People believe what they want to believe – as Julius Caesar wrote in his account of the Gallic Wars. It’s human nature, amped-up more than two millennia later on social media so biases and conspiracies go unchecked and unchallenged, made potent by algorithms in echo chambers that reward outrage – owned by powerful companies with rampant agendas to make unfettered profit from total political, economic, and social disruption. Breathe. Trust in institutions has crumbled: the police, the courts, the media, even governments. Scientists lie; there is ‘fake news’ and fake news – so just don’t believe a word. And, stop the presses – because AGI is here.

That was the headline (everywhere) yesterday evening: that Nvidia (AI) supremo Jensen Huang, echoing Sam Altman last month, has called it – that we have somehow achieved AGI. But what does that mean? It brings to mind HAL from 2001 and Skynet from Terminator – art created by baby boomers, inspired by black-and-white footage of moon landings and atomic bombs, and a technicolor sci-fi movement going back to Frankenstein and the first industrial revolution. But it was always just art; the tech has suddenly advanced so far and fast that the sci-fi (or business of it) seems almost real. But is it? Because AGI is a theoretical concept – believe what you want to believe. 

For Huang and Altman, easy to love and hate on social media, the argument goes that there is enough AI out there to surpass human intelligence. “I think it’s now. I think we’ve achieved AGI,” Huang told computer scientist Lex Fridman. AI could run a company, he suggested – although agents come and go, he said, and they couldn’t build a company like his. So there is that – human intelligence still wins, let the record show. And Microsoft chief Satya Nadella has said the opposite; that the industry is “not anywhere close” – whatever Huang might say. But there are noisy advocates for both sides on social media, and very little in between.

Check your slop-feed; the debate around AI, like everything, is polarized: it is either grossly overhyped, or on the point of armageddon (or economic rebirth or corporate nirvana). Frontier models and software agents are fancy pattern-matching systems; statistical approximators, lossy compressors, terrible thieves of all human intelligence. Plus, they are designed to hallucinate. There is no AGI, no HAL – no self-aware digital mind. The talk about AGI, or clickbait coverage of it, drives unrealistic expectations. PwC has just told staff to get on board, or go jump – which also needs explanation. On social media, especially, the human response lacks reason. 

Which is the problem, here: the human mind, whether it is its perception of reality, or its programming of it. Because AGI or not, these pattern matchers are getting smarter, and increasingly joined up. Even inter-linked functional non-AGI at scale, where agents code agents, will get out of hand, and things will go wrong – and so, whatever you call it, guardrails are important. There is a video doing the rounds on social media of a humanoid robot gone rogue in a California restaurant; it gets too close to the diners, so that its dance routine looks like a kung-fu attack, and it has to be frog-marched out of there by an embarrassed waitress. Online, everyone has something to say.

The point is that systems are designed, not inherently rogue. Automation failures happen because humans misconfigure, ignore, trust – not because the system has agency. As it is, ‘rogue AI’ is mostly a shorthand for preventable engineering issues. Human error – in perception, whether in the bubble or the guts of the system. Even at scale, careful design and testing, and in-the-loop provisions, can prevent cascading errors. HAL-style apocalypse, where the ultimate mission is concealed from the crew, assumes perfect leverage and uncontrolled agency – which real systems do not have. Having said all of that, human imagination is correct: downtime disrupts, somewhere. 

Total human-style paranoia is the correct response – and regulation and guardrails.