The Emperor’s New Neural Networks: Why GenAI Isn’t As Smart As You Think

Remember that thought experiment about infinite monkeys with infinite typewriters eventually producing Shakespeare’s complete works? Well, congratulations — we’ve basically built that, just with better PR and VC funding. Welcome to the world of Generative AI, where we’ve replaced the monkeys with linear algebra and called it intelligence. And yes, I’ll be talking quite a bit about AI not being intelligence. Strap in.

The Not-So-Secret Sauce

JLaw wants to know what you mean — Credit Sony Pictures

You know me, I love definitions because words have meaning.
“Artificial Intelligence” really covers a broader term that can be loosely explained as “using tools/computers to execute intellectual activity at scale”. It is not intelligence in the sense of “the ability to acquire and apply knowledge and skills”.

So let’s be clear about what’s actually happening under the hood. These “intelligent” systems are essentially sophisticated pattern matching machines, playing an incredibly complex game of “what word comes next?” It’s a bunch of IF loops, on algebraic steroids. They’re not thinking, creating, or understanding — they’re predicting based on statistical patterns they’ve seen in their training data. It’s like having a friend who’s really good at finishing your sentences, not because they understand you, but because they’ve memorized your speech patterns.

Garbage In, Shakespeare Out?

Sweet, sweet digital garbage — Photo by Carl Campbell on Unsplash

Speaking of training data — here’s where things get really interesting (and by interesting, I mean concerning). These models are trained on vast amounts of internet content, which is about as curated as a teenager’s bedroom. We’re talking about everything from scholarly articles to Reddit threads about conspiracy theories, all thrown into the same digital blender.

And don’t get me started on intellectual property. Just kidding, I’ll be ranting about that in the next section.

The quality of this training corpus is, to put it mildly, questionable. It’s like trying to learn about world history exclusively through memes and YouTube comments. Sure, you might occasionally stumble upon something brilliant, but you’re just as likely to generate convincing-sounding nonsense.

The Great IP Heist

All your text are belong to us — Photo by Markus Winkler on Unsplash

And let’s talk about the elephant in the room: intellectual property. These models are trained on countless copyrighted works, personal blogs, and private communications, often without explicit permission. It’s as if we’ve created a digital burglar that breaks into the library at night, photocopies everything, and then tries to sell you remixed versions of what it stole.

The legal and ethical implications here are staggering. We’re essentially building commercial products by scraping and repurposing other people’s intellectual property. It’s like claiming you’re a chef because you can throw everyone else’s recipes into a food processor and press blend.

So when Sam Altman of OpenAI clutches his pearls when he discovers DeepSeek AI is trained with ChatGPT, he fails to see that his own platform would not even exist if it hand’t scraped data off the open web, without asking for consent or respecting robots.txt guidelines.

Privacy: The Final Frontier (That We’ve Already Invaded)

Uncle Sam is watching you — Photo by Tobias Tullius on Unsplash

But wait, there’s more! The privacy implications of these systems are enough to make George Orwell say “I told you so.” Personal data, private conversations, and sensitive information are all potentially part of these training sets. Your private blog from 2008? Probably in there. Those forum posts about your medical condition? Yep, those too.

Those videos with your face and voice on them? Congrats, you can now be deepfaked for identify theft, cleaning out your bank account, and Goddess knows what else.

It’s like we’ve created a massive digital eavesdropper that remembers everything it’s ever heard and isn’t particularly discerning about how it uses that information.

The Heat Death of Innovation

Stuck in time, like your data — Photo by Aron Visuals on Unsplash

Generative AIs and LLMs are impressive content recycling systems, sourced on whatever data they were able to feed on. This means that, as people get savvy to how those bots work, systematic IP address blocking will take place so that no new fresh content will make its way to the data training pipelines.

Remember when ChatGPT only offered output based on content from 2017 and earlier? This could well mean “pre-2025” at the time of this blurb, especially if the IP blocking becomes a habit with web hosts and site owners. If that comes to pass, and it needs to, every LLM-based output will be based on stale, outdated content.

The only to make this work is for sites, and individuals, to start monetizing content and data so that AI companies pay their fair share. After all, they’re making money off your content.

The Reality Check

Photo by Elimende Inagella on Unsplash

Don’t get me wrong — these systems are impressive technical achievements. The mathematical and computational frameworks behind them are genuinely revolutionary. But calling them “intelligent” is like calling a calculator a mathematician. They’re tools — incredibly sophisticated tools, but tools nonetheless.

What we’ve built isn’t artificial intelligence; it’s artificial pattern matching on steroids. It’s the world’s most impressive autocomplete, trained on the world’s biggest dataset of human-generated content. And while that’s remarkable, it’s not the sentient digital brain that some would have you believe.

Moving Forward

Data wasteland — Photo by Lukasz Szmigiel on Unsplash

As we continue to develop and deploy these systems, we need to be honest about what they are and aren’t. We need to address the serious concerns about training data quality, intellectual property rights, and privacy. Most importantly, we need to stop pretending that we’ve created actual intelligence when what we’ve really created is a really good party trick.

As a species, we’ve always created tools, extensions of ourselves that would make tasks easier. What’s what made us different from the rest of the animal kingdom. (Gen)AI is not another “us”, not another intelligence no matter how you look at it.

So the next time someone tries to convince you that GenAI is thinking, creating, or understanding, remember the monkeys with typewriters. We’ve just replaced the monkeys with matrices, the typewriters with TPUs, and called it a revolution.

The emperor might have new neural networks, but they’re still not clothes.