, where anyone can simply look up the truth, you’re mistaken.

The body of existing knowledge in every field is actually a complex set of intersecting conversations, some of which have become more settled than others. What I typically wound up doing amounted to reporting on what was comparatively settled and finding a middle path through remaining controversies.

2. Good maps are better than vast stores of facts.

My personal processing ability varied based on my existing skills and experience. But I realized something interesting and unexpected along the way: the places I knew best weren’t always the ones where I made the best guide.

Sometimes a newcomer gives better directions, for roughly the same reason that a simpler map is often better than a detailed one. A map’s purpose is to get you where you need to go, not to provide a perfect picture of your surroundings.

During my time at KnowledeNews, my partner and I became crystal clear on this. We realized that we couldn’t begin to provide comprehensive knowledge of the topics we were taking on—the importance of free speech, the ancient history of Iran, the potential implications of nanotechnology.

We talked daily about the “schema”—the literal maps, the metaphorical blueprints and framing perspectives—we could provide that would help our readers start to get their heads around such complex topics.

The human mind—mine, yours, anyone’s—is more like a filter than a container. To provide a grounded, intelligent, usable schema that enables minds to better sort information and understand the world is valuable. That’s what our subscribers were willing to pay us for. It wasn’t “the knowledge behind the news” so much as “the schema to better make sense of it.”

3. There isn’t enough time in the day.

The hardest problem I faced wasn’t about me or the knowledge I was trying to make sense of. It was the size of the aperture into which I had to fit what I found—in my case, an email newsletter.

Even 20 years ago, people were drowning in information. What they needed weren’t more facts or “takes” but tools to help them think better—to perceive parts of reality that were otherwise invisible, to reimagine what they’d only seen one way, to answer questions they hadn’t yet posed for themselves. And they needed it all delivered through an aperture that was always closing fast.

AI claims to be able to replace the encyclopedia, the library, and even the World Wide Web—to be faster and more comprehensive than any human could ever be. But it can’t and doesn’t solve for the fact that knowledge keeps moving and changing. It can’t and doesn’t solve for the fact of its own processing limitations and biases, much less for the ones inherent in my brain and yours. And it certainly can’t and doesn’t solve for the fact that time and attention are limited and perpetually vanishing—that we can’t ever hope to open our minds wide enough to take in all of the context we would need to fully understand.

4. Better thinking is as good as it gets.

Knowledge is hard to pin down. Thankfully, thinking is a joyful and worthwhile journey, filled with discovery, creativity, and wonder. Even when you find an answer, it invariably leads to more questions. And that, too, is wonderful.

The world moves, and people figure out new things, and I am able to move with it and share what they learn—to learn from them and to teach what I learn to others—because I keep thinking.

Also, we do our best thinking together. The exchanges through which we share knowledge, ask good faith questions, listen carefully, and learn new things are powerful and precious. They are us at our best and brightest. They are how we come to understand better.

THINKING IS BETTER THAN KNOWING

The promise today is that AI can do your knowing for you, that it has all the facts you need and can supply them whenever prompted.

But that’s the ruse, folks.

The point was never really to become all knowing. We don’t have the time, the attention spans, or the energy for that. And it’s not clear what that even means—because once you start to question, there’s always another question. Once you start to wonder, there’s always a further wonder.

At some point you start to realize it’s wonders all the way down. And if you don’t then you are missing the larger and more important point.

The point is not to capture it all and squirrel it all away, to permanently clarify, classify, digitize, and monetize. The point is to learn how to think together, to solve interesting puzzles, discover new wonders, and create better lives.

The end goal isn’t knowledge-as-power, it’s learning as liberation. The point is to cultivate the capacity—and the freedom—to think, explore, discover, create, and be. The point is also, necessarily, to create the cognitive and cultural connectivity to do all of those things together with other people.

The lesson I learned while trying to be ChatGPT is that knowledge isn’t something I could hope to deliver, unspoiled, in easily consumable packets. All I could do was gesture and frame and help people find a way in to the ongoing conversations and explorations.

For breadth and depth and speed, I was no match for ChatGPT. All I could do was wonder and help people discover. All I could do was learn and teach, make sense and share. But when I did that well, I was arguably better than any artificial intelligence.

We all are. We all can be.