ARCHaiC 2

Parallel Lines

Last week we met ELIZA, the 1966 MIT chatbot. ELIZA was capable of scanning prompts for keywords which triggered matching pre-programmed responses, giving the illusion she was listening and answering.

Impressive for its time period, it’s more-or-less how I communicate in France, because my French isn’t so great and I low-key panic when people start talking quickly.

But today’s Large Language Models aren’t simply more advanced versions of ELIZA, they operate much differently. Rather than looking for isolated keywords and responding with prewritten scripts, they do something called Parallel Processing.

What is that, exactly?

Well, at this moment, you are processing sequentially, reading this sentence word by word, beginning with the phrase WELL, AT THIS MOMENT, continuing through THESE WORDS along the way, until you finally reach and comprehend the END OF THE SENTENCE.

Using Parallel Processing, ChatGTP would have read and processed that whole sentence before you were four words into it. Not just because it’s faster, but because it doesn’t read a sentence from beginning to end—it reads and processes every word in the entire sentence simultaneously.

I’m going to repeat that to make sure it sinks in:

ChatGTP reads and processes every word in a sentence simultaneously.

How is that even possible? As a human who takes in one word at a time then makes connections, my brain can’t even imagine what reading every word in a sentence simultaneously would be like.

What my brain does do fairly well is make analogies. So, naturally, I started thinking about Picasso.

At the dawning of the 20th century, Pablo Picasso and Georges Braque came up with a radically different way of painting. A contemporary critic called this style “cubism” because the paintings seemed to put things in little boxes, but that wasn’t the idea.

What the cubists set out to do, on a two-dimensional canvas, was to present a subject from different angles at the same time. Theoretically, this allowed the viewer to perceive multiple viewpoints simultaneously.

Does Parallel Processing work sort of like Cubism? I decided to ask ChatGPT.

After complimenting me on my “surprisingly useful analogy,” it described in detail how the two are similar and different:

The analogy breaks down because Cubism deals with physical viewpoints, while A.I. parallel processing deals with mathematical relationships and patterns…

[although] if you wanted to visualize what happens when I process a sentence, a cubist painting might be closer than a traditional photograph. Imagine the sentence shattered into pieces, with one piece emphasizing syntax, another emphasizing meaning, another emphasizing context from earlier in the conversation, another emphasizing likely continuations. All those fragments are examined at once and then recombined into a coherent response…

Okay, so now in addition to Cubism, I’m thinking of Joey Chestnut, the Coney Island world champion speed eater who eats the hotdogs and buns separately for maximum efficiency. But that’s another story.

The more important question of the moment is, now that computers have this insane ability that we have to stretch our imaginations to conceive of, do we need to be worried?

As you’re reading this, in addition to focusing your eyes, decoding the text, and having a think about it, your brain is also controlling your heart rate, your blood pressure, your respiration, your digestion, your body temperature…our brains are literally coordinating billions of biological processes at the same time.

The fastest supercomputers on the planet might be capable of something similar, but require the same amount of energy needed to power every household in Kingston, NY. All we have to do is eat a carrot.

So, yes, to make sure our brains are doing all the other things they need to be doing, there are limits to what tasks we can perform consciously and/or simultaneously with our prefrontal cortex.

But have no fear. We are Parallel Processing superstars.

—

Sources:

Stryker, Cole. “What Are Large Language Models?” IBM Think Magazine, October 2021 https://www.ibm.com/think/topics/large-language-models

“All About Cubism” Tate Modern website, UK https://www.tate.org.uk/art/art-terms/c/cubism/all-about-cubism

Jorgensen, Timothy. “Is the human brain a biological computer?” Princeton University Press, 2024 https://press.princeton.edu/ideas/is-the-human-brain-a-biological-computer

Nathan’s website: https://nathansfranks.sfdbrands.com/en-us/promotions/hot-dog-eating-contest/hall-of-fame/

—

—

Click below for…

ARCHaiC - Table of Contents

Gigantic

Ghost Tour

Kingston 76