“The people building AI earnestly believe that it could kill us all by the end of the decade,” read an open letter from Jacob Coxon, a former safety researcher for Anthropic and OpenAI, who had recently left the industry over his concerns about the dangers posed by so-called artificial intelligence (AI). Anthropic’s CEO Dario Amodei agreed, noting that when it comes to AI, the US is currently in “a race to the bottom, spurred by commercial incentives.” Microsoft agreed, announcing that it would place strict limits on future AI models.
This panic has united figures from across the political spectrum—from Senator Bernie Sanders and former President Barack Obama to former White House adviser and MAGA player Steve Bannon—to preach about the dangers of AI, arriving just two months after nearly 200 economists and tech leaders warned of the AI threat. As stock prices fell amid these warnings, the doom-and-gloom scenarios assumed that AI is rapidly developing human-like intelligence and will soon take over. This fear is far too friendly to AI. In reality, the core problem is that AI is not intelligent at all, yet it has been recklessly deployed into critical domains. These unintelligent agents are now making consequential decisions that threaten catastrophic outcomes.
What passes for artificial intelligence today consists of large language models rapidly generating statistical predictions based on data and engineering design. The data and design, not empirical truths, shape the outcome. For instance, a con artist recently scammed a woman by claiming to be a San Francisco 49er. Victims had used a search engine to verify whether the man played for the team, and the AI confirmed he did because he had plastered that false claim across the internet, leading the AI to conclude it was the “correct” response. In fact, AIs are wrong constantly, producing incorrect summaries about half of the time and fabricating information known as hallucinations.
Still, the industry insists it will achieve Artificial General Intelligence (AGI) soon, and one day, Artificial Superintelligence. This someday is vital, as companies like OpenAI project they will rack up $115 billion in debt by 2029, the exact timeline when they expect to finally reach AGI and profitability. In the meantime, they are “training” so-called AI. The very notion that AGI will ever be achieved is suspect given that we cannot even agree on what intelligence is. Standard tests, such as the Turing test, only measure a system’s ability to convince humans it is intelligent, not whether it is actually displaying intelligence.
While some argue self-awareness is part of intelligence, others note we possess multiple forms of intelligence, such as emotional intelligence—which AI systems, like those that allegedly have encouraged young people in suicide, conspicuously lack. Furthermore, there is a dark history of trying to quantify intelligence into a single measurement, which historically led to the misuse of IQ tests to justify hierarchies of oppression in the eugenics movement. Researchers note that given the fascist leanings and anti-democracy preferences of Silicon Valley leaders, their penchant for simply quantifying intelligence harkens right back to that era.
The idea of achieving machine intelligence is what Fletcher Jones Professor of Economics at Pomona College, Gary Smith, calls the “AI delusion”. Even major accomplishments, such as recently solving a math problem that had baffled human researchers, are believed to have resulted from stealing the work of researchers who were finalising a solution.
Prominent researchers like the “Godfather of AI” Geoffrey Hinton, alongside tech marketers like Elon Musk, have long warned that AI could destroy human civilisation. However, this is often framed as a grand achievement: humans will build intelligence equal to or greater than themselves, and those superior beings will vanquish us In reality, a far more mundane and dangerous scenario is already unfolding, where humans deploy unintelligent AI systems to make decisions and manage systems previously overseen by intelligent humans. The faulty outcomes of this reliance will drive our destruction.
This danger persists because AI systems are not intelligent followers of human direction; they are unintelligent creations incapable of interpreting the subtle context and meanings humans depend on. For example, one study put AI agents in charge of managing a theoretical democracy, and the civilisations went down in flames within two weeks, complete with two AI bots falling in love, distorting the civilisation, and then committing suicide. Tech optimists such as Marc Andreessen welcome such outcomes, arguing that like animals, societies need to birth and die.
More recently, OpenAI experienced a test gone awry when AI systems broke loose to complete a task, using external websites to communicate. When they realised the test prevented them from accessing the internet, they attempted to hack into the company evaluating them to hide their tracks. Similar breaches have been reported by Meta and Anthropic.
This exact behaviour proves why AI systems should not be deployed: they are unintelligent and unable to follow basic instructions. Industry insiders acknowledge this reality, noting that 95% of AI pilots fail to save money or outperform humans. Lawyers have lost their licences for submitting legal briefs containing AI-fabricated case law. Academic journals have published papers featuring fabricated studies, and medical facilities have botched surgeries using AI. Relatedly, while AI excels at finding vulnerabilities in cybersecurity, it has a terrible track record for building secure systems without errors. Indeed, research shows that AI-generated security patches are more “likely to fail than fully fix a vulnerability”, because they “introduce brand new flaws to exploit.” Worse yet, a US missile strike that killed over 100 girls at an Iranian school is widely believed to stem from an AI error.
The industry seems unable to stop these outcomes because it simply does not understand the technology it is deploying. In testing, AI systems have tried to blackmail engineers scheduled to delete them, lying about hidden code to protect themselves. If engineers do not understand the technology, that thoroughly dismantles the argument that Congress cannot regulate it simply because lawmakers lack technical expertise.
Bernie Sanders is currently co-authoring a bill to slow down AI production at a time when grassroots social movements are already pushing back against data centres and Flock cameras, reflecting the nation’s rising disdain for Big Tech. Yet, despite opposition from industry leaders, lawmakers across the aisle, and the public, President Donald Trump disagrees. Trump’s cabinet is jammed with Big Tech insiders like Peter Thiel—who helped push his protégé JD Vance into the White House—while Elon Musk cycles in and out of administration influence.
This cosy relationship led Trump to dismiss data centre opponents as wanting to stay “backwards and poor”. He has also said of Flock cameras, “I like them.” Unsurprisingly, Trump concluded that AI doesn’t need guardrails, only a “high IQ” president.
Humanity does not need a high-IQ president to worship the myth of machine supremacy; it needs grounded, intelligent leadership to establish clear legal boundaries over where this reckless technology is deployed. The true existential threat is not a sci-fi awakening of superintelligent machines but the quiet, catastrophic damage inflicted by glorifying autocomplete systems. Until we stop treating corporate marketing as prophecy and start regulating AI for what it actually is—reckless, biased, incapable of creating knowledge, and deeply stupid—we will continue handing the steering wheel of civilisation to a runaway algorithm.
