A Thousand Words : Part 1

“A picture is worth a thousand words”.

Is it though? How do you know it’s worth about a thousand words? If you wrote a thousand words describing a picture, could you accurately do so? If someone else read your description, how accurately could they reproduce the picture? With a reasonable amount of certainty, you could say that it might be similar, but definitely not same.

That’s sort of the purpose of this blog - to question whether language and thought are the same thing. To see if we can appropriately differentiate between the two while also recognising that they work in tandem with each other.

“Why just use a thousand words? Why not more?” You go ahead and describe the same picture with ten thousand words. Here’s a short example-

Image generated by DALL-E

Suppose you saw this picture and described it to a painter-friend as “A beautiful mountain landscape during sunset.”

Now your painter-friend might create this-

Image generated by DALL-E

Two different images. But your painter-friend isn’t wrong. You could say - “well, that’s probably because I haven’t described the picture in sufficient detail.” Maybe you would describe the “beautiful mountain landscape” some more - like the positioning of the mountains, whether the sun is in the image or not etc. But that just gives rise to more questions - how many mountains? Where exactly do you place the sun? etc. Although you’ve narrowed down the possibilities and might be in the same ballpark, you’re still quite limited when trying to express what you think when using language and are bound to hit a wall.

Theoretically, you might be able to perfectly reconstruct the image with more structure in the language (say you mention RGB values of each pixel), but at that point your painter-friend is more of a computer and less “human”.

This leads to a much deeper question - possibly one that you may have come across yourself - does thought require language? Or is language merely used for communication?

In other words, do you think in terms of language? Or is thought “something else” which then gets translated to language for the sake of communicating thought?

We used language to describe a picture, but it’s now really hard to reconstruct the original picture from just the language. Kind of like flattening a 3D image to 2D and trying to reconstruct the original 3D image. You lose information during the 3D->2D step, which seems very similar to describing a thought or a picture in words.

So far we’ve only discussed the thought->language direction, but the reverse is also true. Words are sometimes used to actually spark imagination and thought. I’ve read all the Harry Potter books but have not watched a single movie for that very reason. Watching the movie would rob me of my own imagination and interpretation that came from reading the books. To me, that’s probably the most significant benefit of reading a book over watching a movie - books are the only way you can experience the story in the first-person. I think it’s beautiful that words can be used to that effect.

Language also brings clarity to thought. Obviously this is highly dependent on the author/speaker, as this could also bring about more confusion, but with the right wording/verbiage, thoughts can be communicated with a high amount of clarity. A very popular method of debugging code in software engineering called Rubber Duck Debugging uses this - it’s a method that forces you to lay out the logic of your code by explaining it line-by-line to some random object, like a rubber duck. Something similar has probably happened to you many times - sometimes you describe a problem you can’t solve to your friend, and as you’re explaining it, the fix suddenly hits you. The structure of language, although limiting, is also necessary to tame thought.

This is probably why language is so confused with thought. Language is used to communicate thought. Language is also used to bring clarity to thought, refine thought and might even be used to spark thought. But according to research, it isn’t thought.

I did some reading and found that in some research published last year, researchers at MIT found out that the human brain has two separate areas/neural-networks that process thought and language. They found this out through carefully thought out experiments and tasks requiring thought and language and used scans to see which parts of the brain were being activated during these tasks. They also show that language has been implicitly optimised for communication - shorter sounds/words for describing common everyday events and so on. Research like this clearly shows that although language and thought may have co-evolved, they are treated as separate entities by our brain.

In the picture-words example, say, instead of a picture we had some other form of art - like music. Lyrics, just like words, can invoke emotion. But lyrics with sounds of musical instruments can take this to a totally new level. Now combine lyrics, sound, and video and you can sometimes find yourself crying in a movie theatre. Why stop at art though? What about when we behold beauty in another person? What about when we are saddened and hurt by acts of brutality but inspired and overjoyed by acts of kindness? Does it not stir our emotions, sometimes to the extent where it steals our breath away? Perhaps you’ve been there too - just beholding. No words. No language.

In part 2 of this blog, I would like to explore this further. With the rise and success of language models like ChatGPT etc, I want to explore whether these models actually think and reason based on some amazing recent research. And if these models can talk, solve math problems, write code, paint and create music like humans, then, intellectually, what makes us as a species uniquely human? I find it fascinating that all this cool research in machine-learning eventually leads us to learn what makes a human human.

By mere coincidence, the length of this blog is exactly a thousand words.

Thanks to Anush Kini and Roshni Poddar for reading drafts of this and providing valuable feedback.