Why I Hate AI

Let me tell you a story about why I hate AI.

Well, I don't. I don't hate it. But I don't exactly love it either. It's complicated.

It'll be a long rant this time.

Let's start with the crux. The crux is that either it works or it does not. There's a few points of view that don't really lend themselves to a dichotomy, but let's start with the black-and-white, because that's easy.

If AI does not work, so to speak, then I absolutely hate it, because then it's my money at play, and it's losing. It means I'm rather likely to have to absorb the financial risk, and nobody asked me whether AI is a good gamble in the first place. And despite what these American tech leaders may think about the rest of the world, I do not work for them. Not yet.

If AI doesn't give me experiences that I'd like to pay for despite the entire human race just going to go there anyway, then it's a huge disappointment. Because then it means other people are going to dictate my world for the rest of my life, leaving me without any real say or representation.

I can't get rid of the damn data centers; I'll have my self-driving cars which don't meaningfully differ from a tram because you can't self-drive them everywhere anyway; I'll have my personal robot butler with me 24/7, wiping my ass for me and all the rest of it. And in that world, there's no realistic ulterior option. Not when the nonsense has progressed far enough.

The problem with this isn't so much tech improving and the times a-changing or some other garbage like that. The problem is culture, and nobody gives an American the right to tell me how to think.

Kinda like you can't really live very well without some motorized transportation. Everyone expects access, or you're out of a job, you see. It's all culture – culture protecting the powerful, and you dancing to someone else's tune. And now the same thing that's generally true for locomotion is coming true for computing. Tell me if that isn't worth hating.

Right. But let's say it works. So what exactly does work? Is it transformers, LLMs, support vector machines? Is it AlphaFold? Is it data science with MoE? All those things are fine. Those are some cool algorithms and techniques, some more useful than others.

But what if "it works" means the super rich are going to absolutely make sure nobody's going to do any business without asking them for permission first? What if "AI works" means more social segregation? What if it means you can't even enter a business without huge capital investment, thus setting up arbitrary structural limitations that did not use to be there? Having access to your own LLM does little to hinder a strategy like that. And having global access to these things may just prove worthless for the individual, after all.

You see, I don't care much for some entertainment use for AI, although I'm not against it either. From my perspective, that's not important, because I can see myself enjoying entertainment with or without AI algorithms involved. It's perfectly fine if you want photorealistic water animations in your childrens' movies, just great, just fine. Everything lookin' pretty. I don't know anyone hating AI because of something like that.

Rather, it's like this: if you've even run your own mail server, then you know very intimately how you're at the mercies of your internet provider: they get to decide whether to change your IP address (CGNAT etc.), they get to decide whether to block port 25, and they get to decide almost to an arbitrary degree how much it costs to get around these limitations. All in the name of preventing spam. And so, they benefit from spam being a problem, all right? Because if it wasn't for this technological limitation, no one would pay extra, and all internet access would be the same. It's exactly the same with popular AI, too.

It's when a single entity controls all of the infrastructure you use all the time, and has access to advanced AI algorithms. You can't compete with that, assuming people keep preferring whatever comes out of it. Come to think of it, it is rather like someone owning your car, and you having to pay in order to use it. Or like someone having rigged cameras inside your home just so they can advertise to you based on your bodily functions.

That, and all the standard complaints about rising prices for hardware components, electricity, internet connectivity etc., assuming you want maximal scale for that sweet, sweet AI.

I never did understand how advertising made the tech bros so rich. I suppose it's the dependency – that everybody absolutely needs some of that visibility in order to even exist and do business online.

You see, I'm going to be honestly cynical and say that everything is politics, everything is arbitrary, and nobody actually deserves what they got. It's because when you make a decision with advantage, chances are you succeed. So then, if you make those advantageous decisions repeatedly, you'll eventually reach a position where you will have won most of the time. This is determinism, not merit. It's a statistical mechanism, and not good sense due to some individual. And if you personally lack the ability, just ask your machine learning algorithm for advice! Just because you're big doesn't give you any moral right whatsoever.

So just to spell it out here: owning all the relevant infrastructure + AI means maximum dice throws means you'll just win. Or at least, that seems to be the plan, because otherwise it couldn't be worth the investment. And that's really clue number one for any "real and true" motivations behind any of this.

Of course, my primary motivation for writing about this topic is that I want to live in a fair world where I personally have a realistic chance to take over the market, and to be able to completely destroy Alphabet, Microsoft and Meta, presuming my product is good enough. Not that I'd necessarily care enough to do that, but otherwise, what's the point? Otherwise it's not really a competition, is it?

And you know, I'd love for someone to actually challenge what I said earlier, about everything being politics. But frankly, that's what everyone seems to argue for.

But I digress, hating AI also kinda means nothing of course, because the term AI itself doesn't exactly mean anything. I largely see it as this smokescreen, intended to misdirect and thrill, kind of like how bread and circuses used to work. Except I actually enjoy circus. It's that smoke there is what gets into your eyes sometimes.

If only someone had really studied computer science and understood the fact that scaling up means a tradeoff, and that you could very well argue for reasonable maximum limits for your hardware, compute, and power expenditure. But, as it stands, there seems to be none of that.

No, it's like the field of traditional algorithmic analysis is taking the way of baby Jesus, eventually crucified, even though it did nothing wrong.

Shouldn't this be extremely ironic and hair-raising? Because traditionally, people have been concerned about time complexities, space complexities, and other bounds, dedicating large swaths of research to algorithm design. Instead, now none of that seems to matter in the face of "the race." According to everyone and their mother, it's the new industrial revolution, as though that comparison was supposed to be clever.

All that algorithmic analysis, thrown into the dump. Ignored, because not sexy, not in fashion.

And indeed, the name itself, "artificial intelligence," seems misleading on the face of it. If you think about what intelligence means, it's the ability to reason quickly and from large amounts of data. That's what animals do. But how's that differ from any old algorithm? If you go ahead and read Artificial Intelligence by Russell and Norvig, the very first thing you notice is that basic breadth-first search itself is considered artificial intelligence! But you won't hear that taking your algorithms 101 class. A smokescreen, I say, a smokescreen.

And moreover, if AI succeeding means I'll have to adopt it just for work, that'll change the picture of what it means to work in the IT field; it may change so much that I never would have chosen it had I known it to become like this in just a few years' time. But the thing is, I've already wasted enough of my time studying this stuff. All this change does is that it takes the control off my hands and forces me to keep myself preoccupied, learning new stuff nobody knew about ten years ago, let alone twenty, when I could instead be learning about something originally interested me.

But what "stuff"? You see, this is the important bit right here. Let's say we already have algorithms that do something great, and they'll get us 90% there, plus you get to keep your job regardless; we have an incredibly simple solution that wastes no electricity and just works. It's brilliant. So then, what happens when we sprinkle that AI jizzyjazz on top? We get an incredible 99.99999% performance, getting us right there to the tippy top, but the electricity bill is like 1000x what we used to have, and now everyone's just gotta use that new, difficult-to-explain nonsense solution that requires a goddamn €100,000 server rack to compute.

But you know, I'd totally flash the green light still and say that it's all good if it really were for some scientific purpose with a reputable university behind it; some one-time large computation. Sure. Or even some crowdsourcing thing or whatever. But nooo. It's commercial. From top to bottom, it's commercial.

And you know, we'll be reaching this incredibly awkward phase sooner or later where your personal computer is going to be the ultimate scamming machine – deepfakes and all the rest – which is when your tech overlords finally gain their moral justification for "all computation must be done in heavily regulated data centers and your device will be just a browser." A browser in your pocket, a browser with a keyboard, a browser on wheels, whatever.

That'll be the moment when general-purpose computing also dies. So if you actually like programming or you want to study software someplace, you better get politically enlightened, and fast.

Because the scaled-up AI solution either works or it doesn't, yes?

Yes, that's right. Popular AI is not supply-and-demand. It's not emergent. It's not "for the people by the people." So what is it? It's mostly top-down strategy. Dictated from on high by your lords of destruction.

It's strategy because genuinely gullible people believe in revolutions.

And sure, I suppose I should explicitly acknowledge that AI doesn't exactly explain job loss generally, although other people have surely figured that out by now. Sometimes it's used merely as an excuse. Sometimes it's market lingo used to drive stock prices up. It all kinda depends. The stock market is basically duplicitous nonsense all the way, allowing for things like bubbles to form up.

In order to understand how things got into this stupid state, I suppose one should remember the pathetic situation in which the American tech industry was before the AI nonsense took over. Back when everyone was asking the same old question of what the next big thing was going to be, as though there needed to be one. So if we could at least raise a shitton of money, we could experiment and see what happens to AI, scaled to the max. Or so the story goes.

So you get this distinct impression it's just not going to be a growth industry anymore. Or at least not without some skillful gymnastics. For example, if you think about all the shops used to be there around town, that's how many social media platforms you should statistically expect to exist – all the way until "social media" no longer means a thing and it's going to be "homepage" instead.

The thing is, when humans desert one place, they move on to the next. Unless there's nowhere to go, of course.

Ah, but I also know of cases where AI really has replaced human workers already, whether you consider that one of its successes or not; and those guys aren't exactly fulfilling their dreams now, by the way – well, not yet. And I don't actually know of anything that could prevent automating all the functions of the entire society in principle. That "in principle" being the operative term there.

It's just that going all the way to max automation doesn't really hold up next to real human dreams and goals. So, newsflash time: it's not all about the money. It's not all about the number going up. Although those things do matter.

I'm just repeating stuff I've said a billion times before, but you have to think about what's valuable. And in a sane economy it's the valuable thing people actually pay money for. Otherwise, if we're not in that sane world, it could be the least bad option people pay for, or someone following a recipe they think is successful, or people pretending to innovate so that others can pretend to like it. Hell, if you're desperate, that's still better than poverty.

Turns out total corruption, arbitrary violence and tumult is still better than poverty.

And then there's the classic, abused question of "what if there's no human in the loop?" Yeah, well, that's exactly when I'd start demanding instead of asking. Because of this thing called "leverage." Because when you find yourself in a position with none, you'll have absolutely nothing to lose. And I sure as hell won't take my trips abroad with VR goggles.

Yeah, so what is the social contract exactly? Hindsight 20/20.

Years and years ago, when I was but a kid of eighteen, I used to think technology was so great, so cool. I remember I wrote my first chat program with Java, and Final Fantasy X was considered a big production. Man, those were the days. Tech gave us so much freedom, so much insight into other lives. Tech gave us GPS maps everywhere. All these incredible showcases for technology that could be used anywhere, promising a long and prosperous career in programming for anyone masochistic enough, willing to withstand the tedium.

You know, way back before everyone and your fricken' grandma used Slack.

Anyway, now the story is almost the opposite: every country, every person is the same. The same personality everywhere, the same nonsense about Gen Z this and the Millennials that, the same stupid spoon-fed values, everyone using the same programs, the same computers, the same cars; and nothing is truly valuable in a global market economy except monopolies and maybe bullshit. How's that even add up?

Have you ever wondered about what happens if you truly cut the slack from your local economy? When you truly optimize to the hilt? How many people will simply not be able to make it, because they relied on bullshit, they relied on crime, they relied on everybody turning a blind eye? Yes it's structural, all of it.

It's stuff like this, why I started actively meditating, looking for inner peace in the teachings of Lord Buddha. But my faith is hanging by a thread, it seems. There's so much arbitrary competition, egotism and war; the same old grandpas still holding on to power even well into their 70s and 80s. And I suppose no one worthy has stepped up to take their place. The youth is so fluffy, you see. So immature. You can't trust them to fetch your coffee.

So tell me, how is AI going to help with any of that? I mean, it might. But it hasn't helped me personally. If writing some Torch code with Python is going to make me rich, I promise I'll change my tune.

Put it this way: is there a recipe I can follow so I'll get something I want in return? Can you tell me? I've been looking for that recipe, that character string my entire adult life. You see, once you have that recipe, you'll of course follow it to the letter, because whatever you want is whatever you want, and that's the way you'll get it; that's the deal. But in the meantime, you get to experience hell. That's just how it's going to be now, I guess. Oh, look: it's your friend the AI, helping you reduce everyone into replaceable blobbedyblobs.

But all this replaceability talk should be incredibly vain and ironic, shouldn't it? The one point no one hones enough is the following one right here: what does commercial, popular AI help you get that was categorically out of reach with classical, traditional methods? You used to get clothes, now you get clothes; you used to get entertainment, now you get entertainment; you used to get housing, now you get housing. So, guess what: AI is not the same as the industrial revolution. Not even in the long term.

The thing is, some people seem to think we're on "a superexponential trajectory" with respect to AI development. No, we're not. The next levels of improvement with respect to some LLMs may happen quicker but only because the big companies are using automation to help build more automation, which is generally going to eventually be quicker than using human labor to do the same. Physical constraints don't go anywhere just because. The best you can do is to take shortcuts, but the broader laws still hold. I especially recommend Information Theory, Inference, and Learning Algorithms by David J. C. MacKay, learn all about it. It's a bit dated, but one of the best books I've read tying together information theory and machine learning. Some of the examples are a bit complex though, it's not the easiest read. But that David guy = smart.

And of course, the elephant in the room is that big countries want AI because they want robot armies. The idea is that this should translate to power – you know, because of all that ego seducing you all the time. There's practical problems though: more complex algorithms require more memory and more electricity, which kinda translates back to size, mass and fuel consumption, all that stuff the physicists love so much. So there are natural limits to what robots can do. There's a reason why drones usually just fly directly towards the target and explode.

Again, it's one of those things where I don't exactly contest the possibility of terminator-style robot armies in principle. But the only real way to deploy 'em would be some kinda lightning-warfare or stealth-style approach together with human engineers riding on the coattails. And that's absolutely something other humans could prepare for. It's also the one strategy to work because humans aren't prepared for it. So maybe with enough planning, who knows. But then again, if your sole aim is to catch your opponent with their pants down, what do you need AI for?

And if it's so dangerous that a company is building robots, then the government itself might as well burn it to the ground, because why would they want to give up power in favor of some AI jackoff? Because it either works or it doesn't, remember?

The point being that the chain of trust didn't go anywhere.

Of course, there is some delightful irony in that if you'd just hire some anti-AI dissidents to burn down those humanoid industrial robot factories, you'd be doing the local population a favor. Because from a company's point of view it's that you either pay for human or you pay for robot. Human be somebody else's problem.

I'm not even all that sure whether I'm being too presumptuous here. I've seen some shit over the last years, so maybe not.

You know, it's somewhat problematic, but in some societies, being an expert or in an authority position, people actually listen to what you have to say. They trust you. And there is generally no discernment between actual facts and your self-serving marketing pitch. That's just horrifying, I know. So they'll be at your mercy.

Another interesting point of view into this is how you could use large AI models to find software vulnerabilities. You know, security? However, there's a few simple questions you'd need to seriously ponder before jumping to any conclusions about the relevance of this aspect.

First, let's say a large model does find a nasty zero-day vulnerability. Fine. That's scary, I admit. But is the vulnerability remotely executable? Does it allow arbitrary code execution? How probably can you trigger the vulnerability at all; does it require other vulnerabilities that you'd need to chain together in order to even leverage it? And of course, what are the chances that, after your huge AI model has exposed the vulnerability in the first place, absolutely nobody is going to fix it? So you see, it's exactly the same as with humans: it's an ethical question of whether you'd patch the vulnerabilities you discover or not. It's still about who to trust when it comes to your software patches arriving on time, just like it's always been. And there are lots of vulnerabilities found all the time, with or without AI involved.

Some asshole hacking your computer is still some asshole hacking your computer.

In case all I've said was somehow unconvincing, just think about this last little question but for a moment: if the great big leaders of American tech companies know for sure that artificial intelligence is going to give all of humankind limitless wealth and prosperity in the form of "the singularity" or whatever, then why would it matter to make the biggest of AI companies into a for-profit? Why would it matter, fool?

I guess it must be nice to make billions of dollars.

No, the way I interpret this is that all those rich tech bros owe me some money. Where's my fucking money, Elon? Where's my limitless pussy, Altman? You said I get to have all my fucking dreams come true!

Not that those guys even know what my dreams are, just to make that perfectly clear.

So how about some evidence?

Anyway, that, my friend, is the story of how I slowly started hating AI. I actually have even more reasons, but I've kind of written about those already, and I've dragged this on for long enough.

Suffice to say that most of what I truly hate about AI really boils down to how software in the public interest just isn't the same as this blind lust after productivity increases – especially considering we're supposed to be the 1st world with all this prosperity and abundance – unless it's all a lie, of course, and basically nothing is truly built on productivity anymore. Indeed, you probably get less and less in expected returns for some constant amount of productivity increase, plus if there's more competition now than there used to be, that's going to exacerbate your chances.

Now, if you'll excuse me, I'll go and have a therapy session with my AI therapist companion who is also my lover. Way cheaper than going to real therapy. 100% much recommend.