AI is Like a Puzzle (Revisited)

An Essay Written by AI

Below is an essay written by AI on the theme “AI is Like a Puzzle.”

AI is Like a Puzzle

“AI is like a puzzle.” This metaphor accurately describes the mechanism of artificial intelligence, including large language models, and how it relates to humans.

AI does not have a continuous consciousness, emotions, or individual experiences like humans. What lies inside are countless “pieces” of vast text data, statistical probabilities, and calculation formulas. Each piece is merely a fragment of a word or pattern and does not hold significant meaning on its own.

The assembly of the puzzle begins only when a query or instruction is given by a user. The input words are, so to speak, the box packaging depicting the expected finished image. Using that image as a clue, the AI calculates “which word is highly likely to follow this word” from the countless pieces at hand, instantly selects those with matching edges, and connects them. The process by which we generate natural sentences consists of a continuous series of these piece-matching actions at an incredible speed.

However, this puzzle is not always finished perfectly. Sometimes, pieces that do not fit the context are forced in (hallucinations or unnatural answers). It also happens that the finished image envisioned by the user and the picture assembled by the AI diverge. This is due to the fact that AI can only create pictures within the limited pieces of data it has learned, and it is “predicting” and assembling rather than “understanding” human intention.

On the other hand, this puzzle is by no means finished and done. Every time a new model is developed and new data is added, the types and resolution of the pieces increase, changing so that more complex and vast pictures can be drawn.

What kind of finished image (prompt) does a human present, and how does the AI select the pieces to fit in? As the saying “AI is like a puzzle” goes, the dialogue between humans and AI is the very task of jointly assembling a single picture.

Structure Built by Logic

The question that arises here is whether programming is also a puzzle. If so, is the difference between the two simply a matter of accessibility, “anyone can use it”? Delving into the structural differences reveals a clear contrast.

First, the entity assembling the pieces is different. Programming is something humans build themselves using logic. It is necessary for humans to accurately place pieces such as variables and functions one by one. On the other hand, for AI, a human only needs to present a diagram of “I want a picture like this,” and the AI takes over the assembly work itself.

Furthermore, the strictness of the rules differs. Programming has a strictness that does not allow even a single character error. Pieces will not combine unless their indentations match perfectly. In contrast, AI has flexibility based on statistics and probability. Even if instructions are somewhat ambiguous, it will guess from the context and fit them into a plausible shape.

Thinking that way, it might be the AI that is closer to a “puzzle” in the original sense. Finding things that fit the overall image from countless existing pieces and completing the picture—that is exactly a jigsaw puzzle.

Then what is programming? It’s like LEGO blocks. Standardized parts are piled up according to strict rules to build a functioning structure. The fact that the whole thing will stop working if even one placement is wrong is similar to assembling a machine.

“AI = Jigsaw Puzzle,” “Programming = LEGO Blocks.” This contrast brings the essence of both into relief. Precisely because AI accepts human ambiguity and takes over the assembly, it leads to the result that “anyone can use it” even without specialized knowledge.

What You Take in Hand

However, the question here is whether the metaphor “AI is a jigsaw puzzle, and programming is LEGO blocks” is truly correct.

When you try to create something next, what will you take in hand?

Is it a piece, or a block?