No. 68 — They All Saw a Cat ⫶ How Big is YouTube? ⫶ Moonbound
and the computer saw A CAT
My name is Linda. I write a bi-weekly newsletter about computer science, childhood, and culture.
Last fall, during toddler time at the American Library in Paris, the librarian read a book called They All Saw a Cat by Brendan Wenzel. In it, an inquisitive cat wonders in the world. We see it through the eyes of a dog, a flea, a child, and many, many more.
I’ve mentioned the book in passing, but looking at the captivated crowd made me want to get it for our home collection, too. It’s a classic. (Here’s a readout if you are curious.)
And, like best books, it sparked a lot of thoughts. Many companion activities teach students perspective-taking and building empathy, but I want to use this book to do something different.
With the emergence of AI, we need new vocabulary, representations, and similes of what we talk about. Is AI like a JPG? A Stochastic Parrot? A Compression Engine? The discussion is ongoing, and early childhood educators ought to be a part of it.
So what might a computer see if it, in turn, saw a cat?
Stephen Wolfram has a lovely research paper on the exact problem, with many cats in party hats, some code, and math. In the text, he shows how neural networks create an image to fit the prompt, a cat in a party hat, starting with random pixels and then iteratively forming the image. Wolfram shows us how to travel within this space.
It’s hard to visualize this space since the neural network Wolfram Alpha uses has 2304 dimensions - and we humans can only perceive three at a time. But one can slice through a two-dimensional space and find these fascinating, collage-like examples.
The images here are “statistically reasonable” based on the millions of cat images we humans have put on the web.
What are all these things around the cats? Let Wolfram explain:
But the fundamental story is always the same: there’s a kind of “cat island”, beyond which there are weird and only vaguely cat-related images—encircled by an “ocean” of what seem like purely abstract patterns with no obvious cat connection. And in general the picture that emerges is that in the immense space of possible “statistically reasonable” images, there are islands dotted around that correspond to “linguistically describable concepts”—like cats in party hats.
It means there is a whole space of non-cats, almost-cats, not-quite-there cats, and cat islands that we don’t see and don’t have words for but still exist in the manifold matrices of the neural network.
And I love this idea so much.
I will run this experiment with children later in the spring and share the results, but you can now participate. Play with me! The style is free: collage, pens, play-doh, code, gen-ai, and matrices...
My prompt, with further guidance, is below.
The cat walked through the world, with its whiskers, ears and paws…
and the computer saw A CAT,
and the algorithm saw A CAT,
and the neural network saw A CAT.
Yes, they all saw the cat.
Start by reading “They All Saw a Cat” aloud. Enjoy the rhythm, the illustrations, and the different perspectives.
Ask children to first draw their own interpretation of a cat. Compare these drawings.
Now, what about a computer? How might a computer, an algorithm, or a neural network see a cat? Celebrate the different ideas. This activity is less about right/wrong answers and more about developing empathy and early mental models.
Compare the different perspectives of humans and machines. Share observations and interpretations. Group works according to different themes that arise.
Linked List
In computer science, a linked list is a linear collection of data elements whose order is not given by their physical placement in memory. But here it is a selection of things I’ve been reading lately.
From Czech writer Karel Čapek, inventor of the term robot, comes a 1935 short story on From the Point of View of a Cat. The cat, obviously, doesn’t care.
How Big is YouTube? I love a good story of technical/practical problem-solving, and this hits all the marks: to estimate the size of YouTube, Ethan Zuckerman & team “drunk dialed” random video URLs and counted how many were valid. The answer is roughly 14 billion videos, but once again, the answer is less interesting than the question and the process.
Robin Sloan’s new book, Moonbound, is out for preorder! I love the Penumbra universe and can’t wait to dive in. Meanwhile, the best recommendation line from his last newsletter is “Claude Shannon would have loved Death Note.”
Classroom
I’m hoping to surface and share stories from all of you and I’d love to see your creations! Here are a few teachers using Ruby in creative, fun and inspiring ways.
Ms. Leedy’s students practice using Ruby’s map and giving step by step instructions.
And ZMI lists top books for Artificial Intelligence and includes the German edition of Wenn roboter zur Schule gehen! And no, unfortunately still no news on translation in English :(
