The AI Takeover of Humanity

Technology journalists, the mainstream media, and even the Pope are warning about the threat of AI taking over or destroying humanity. This fear is largely fueled by the AI industry itself. In the scramble to inflate their valuations before their impending IPOs, they're trying to trick everyone into believing that the statistical language models they're developing are far more capable than they really are.

When people think about AI taking over, they envision Skynet from Terminator, SHODAN from System Shock, the machines from The Matrix, and the thinking machines from Dune (HAL 9000 usually falls into this group, but if you've read 2010, you know HAL wasn't evil). All of these are artificial intelligences that developed sapience and used their superior intelligence to either wipe out (or at least attempt to) or enslave humanity. What we're referring to as AI right now isn't that. What's being sold to us as AI isn't even intelligent. ChatGPT, Claude, Grok, and other popular software packages being billed as AI are really just large language models. They use statistics derived from training data to guess which word should come after the next based on provided input. This results in grammatically correct responses (which is a miracle considering the training material) with dubious correctness.

AI does stand a real chance of taking over humanity. However, it's not because AI is sapient or even somewhat intelligent. It's because the majority of humans are incredibly gullible. The best illustration of this fact in the context of this post is the ELIZA effect:

In computer science, the ELIZA effect is a tendency to project human traits—such as experience, semantic comprehension or empathy—onto computer programs. ELIZA was a symbolic AI chatbot developed in 1966 by Joseph Weizenbaum that imitated a psychotherapist. Many early users were convinced of ELIZA's intelligence and understanding, despite its basic text-processing approach and the explanations of its limitations.

History doesn't repeat itself, but is rhymes. The main limiting factor for ELIZA was the technology of the time. Computers in 1966 were far and few between. That meant the number of gullible people who interacted with ELIZA were limited. A secondary limiting factor was morality. Joseph Weizenbaum didn't lie to people about ELIZA being more capable than it was. He wasn't trying to increase the valuation of some company so he could make off like a bandit after an IPO. Instead he pointed out that the people attributing human traits to ELIZA were gullible fools (although he probably used a nicer descriptor).

The unwashed masses (I'm not going to refrain from mean-spirited descriptors) seem to be in a hurry to outsource their thinking to anybody or anything. Historically this has taken the form of the masses parroting whatever a philosopher, religious leader, or politician said. Today it's taking the form of parroting whatever an AI generated. Asking a human being a question often results in them asking an AI and parroting the answer.

The masses aren't relying on AI only for their talking points. They're listening to AI generate podcasts and music, reading AI generated novels, and watching AI generated videos. Soon they will rely on AI to tell them how to spank the monkey. Writers (I use the term loosely) are letting AI generate their articles and novels.

Companies are also rushing to outsource as much work as they can to AI. Amazon is pushing employees to use AI so fiercely that its employees are gaming the system by using as many token as possible to improve their performance metrics. The same link mentions how Meta is ranking employees by how many tokens they use. Even if they're not necessarily making it a performance metric, many large companies require employees to use AI. Soon companies will use AI for marketing and creating influencers (looking at human influencers, this might be an overall improvement).

I could go on, but I believe I've illustrated my point. AI stands a real chance of taking over humanity, but not in the way most people are predicting. We're not looking at a sapient hyper-competent AI emerging in the near future and executing a brilliant plan to eradicate or enslave humanity. We're looking at gullible humans outsourcing their thinking to the current batch of statistical models. Shackled humans won't be prodded into forced labor by armed robot overlords, they'll continue wandering aimlessly as they stare at their smartphones. The catalyst won't be machines developing sapience, it'll be human stupidity and laziness. Our future isn't Terminator, it's Idiocracy.

The only saving grace will be when the AI companies finally start charging profitable rates. Right now everybody is living high on subsidized rates meant to develop massive user bases. Eventually these companies will need to make money. Being technology companies, their preferred form of making money will be offering less for more (enshitification). That's when we'll see whether the unwashed masses and companies will continue to outsource everything to AI or instead see AI for what it really is: a tool that is appropriate for specific jobs.