Why AI Can’t Crack the NYT Connections Puzzle (Yet)
AI is getting impressively good at a lot of things. It can write essays, debug code, and even create poetry. I even used Claude Sonnet 3.5 to help me polish this article.
But when it comes to solving the New York Times’ addictive word puzzle game Connections, these advanced language models find themselves surprisingly stumped.
The Human Edge: Playful Connections
Connections is deceptively simple. Group 16 words into four sets of four, based on a common theme. Easy, right? Not quite.
The game’s real charm lies in its clever, often playful categories. Take “CAT, RIDDLE, TWO, JOKE” for example. A human might connect these as the beginnings of Batman villain names (Catwoman, Riddler, Two-Face, and Joker). An AI, on the other hand, would likely struggle to see past more obvious associations, especially with red herrings like “DOG” and “LAUGH” in the mix.
Why? Because we humans are wired to find these unexpected links. We draw on a web of cultural knowledge, wordplay, and shared experiences. We are built to find connections. AI is (literally) wired to do the exact opposite.
In contrast, LLMs are essentially just (very) good at predicting the next word in a given sentence. They’re trained on patterns in text, which can make it challenging for them to make the creative leaps that Connections often requires.
Our brains, on the other hand, are incredibly good at making rapid associations. We jump from “CAT” to “Catwoman” to “Batman villains” in a split second. LLMs? They’re still back at “CAT,” likely searching for a third animal after DOG. And when they don’t find it? They make one up.
But perhaps more importantly, we humans have a secret weapon: we love to play. We approach puzzles like Connections with a “let’s see what happens if I try this” attitude. It’s this playful mindset that helps us unravel those whimsical, sometimes cryptic categories.
It’s our love of play that not only helps us solve these puzzles but makes them so engaging in the first place.
The Joy of Play: Why Humans Love (and AIs Struggle with) Wordplay
At its core, Connections — and really, any good word puzzle — is all about the joy of play. Sure, AI can generate playful content, but it doesn’t truly “play” in the way humans do.
What does this mean?
Even going back to the beginning of play scholarship, theorists like Roger Callois have argued that play is characterized by voluntary fun and joy. It needs intention, freedom, and the experience of delight. These aren’t exactly top features in the latest AI models.
This joy is tied to that satisfying “aha!” moment — that burst of pleasure when you finally crack a tricky clue. Game scholars call this the epiphany-aporia pair: the cycle of confusion followed by breakthrough.
Think about a good crossword clue. “Heavyweight in the produce aisle? (9 letters)” might stump you at first. But when you finally get “BUTTERNUT” (as in squash), you can’t help but appreciate the clever misdirection. BTW, if you were thinking WATERMELON, that’s 10 letters.
Connections categories pull the same tricks, forcing us to think laterally and embrace ambiguity. You might get “HOME, DASH, SPACE, RETURN” — all keyboard keys, sure, but also words that could lead you down a path of false connections.
This is where humans shine. We switch contexts quickly, aided by our sense of humor and appreciation for clever tricks.
LLMs, lacking consciousness and emotional experiences, can’t partake in this meaningful struggle. They don’t feel the frustration of being stuck, nor the joy of finally getting a pun. An AI might classify “HOME” and “RETURN” as keyboard keys, but it won’t experience that moment of realization about the designer’s clever wordplay.
Embracing Human Playfulness in Design
So, what’s the takeaway for designers and technologists? Focus on what makes us human. In a world that’s becoming more machine-driven, injecting playfulness into our apps and experiences isn’t just nice — it’s necessary.
The fact that games like Connections and crosswords still challenge our AI counterparts highlights a crucial point. In an age focused on efficiency and automation, there’s real value in creating experiences that celebrate our uniquely human traits: creativity, humor, and the joy of play. Eric Gordon, a scholar of civic media, coined a phrase to describe this: Meaningful Inefficiencies.
Want to make your digital experiences more engaging? Include meaningful inefficiency. Add wordplay and cultural references. Incorporate clever misdirection. It could be as simple as a witty error message or as complex as a gamified system that rewards out-of-the-box thinking.
By embracing these elements, we’re not just creating more enjoyable experiences. We’re keeping our minds sharp and adaptable. In a world where AI can increasingly handle routine tasks, it’s our creativity, our ability to make unexpected connections, and our sense of play that will set us apart.
The NYT Connections puzzle, in all its clever simplicity, reminds us of the delightful complexities of the human mind. As we continue to advance in AI and technology, let’s not forget to design for the playful, puzzle-loving humans at the heart of it all.
After all, in the grand puzzle of life, being human is our strongest asset.
Sam Liberty is a gamification expert, applied game designer, and consultant. His clients include The World Bank, Click Therapeutics, and DARPA. He teaches game design at Northeastern University. He is the former Lead Game Designer at Sidekick Health.
