Photo courtesy of Cassie Kozyrkov

How a high-level Google data scientist found success (and a fulfilling creative outlet) on Medium

An interview with Cassie Kozyrkov


Cassie Kozyrkov is the Chief Decision Scientist at Google. So what does that mean she does all day?

“Everyone asks me that!” she says.

“The truth is, what I do day-to-day varies a lot, and you’ll be disappointed to know what the one constant is. The true misfortune of human adult existence is no matter what your job is, most of it is email. And meetings. That’s what one does all day: email and meetings. But if you’d like to know what the theme of my work is, let me answer that instead. I spend my time doing two kinds of things. The first is helping important people make their biggest decisions. And the second is designing applied data science processes. I think of data science as the discipline of making data useful, but it doesn’t become useful all by itself. What I do is make data science useful. Decision making is turning information into hopefully better action. And information in electronic form is data, so both of these themes come together when a decision requires the support of a data science project.”

Kozyrkov grew up in South Africa, where she began her undergraduate studies in statistics at Nelson Mandela University before moving to the United States to complete a degree in economics at the University of Chicago. She then earned graduate degrees in mathematical statistics, psychology, and cognitive neuroscience from Duke University and North Carolina State University. A renowned artificial intelligence and data-science thought leader and frequent keynote speaker at technology conferences, she has been selected three times by LinkedIn as the #1 Top Voice in Data Science and Analytics.

Kozyrkov began posting on Medium in 2018, and currently has some 150,000 followers. I spoke to her about how she got started on Medium, why she writes on the platform, what advice she has for other Medium writers, and more. Our conversation has been lightly edited for length and clarity.

Jon Gluck: First things first. What got you started on Medium?

Cassie Kozyrkov: My first big keynote speech was at The Next Web, in Amsterdam, in 2018. And wow, everything went wrong. It was this massive “coliseum” stage, a giant round space with the audience all around you. You’re supposed to look for a few minutes in one direction, then gracefully turn about 90 degrees and stay focused there for a while. You’re not supposed to spin around and around like a dervish, which is precisely what I did. I was also under a tight time constraint, but to my horror technical difficulties delayed the arrival of my slides on screen by a few minutes. Talk about pressure!

After it was over, I wanted to do something that would make me feel better about the experience. Someone said, “Try this Medium thing,” so I made an account. I remember standing there, in the grassy garden area outside The Next Web, with my laptop. There wasn’t even a chair; I was at a standing table. But I took my script, polished it up a bit, added some pictures, and called it, “The Simplest Explanation of Machine Learning You’ll Ever Read.” I was amazed that so many people liked it. Then I noticed that some of the commenters said the way I used the term AI was technically wrong, so the next morning, while I was eating pancakes in Amsterdam, I wrote a response called, “Are You Using the Term AI Incorrectly?” From there, we were off to the races.

JG: Why do you write on Medium? What’s your motivation?

CK: People tend to see me as a statistics nerd or a corporate leadership type. And they’re not wrong. Once a statistician, always a statistician, I say — you never get the grumpy out. But in my heart of hearts, I’m an artist, and words are my art. Scripts for talks, blog posts, articles: that’s artistic expression for me.

JG: You don’t use Medium to build an audience or to make money?

CK: If I write a post and it doesn’t do well, I’m certainly disappointed that no one else saw what I saw in it. I’m thinking of two stories, for instance. One of them I polished for a month. I was like, This is a masterpiece. The other one I typed up in 45 minutes. When the 45 minute one eclipsed the one I thought was really thoughtful and useful and good, I was disappointed. But that’s not going to stop me. If I feel inspired to make another one like it, I will. The great artists didn’t care if their work sold, and I feel similarly. On my LinkedIn I’ve got over 500,000 followers. On Twitter, I’ve got something like 41,000. I put something on the former and get thousands of likes, then tweet something similar and get, like, five. You’d think I’d abandon Twitter, but I don’t. I put similar effort into them. I just enjoy sharing stuff with people.

JG: What do you like best about publishing on Medium?

CK: I like the way the platform feels. I’ve actually found myself drafting emails on Medium just because I just like Medium’s interface. And I like what Medium stands for, the decisions your team has made to nurture a respectable platform for knowledge. The easy path in social media is promoting the worst content, the cheapest, tackiest, lowest-effort stuff. That’s not what you get on Medium. You can actually find content you can build your brain with. I appreciate that, both as a reader and a writer.

JG: How do you come up with ideas for your posts?

CK: Almost everything I write is inspired by a situation where a human either asked me a question they were struggling with or a bunch of humans were having a conflict, often in a meeting, because they didn’t understand each other. I figure if a few people are having a problem, maybe others are, too.

And it’s okay if the problem is niche. Most people aren’t lying awake at night trying to figure out what the curse of dimensionality means, but someone asked me about it and I enjoyed being able to express it simply and straightforwardly so that a student of machine learning could feel it click.

JG: Do you have a favorite story you’ve posted on Medium?

CK: I think the one that probably has done the most good is The Ultimate Guide to Starting AI. The story focuses on the decision-maker’s perspective on ML/AI; it’s a reminder of how important the decision maker is in ML, how they need to be actively involved in guiding the work to a worthwhile purpose. Another one is called Forget the Robots, Here’s How AI Will Get You. It’s about why AI can be more dangerous than traditional software. AI can be a thoughtlessness enabler. That’s both why it’s awesome and that’s why it’s dangerous. It’s wonderful when it allows a person to get their personal project automated with less thought and effort but it’s dangerous when the project isn’t just personal. When it affects other people. Because if you allow people to be thoughtless at scale with impact on millions of lives, that’s a terrifying thing. Two more favorite articles for the nerds — which I am one of! Statistics for People in a Hurry and Explaining P-Values With Puppies. Both explain statistics concepts in a fun way.

JG: How did you learn to write the way you do?

CK: As an undergrad at the University of Chicago, I took a class with a writing emphasis where the professor did something amazing. He said, “I’m going to start everybody off with a C+. No matter what you hand in, your grade will be C+. Then you can rewrite this same piece as many times as you wish. In fact, you can keep going after the class is finished and I will retroactively adjust your grades. You can keep going up until your graduation week. But if you want that A, you’re going to have to work for it.”

At first, I tried rewriting this essay on my own, but I couldn’t get above a B-. But I’m a stubborn creature and I wasn’t going to allow my transcript to be blemished by anything that didn’t start with the first letter of the alphabet. (By the way, knowing what I know now, I’d probably intentionally aim for C+ grades and take more classes to learn more. I wouldn’t let a GPA get in the way of optimizing my learning. But that’s the kind of wisdom I had to wait a couple of decades to acquire.) In my quest for that A grade, I sought out writing tutors and learned how to structure a paragraph, plus some of the other basics of the craft of writing.

I learned that once you have the technical stuff down, you have to add your own voice, the spark that makes a reader recognize the artist behind the words.

After a lot of work and maybe a few dozen rewrites, I squeaked up to a B+. The class had finished by the time I learned that once you have the technical stuff down, you have to add your own voice, the spark that makes a reader recognize the artist behind the words. Aha, this one is Vladimir Nabokov and that one is Toni Morrison. It took me months after that to even start making inroads to an A-. I eventually got the A, but only after many, many rewrites.

If that professor is somehow reading this interview, I’d like to thank him for making me the writer I am today. He deserves an A+ at professoring, for sure (adjusted after graduation).

JG: Do you have any technology tricks you use to write?

CK: I’m not a natural writer. When I was in college, I did well in my math and science classes, but when I had to write essays, I really struggled. I would see that blank page, and it would terrify me. I’d start writing some opening sentences, then I’d look at them and know they were awful. Then instead of moving ahead, I would start polishing those two sentences. By the time the sentences were perfect, I’d already have forgotten where I was going with them and found myself stuck. Classic writers’ block.

At one point, when I was complaining about how hard writing is to my friends, I realized I could tell them what I wanted to say without a problem. Now I use speech-to-text for first drafts. It’s the perfect way to tie my hands behind my back so that I’m physically unable to polish my first sentences, beating the dreaded Blank Page Syndrome.

An unemcumbered ramble into a mic gives me a big chunk of text that I can rearrange and remix. From there, I can produce my first real draft. But we’re not done. I wish we were, but there’s still quite a lot of polish that needs to happen. I smooth out the rough edges and add flair. Then I sleep on it for a while — a day, a week, maybe even a month. When I come back to it, I aggressively take a machete to it and do the final improvements. The finishing touches involve writing the intro (yes, many writers leave that for the end), creating a title, and illustrating the piece. I used to illustrate it with Pixebay, now I use Midjourney. Midjourney is awesome.

JG: Do you use AI?

CK: Hah! Well, speech-to-text is already AI. Good, old-fashioned, traditional AI. The kind of AI people aren’t even excited by these days. It’s amazing what people will take for granted when it comes to technological progress. Speech-to-text AI is such a core part of my process that I can’t imagine blogging without it. And of course I use AI for the illustrations, since that’s what Midjourney is: a generative AI art system.

But that’s not what you’re really asking me, is it? You want to know if I use a large language model like Bard or ChatGPT. And the answer is yes: I use them three ways. The first is to clean up the grammar in the speech-to-text transcript. I add a line to my prompt insisting that the LLM must not change any of the words, it should just do grammar cleanup. Super useful since my favorite speech-to-text software skips punctuation.

The second way I use LLMs is to help me make a catchy title. I feed in my finished blog post and ask for ten title and subtitle suggestions. Then I ignore them all and write my own, a task which is inexplicably easier once there’s a few suggestions to roll my eyes at.

And as the wonderful third way I use LLMs… I use them to get angry. One of the easiest ways to be inspired to rant into a microphone is in response to Someone Being Wrong On the Internet. Alas, dealing with fools in the flesh always leaves a bad taste in everyone’s mouth. I prefer not to have to have beef with any particular human being. So I simply ask an LLM to write a blog post in my own style on my topic and then I get to enjoy just how offended it makes me, both because the writing is subpar and because some nuance of the topic has invariably been flushed down the toilet. That gentle rage is a great muse for my own — much better — article on the topic.

The moral of the story is that there are many creative ways to use AI that don’t involve brainless, hands-off plagiarism. Global bans on AI seem pretty silly; to stay consistent, those folks ought to ban computers, too. And pens. All these things are just tools that reduce time spent on everything that’s not fundamentally creative. I don’t want to have to make my own papyrus just to be able to put some thoughts down. What I believe publications should penalize isn’t AI use but laziness in all its forms. If a content creator makes lazy content — whether it’s generic garbage automatically churned out by AI or simply old-fashioned lazy writing — they show a lack of respect for their audience. I’m all for publications banning disrespectful writers so that the voices who are genuinely trying to serve their audiences don’t get drowned out in a deluge of rubbish.

JG: You mentioned story titles and using AI to help generate them. How do you ultimately decide what they should be?

CK: I don’t write stories that are clickbaity, but at the same time, I want people to read my writing. So I compromise by making my titles a little clickbaity.

JG: Do you post on a regular schedule?

CK: Here I have some guilt, I guess, because I’d like to post once a week. I used to sit down very strictly, on a Sunday, as part of the whole ethos of forcing yourself to do a little bit of work consistently. Tackle the work in little nibbles and then, before you know it, you’ll have eaten the elephant. Now what I do, to be honest, is cram. I try to write as many pieces as I can in a short period of time, then I hold some of them back to post later. I wish I could say that I just still sit down and write every Sunday, but my schedule is overflowing these days.

JG: How do you promote your stories?

CK: I should really step up my game and be more intentional about this, but I’m not. I write when I write. Then I tweet about it, I LinkedIn it, and go to bed. I don’t even use those tweet scheduling services. I tried one, but didn’t stick with it. I just haven’t been doing any of those professional influencer things. With my build-it-and-they-will-come attitude, I’m not sure why anyone came at all, but somehow it’s worked for me. It’s far more than I deserve, and I’m very grateful to my community.

JG: In addition to writing on Medium, you’ve used the platform to organize some of your ideas into what you’ve called learning paths. What’s that about?

CK: To help my readers sort through the different topics I write about, I added standardized supertitles to my articles. I’m sure it’s confusing for people who come for my leadership topics to suddenly stumble across a random post about philosophy or, worse, something with equations in it. Now I’m trying to make clear channels for readers to choose from.

JG: You wrote in a recent tweet that your Medium following is 70-percent as big as Barack Obama’s. What do you think draws people to your posts?

CK: As I said earlier, I try to untangle knots, for at least one person and hopefully for more, I put a lot of thought and work into each post, and I try to include an element of irreverence. There are situations in life where reverence is appropriate, but I can’t say I’ve found many of them in data science. A little humor can go a long way.

JG: What are some of the most memorable reader comments you’ve gotten?

CK: In a few cases, people have told me that they started their data science career because of my writing. Others have said they were inspired to take on data science leadership, instead of trying to progress by doing more and more sophisticated technical things. I guess I helped them find the courage to step away from developing the next algorithm and start leading in this space, asking whether that kind of algorithm is needed in the first place. I’m proud of having played a role in that.

JG: Are there people or publications on Medium you would recommend to other people?

CK: I have a very particular take on finding information. I don’t subscribe to publications or people or even read the news as it comes at me. I know that’s a shocking thing to say, but hear me out. None of those sources are designed to have my best interests at heart. They may be benign or they may be trying to get attention by any means, but they’re certainly not customized to what I need to know about. I prefer to get my information proactively. So the way I interact with news and publications, including Medium, is I ask myself, What do I actually want to learn about? From there, I look for keywords I’m interested in. Sometimes a post I find might be from a writer who’s just starting out on the platform and has no following at all, but why should that put me off? If that person does a great job of explaining things, I’m really grateful to them.

As for news, I have a very carefully curated list of keywords which ensures I catch information about global events that affect my future as well as niche news I’d like to be informed of. When I was a college student, I sat down with a few newspapers and audited the fraction of articles that were real news, in the sense that I’d be worse off a month later if I had missed that news story. I still occasionally go through that exercise with my digital news sources. I encourage you to do the same with whatever your news source is. Chances are you don’t have the time to have your time wasted and you might like to use tools to help you block the attention-grabbing stuff with no nutritional value so you can focus on the news that’s actually news.

JG: Do you have any advice for other Medium writers?

CK: If you try to judge your worth by views, you’ll be playing a losing game. A lot of view counts are dependent on rare events, like someone with a big Reddit account talking about your stuff there. That’s a lucky event that you can’t count on the next time you publish. You might publish a masterpiece next and maybe that masterpiece only gets a few hundred views. Don’t take it personally; view counts tell you very little about the quality of your ideas.

If you’re writing for the joy of self expression and the delight of delighting those who get hooked on your stories… then you’ll have the happiest possible time on this and any other platform.

If you’re writing purely for business reasons, then you’re probably going to play all the optimization games where view count is everything. I kind of wish you wouldn’t because you’re making writing worse for all of us. Unfortunately, the actions that get you more views are orthogonal to creating prose you can be proud of.

But if you’re writing for the joy of self expression and the delight of delighting those who get hooked on your stories, simply because it delights you to make other people happy (but you don’t let yourself be too bothered if they’re not), then you’ll have the happiest possible time on this and any other platform. Whatever view count optimizing you’re planning to do, do it in how you pitch your piece elsewhere, not in the act of creating it.



Jon Gluck
The Medium Blog

Jon Gluck is the Editorial Director of Special Projects at Medium. Previously, he held senior editorial positions at New York Magazine, Vogue, and Hearst.