Hype train blues

The AI hype train shows no signs of slowing down. Microsoft AI CEO Mustafa Suleyman recently sounded the death knell for white-collar workers in an interview with the Financial Times, giving us 12–18 months before our jobs are completely absorbed by AI. He explicitly mentioned lawyers, accountants, project managers and marketers. Thankfully not academics or journalists, but given what else he said, I’m pretty sure his bell is tolling for us as well.

That was back in February (so that’s now 9–15 months away), but for some reason, Suleyman’s prediction is getting another run in the business and tech media this week. Perhaps it’s just a financial-press version of the giant-shark attack stories running in the Mirror. Whether it’s human slop or AI slop, it’s slop all the same. Slop about AI. Metaslop.

Like most AI executives, Suleyman is no stranger to hyperbole. Just ten days before the FT interview went to air, he warned against overhyping the capabilities of AI. This was in reference to Moltbook, a just-launched social network where AI agents could hang out with each other and chill, or maybe plot the downfall of humanity. Elon Musk, in his irksome way, called it ‘just the very early stages of the singularity.’ But really, it’s just LLMs given directions to find stuff on the internet that will help them do particular things. And as it turned out, many of the supposedly autonomous posts were not really autonomous, in the sense we might ascribe to humans. Instead, it was a bunch of AI agents who’d been told to act like they were humans, even to act in specific ways, like inventing the kind of code language a child might come up with. Certainly not the birth of a new language, as one AI commentator noted, but ‘a statistical sink where overlapping probability distributions converge on common attractors,’ particularly given the pervasive puppeteering.

Yet, despite urging that these agents were certainly not conscious, Suleyman downplayed the reality. ‘No matter how autonomous it was, in hindsight, it doesn't really matter,’ he said, going on to describe some fairly prosaic (for AI agents) things that they might be able to do in the near future, such as writing their own code or making phone calls to one another. After all, Suleyman has put himself in charge of a new Microsoft ‘humanistic superintelligence’ project, which will ensure that AI is used for good, even if it’s way smarter than us, because it will be restricted to particular usage domains.

And this superintelligence (presumably Microsoft’s) is what is ringing the unemployment bell, heralding the arrival of AI with ‘a human-level performance on most, if not all professional tasks’. But look a little deeper, and the tasks Suleyman imagines here are also pretty prosaic, rote tasks that you do on the computer, like writing word documents, filling in spreadsheets, and looking stuff up on the internet. Medicine, his primary use case, will be transformed, he suggests, by AI which can ‘reason over all the contents’ of a patient’s health record based on their knowledge of current medical science, and come up with an accurate diagnosis most of the time. But it’s not that doctors will be out of a job. Instead they will be given over to patient care. As if that isn’t their most fundamental ‘professional task’!

So, we might ask, what makes this superintelligence, rather than just doing some things really fast, like responding to natural-language prompts with statistically probable answers? If AI is not really conscious, hell, not even really autonomous, is it even intelligent? It might seem we’re splitting hairs about what ‘intelligence’ or ‘autonomy’ really mean, but it’s the vagueness in these terms that AI hypesters trade off. And keeps that hype train rolling on.