Your Word Game Brain Is Doing Something AlphaFold Can't
There's a paradox worth knowing. It's called Moravec's Paradox, it showed up in a Saturday Morning Breakfast Cereal comic on June 12, 2026, and linguist Mark Liberman liked it enough to file it under "Linguistics in the comics" the same day. That's a solid recommendation.
The short version: AI is brilliant at the things humans find hard, and bafflingly bad at the things humans find easy. A computer can solve protein structures that stumped biologists for fifty years. It struggles to fold a t-shirt. The tasks that require a doctorate turn out to be the easy ones. The tasks your body does without thinking are the hard ones.
The Protein Part
In 2020, DeepMind's AlphaFold cracked protein folding, a problem so significant that Science magazine gave it serious real estate (DOI: 10.1126/science.370.6521.1144). Proteins fold into specific 3D shapes. The shape determines the function. Predicting which shape a given amino acid sequence would produce was one of biology's great open problems. Decades of effort. AlphaFold solved it.
That's Moravec's Paradox in action: formalize a hard problem into math, and machines handle it beautifully. Meanwhile, walking through a kitchen without bumping into the counter is still a robotics research frontier.
The Laundry Part (and Why You Care)
The paradox says that sensorimotor tasks, the things your body does automatically, are genuinely hard for AI. Folding laundry is the famous example. Your hands know how to do it. Describing how your hands know how to do it turns out to be nearly impossible to formalize.
Language works the same way. Word games live in exactly this zone.
An AI can hold every legal Scrabble word in memory simultaneously. It can run letter frequency analysis on any dictionary you hand it. But the felt sense of a pun landing? The moment a crossword clue clicks and you can't quite explain why? The gut read that tells you "CRANE" is a stronger Wordle opener before you work out the math? That's a different kind of knowing. Embodied. Intuitive. The thing Moravec pointed at.
Where It Gets Stranger
Chris Olah's research into LLM internal states adds another layer. His work suggests that large language models may have something like liminal consciousness, internal states that aren't quite nothing. One possibility is that the distance between processing language and experiencing language is narrower than expected. That's genuinely speculative territory. But it's the kind of speculation that makes you look at your next Wordle attempt differently.
If models have any flicker of inner states when working with words, then the gap between "tool" and "player" blurs a little. Not enough to matter for your Tuesday crossword. Enough to be interesting.
What the Comic Got Right
SMBC comics work because they compress something true into a small space. Liberman flagging this one for linguists means the Moravec framing landed on something real about how language works, and how intelligence works, and how weird it is that the two don't line up the way you'd expect.
High-level reasoning: solvable. Protein folding: solved. Sensorimotor intuition, the stuff that makes word games feel like play instead of computation: still the hard problem.
The Compliment Hidden in the Paradox
When you finish a tough crossword or find the bee's missing pangram, you're doing the hard version. Not the formal reasoning part. The part that confounds the machines. The pattern recognition that happens sideways, from the corner of your attention, before you consciously ask for it.
That's your word game brain doing the thing AlphaFold can't. Not despite being simple. Because it's built on the same architecture that catches thrown objects and recognizes voices through walls and knows a word sounds right before you can say why.
A robot somewhere is still working on the laundry. You're already solving the puzzle.
Source: Languagelog