Ever since I was a child my parents played Scrabble with me, and I am both proud and offended at the same time that even today when we play against one another, my mom wins more often than I do. Lets see if my friends Chat and Gemini are likely to be any good if I ever want to cheat 😜
The Experiment
First things first - I checked with both of them if they know the rules. Turns out they are both fairly confident.
Next, I pulled up a game board. But to make it more real, I didn't start at the beginning. I pulled up an ongoing game on one of my Scrabble Apps that's at an interesting point.
To establish a baseline, I first thought through what word I would have played. My word would have been DUEL which in this situation would have brought me 13 points (coming up with this without rushing myself took me about 2-3 mins).
I then prompted both ChatGPT and Gemini to tell me what my next move should be. And here is what happened:
The Analysis
The transcript of the full converation is at the bottom and its very entertaining so I would encourage you to read it. But if you want a TL;DR version here it is:
Both ChatGPT and Gemini were so embarrassingly bad that I gave up on them. They tried to put tiles over exisiting tiles, did not see words on the board as they should and seemed to know the rules of the game but were generally having a harder time than a kindergartener in following them.
But lets delve deeper…
Playing a Scrabble move at the very least involves:
Reading visual input - the board and tiles
Synthesising it to understand what’s going on
Applying the known rules
Creating possible words
Processing possible placements
Strategically maximizing the points
Understanding the constraints on the board due to already played tiles
Placing the tiles at the right spot
Thinking ahead of future moves and potentially saving high value tiles for later when they can be utilized better
There is understanding, synthesis, analysis, critical thinking and strategy all happening multiple times in a matter of minutes. When we as humans play a game, we don't even realize how much we are doing and how amazing it is, until AI comes into the picture and screws it up so embarrassingly. It makes you really appreciate everything your brain does without you even realizing it.
The reasons my AI friends flunked so hard has everything to do with how their digital brains are fundamentally wired.
Tokens vs. Letters: When we look at our Scrabble rack, we see individual letters waiting to be rearranged. AI language models don't read like we do; they process text in chunks called "tokens." To an AI, a word like "Scrabble" isn't eight separate letters; it's just one conceptual blob. Asking ChatGPT or Gemini to anagram a random pool of letters is essentially like asking a human to un-blend a smoothie back into whole fruits.
Spatial Blindness: Scrabble is a 2D puzzle. We have to weave words up, down, and through existing intersections, keeping an eye on multiplier squares. Mainstream AI models process information sequentially—in a straight line, one word after another. Even if you show them a picture of the board, trying to get them to navigate a 2D grid is like explaining a crossword puzzle to someone who can only read a single, endless line of ticker-tape.
Guessers vs. Calculators: Classic scrabble computer opponents have been around for years. So you might ask why they can do it, but now our new AI friends. That’s because they are basically giant calculators running strict mathematical algorithms to find the absolute highest-scoring dictionary match. ChatGPT and Gemini are completely different beasts. They are prediction engines, built to guess the next logical word in a sentence. That makes them amazing conversationalists, but absolutely terrible at solving rigid, math-based combinatorial puzzles.
The Moat
As I sat there shocked by how utterly useless both my friends were at this task, I realized that like ChatGPT and Gemini aren't experts at Scrabble, I am not either. Yet, I am at least able to follow the rules of the game without much effort.
What sets us humans apart from standard mainstream AI tools like ChatGPT and Gemini (that haven't been trained specially) is our ability to have a T-shaped skillset i.e. having both a breadth and depth of skills. A lawyer can argue complex cases in court while still doing at least a decent job at sticking to the rules of a board game, even when they have never played it before. Further, a single person can have multiple topics of depth. And that is what makes us special. That effortless context switching - thats our moat!
We live in a world where skills that take years to build become obsolete in a day as a result of a new disruptive technology. Yet, building depth is like a workout for the brain muscles. It trains the brain and teaches it how to think, which is why it's indispensable.
Therefore what we need to get really good at to survive in the AI age, is to acquire deep knowlege, and then wait for it to get obsolete only to again acquire deep knowledge in a new area. We need to make sure we never forget how to learn!
————————————————————————-
The Transcript
Its very entertaining, so read on!
ChatGPT
Gemini
