A Pseudo-Literate Parrot

Both hopes and fears concerning AI are ignorant and misplaced

A commenter helpfully describes the complete ignorance concerning what AI is and how it functions in colorful rhetorical terms:

The levels of ignorance on How AI Works are matched only by the snakeoil 'engineers' and salesmen who peddle the AI as anything but a literate parrot.

I've played extensively with AI models, read the literature, built/trained new models, coded using the concepts, and compared LLMs, SD/etc visual models, and more. I still use them for 'shortcuts', but I've learned to trust them no further than a wily slave laborer who hates his master and looks for any opportunity to put one over on them, and who will lie with a straight face, and swear eternal fealty.

LLMs do not store facts. EVER. They store map points in a vast mega-dimensional landscape of uncharted axes of undocumented spectrums. Not only is everything relativistic to it, but it always finds the path of least resistance thru that jungle of cloudy 'concepts' it holds. That's the non-technical description of what is so well documented by 3blue1brown (recommended to all!): it's a big table of probabilities. It's no more true than rolling a D100, consulting a table in the appendix of the D&D reference and declaring that you've rolled a critical hit or failure. It's fiction, entertaining, but at the end of the day, pure fantasy gaming.

I’d even go a little further, because the parrot not only doesn’t understand what he’s saying, he isn’t actually literate either.

I’ve tried to explain this in terms of pattern recognition and matching to people, but been largely unsuccessful, probably because most people are binary thinkers and incapable of thinking in probabilistic terms. But it’s not really that hard.

Let’s limit the discussion to AI-generated music and simplify it even further. Imagine that there are only 8 musical notes. The factorial function tells us there are n! ways of arranging n distinct objects into an ordered sequence, so 8! means there are 40,320 different ways to combine these notes, from 1,1,1,1,1,1,1,1 to 1,2,3,4,5,6,7,8 all the way through 8,8,8,8,8,8,8,8.

Now, through training, the parrot is taught which sequences are common and which are not. It recognizes that sequences that are all odd such as 1,3,5,7,7,5,3,1 or all even 2,4,4,2,6,6,2,8,8,6 are common while those with just one number repeating are uncommon. It doesn’t know what these notes are, or have any idea what they represent, it just knows which numbers go into which sets, and which sets usually follow each other and which do not.

Then, through algorithmic restriction, the parrot is instructed which sets and sequences of sets should be favored and which should not. So, for example, the programmers might tell the parrot not to whistle a backwards set unless the set is first counted forward.

This means that when you “roll the dice” by requesting a musical track, the parrot will usually whistle a sequence that is both a) common and b) programmatically favored out of the 40 thousand possibilities. But it’s not doing any thinking or reasoning or copying at all, it’s just consulting a very big table, very fast, and quickly producing a result that sounds credible to the average human listener based on the way it follows past patterns known to be popular.

And, of course, the gargantuan number of possibilities means that accidentally recreating a pre-existing melody is almost impossible. A normal human voice has a two-octave range; therefore there are 620,448,401,733,239,439,360,000 possible combinations of those 24 notes. Multiply the number of notes in the average 3-minute pop song by the number of tracks, then calculate the factorial to see exactly how absurdly mathematically improbable it is for Suno or Riffusion to reproduce even a very simple instrumental with a basic melody, no matter how tightly its algorithms are restricted.

Interestingly enough, this is exactly how the system I designed for MMO bots designed to realistically mimic human gamer behavior worked. It was just a large probability table that ensured a given reaction in a specific situation would be perceived as realistic, based upon a concept that the original designer of the WWII infantry game Squad Leader called “design for effect” as opposed to “design for simulation”.

One of the greatest game designs of all time.

AI isn’t simulating the operation of the human brain or any other kind of intelligence. It’s simply producing results that are designed for effect, which is why it so readily produces results that look and feel realistic, even though they are entirely fictional.