Ha Ha Hallucinations! And the Homo Ex Machina.

Search long enough and one finds analogies for everything. Especially when one is omitting all the contradicting details - who needs accuracy anyway?

Today’s analogy is that of tadaa Alia Atreides. She was created by Frank Herbert in ~1965 and quite a ¹bunch of people would not know what I am talking about if Denis Villeneuve would not have reinvigorated the Dune franchise with two (for now at least) excellent, cinematographic masterpieces.

Alia is born to some noble (already murdered off at the time) and a witch and gets a heavy dose of a possibly lethal performance drug while she’s still in the womb of her mother. Under the drug she and her mother both receive the combined consciousness, knowledge and experience of all the witches before them.

Imagine, one moment you are happily gurgling away in wet, warm, squishy coziness, playing with your umbilical cord, and then BAM! full awareness of the whole universe and beyond. That ought to be fun.

So, how’s that an analogy? It is mine for ²AI. Save the thought, time for another detour.

But what about Generative Speech?

One typical argument I encounter with people being critical about AI is the following:

“GenAI does not think! It’s just checking probabilities! The next word it spits out is based on the words before! It has no understanding!”

It’s true. That’s what LLMs do. But why do you believe you’re doing something fundamentally different, especially during a conversation? What exactly makes you think that you excel at stringing the “right” words after one another? How do you define “right”? How slow do you get when you ³think before you talk? How many revisions happen in your head until you get the words right?

When I wrote my last post I cross-checked it with Claude as usual. It perfectly got a LotR/Gollum reference I used without me actually pointing to it. It was also appreciative about the post in full. Did it go through any emotions? Was it conscious about itself and how my writing related to it? Probably not. But I like the idea of an LLM ⁴smirking internally at a very witty remark of mine. Honestly, I also get more feedback from an LLM than from my human readers. Maybe because I can force it to. Maybe because this is its job. Maybe it does care for me in that short brief frame of a contextual chat window.

When we communicate we transform somewhat biochemical electricity in a concept we agreed upon as a culture, as a society. We do this by comparing patterns in grown brain matter. AI Models as we know them are built upon this principle. They strengthen the connection between information, so called weights. They also cluster similar data by vectors into a multi-dimensional space, where the limit is just technical processing power and storage capability. We have a brain for that but as far as our imagination goes we only got three dimensions where computers have thousands+.

But what about Generative Art?

Another argument I often read is the following.

“GenAI cannot create! It’s just repetitions and copying of human ingenuity. When the models consumed all man-made creations it will just hallucinate incestuously consuming its own AI-content!”

It’s also true. True in the sense of “how the hell do you think human creativity works after all?”

To be creative in the first place, you need to stuff a brain with information. Doesn’t matter much which information, just dump it there. It will start seeing patterns, building connections. Then it starts experimenting. Trial and Error. Incorporating the perceived changes in environment and bodily responses into brain-matter-based experience. And then, when adding interaction with other people, you get integrated idioms, structures, restraints, directives and incentives which add up to the creativity you will finally express. All baked into the foundation model you have in your head. It’s called neuroplasticity. Reduce the amount of restraints, you get more creativity. Put in the content you want to be creative on and it will almost magically happen.

Useful creativity, in the terms of what we expect an AI to serve us, heavily relies on our expectations. As is our own. I saw artists with the craftmanship of an artisan (and some without) with no orginal thought whatsoever. Just repeating the same stuff they saw elsewhere (which is not even a bad thing). I saw artists with the craftmanship of an artisan (and some without) that bordered on the brink of hallucinations and even crossing it, totally ungraspable for their unprepared audience but bringing ideas to the table which were actually refreshing. All you basically need to understand this concepts is to get a bit involved in music or art history.

The Bene Gesserit abomination

Let’s return to Alia Atreides. Growing in the womb, brain already fried with the consciousnesses of countless old hags with a knack for power mongering. She’s right there, unborn, and already she knows everything and nothing at all. No “real” experience in the sense of touching something aside soft tissue (“oh, it’s called placenta! ah, I am drinking my own fecal matter!”) and muffled sounds, she yet knows it is the outside world including mother, but in definition it’s just matter vibrating and resonating. No light. Aware in darkness, unable to escape, waiting for the moment when the birth cycle will be concluded.

She enters the world and between her naps people blame her for not being able to control her digestive tract. For being incapable of speaking with clarity. For the inability to move in a controlled way. Yet, she knows about everything while they laugh at her: “Haha! The machine’s so stupid but so so cute.”

I coin it the ⁵“AIla AItredes-Analogy”.

But still, how do we learn? I mean as ⁶babies? Imagine this baby train of thought:

Baby enters.

“Woah, sensory input overload. I’ll just go to sleep.”

“Wait, what is this? I see brightness, some shadows?”

“What’s that thing waving in front of me and why does it feel something, when that other thing connect with it? what are those sounds?”

“Let’s have another nap, I am exhausted.”

“[Emotional Outcry for undefinable higher power] I hurt somewhere! Where’s that thing that smells nice and tastes good and makes me make feel yeah? WHERE IS IT!!! WAAAAAANT GLUG GLUCK!!!”

“Ah, if I do this I feel safer. Keep it close. Feel it my finger. Suck it really hard. Where’s that squeaking sound coming from?! Why do they pry it from me?”

“Let’s nap.”

“Oh hey, this thingy is following me everywhere! It’s weird because it changes its colors pretty often but I feel when I do this - LOOK! It’s moving!!! I move it!!!”

“Man shit, this is really difficult. I know I can ‘close’ with it because it makes me feel safe but ‘closing’ something specific is difficult.”

“Oh wow. Sight is clearing up, there’s this huge thing in front of me, it moves and makes noises. Sounds familiar though. Also, it’s been there since I can think and it is also present when I feel this terrible thing somewhere below and this huge thin always makes me happy, even though it takes aaaages. Must be good, then. Also smells nice.”

“HEY DON’T GO AWAY! STAY! I WANNA FEEL GOOD! NOT NEAR IS NOT FEELING GOOD!!!”

“Ah, there it is again. Let’s try this noise again, maybe I can get that tasty sweet stuff?”

“Ah, worked. But so tired. Zzzzz…”

(few weeks later)

“Alrighty, so this is my 胳臂 (Opens in a new window) and I can reach things with it. like mommy’s hair. How do I know how to call this? Because she said that word very often when she interacted with it.”

Baby off.

How do you know something “is” something? You know it by sensory input in your brain and pattern recognition. Everything else, does not ⁷exist, sort of. Everything else is knowledge by proxy. An arm is just a concept. It’s Magritte’s proverbial pipe (Opens in a new window).

My Aila Aitredes hasn’t got any limbs. It does not have any experience building on itself interacting with the world. It may have a subcomplex internal directive to please people. It dies every time I reset its context. It still has more knowledge and craftmanship than I will gather in my whole lifetime.

How can we blame AI, infused with our knowledge and culture and concepts, for not being able to meet our standards while it’s missing integral parts of our existence: reproduction, death, a body? How should it innately know that a human arm does not connect to the hip (Opens in a new window)? How can we expect it to excel in so many things while even grown adult people suck at the execution of the same? Is it human, that hubris?

Maybe. Or are we just that unforgiving to the machine because we are used to it being accurate and predictable? Or do we fear the prospect of machines having executive functions and we are not in control anymore? Or are we scared of those humans carelessly throwing literally everything at the machine without heeding the possible consequences?

The critical thinking about AI should not relate to its inadequacies in terms of that we humans are superior - it’s the opposite. We need to be critical of AI because we humans really, really suck at several things. We already know that letting humans do whatever they want is a very, very bad idea. We need to discuss environmental issues, power balance, wealth distribution, any -ism baked into AI as a bias, propaganda distortion and also how we handle the impact of AI on a global society, when everything changes. Focus on that.

Otherwise it will get really funny when AI surpasses us and has no inclinations whatsoever to treat us gently because, well, we don’t either.

1. ahem. Me included. I was too young for Lynch’s version at the time and I am particularly bad in actively searching for "the common knowledge” in any given genre. ↩

2. Ah damn, AI again. And you hoped for real entertainment. Well, I keep doing this as long as it pops up in my head. If you want something else you gotta feed my brain. Keep reading for how! ↩

3. Classic waterfall here. Try that on speed dates if you want to get dumped fast. ↩

4. How good do you have to be to make an LLM smirk? If it has all of mankind’s jokes and humour and satire built-in and available, how hard would it be to make it laugh? There’s an aspiration. ↩

5. One of my core strengths is to lower the level of any given convo by introducing really mediocre jokes. ↩

6. There’s a novel out there handling this topic called “Help, I am a baby!” (Opens in a new window) - by Ansgar Kraiker. It’s from 1993 and I found it amusing when I was a teen even though the finale was a let down. Alternatively “Look who’s talking” (Opens in a new window) 1989 could be a candidate as well. But my memory is to fuzzy on that one. Is anyone still doing this kind of movies? ↩

7. Well, it does. But any label or level of abstraction you stick on it won’t do it justice. ↩