https://consciousnessofsheep.co.uk/2026/09/16/ai-is-not-a-flyer/
There is a surprisingly versatile technology which, when used correctly, can predict 99.9 percent of bankruptcies. Its not particularly expensive – children’s versions sell for less than £1.00 on Amazon, and you can even download free app versions for your phone. Indeed, I have a free version packaged in with my computer operating software. It’s called a CALCULATOR. And it is a device so rarely used by financial journalists and so-called investment specialists that time and again we have witnessed implausible bubbles being inflated where basic maths would have seen capital deployed somewhere less exciting and where returns are a little slower to accumulate.

The first fracking companies were a case in point. Following the eye-watering price spikes either side of the 2008 crash, hydraulically fracturing the USA’s shale deposits began to look profitable. And with interest rates close to zero (and negative when adjusted for inflation) the fracking companies’ junk bonds looked like a good deal to an investment class with few alternative sources of yield.

Had anyone bothered to reach for a calculator to check the cost of fracking against the oil price needed to break even, they might have seen that fracking was yet another of history’s many examples of people spending billions of dollars to make millions of dollars in return. And there it might have ended… except that the people who matter – the investment class, the financial journalists and the numpties who inhabit our political institutions – work on narratives rather than maths. The establishment media was flooded with feelgood stories about “Saudi America” and the promised coming “century of energy independence” even as what was an old technology and long known shale deposits were being sold as new discoveries and futuristic technologies… it seems we are all suckers for stories about breakthrough technologies.

It turned out there was a huge gulf between the numbers that the fracking companies were selling to the investment class and the more accurate (because they carry legal penalties) numbers they were reporting to the US Securities and Exchange Commission. If the oil price had remained above $140 per barrel, things might have been different. But in today’s precarious and indebted economy, high oil prices force people across the western (i.e., consuming) states to cut back on their discretionary spending (a process which is currently unfolding on steroids now that the Trump administration has disconnected a fifth of the globe’s oil supply). Back then, it resulted in the world oil price falling below $40 per barrel, rendering the first fracking companies insolvent, and even resulting in a few prosecutions for fraud.

With the wells drilled and the equipment paid for, and with oil prices rising again, the companies that bought the bankrupt companies’ assets for cents on the dollar have been able to turn a decent profit from America’s shale plays. Although the rate of production decline has seen a retreat from most of the plays, with only the Permian Basin now promising additional oil. It seems that the USA is closer to a couple of decades of partial energy independence (it still needs to import heavier oils) than the century or so promised by the financial media at the height of the frenzy.

For a while, cryptocurrencies became the next potential means of detaching fools from their money. But while it attracted an army of quasi-religious fanboys and the usual nonsensical hype in investment media, it turned out to be too unreliable both as a currency and as an investment vehicle… and so the money flowed toward Big Tech instead.

The story though, is pretty much the same. Peak tech came and went about a decade ago with the peak of smartphone sales. And the smartphones themselves were about as technologically developed as it was possible to make them without their becoming so expensive that people would refuse to buy them. Indeed, for several years now, consumers have been complaining that the latest versions of smartphones have little in the way of new technology, but involve mere changes to the styling, causing a growing part of the younger generation to give up on them entirely… which is devastating in a US economy in which the mendaciously titled “AI” alone accounts for more than half of the S&P 500.

Step back and it is simple enough to understand the problem. Technologies tend to be driven by changes in the energy landscape, with particular energy sources enabling a suite of technologies – most obviously, the internal combustion engine (which had been theorised more than a century before one was built) was only possible following the production of large quantities of hydrocarbons not required for alternative uses. Diesel and petrol met that requirement. And since they were initially a waste product from kerosene production for oil lighting, they were available (in the USA) to fuel the growth of the automobile industry. Similarly, coal-powered steam locomotives would no doubt have come into use at some point, but the reason they came into use where and when they did was the consequence of an energy shortage and an alternative energy abundance.

In 1804, engineer Richard Trevithick’s steam locomotive pulled a 10-ton cargo of iron along with 70 passengers down the ten-mile track connecting the Penydarren Ironworks to the Glamorganshire canal quay at Abercynon. It is widely seen as the first modern train journey, although it would be a quarter of a century before the Liverpool-Manchester line opened. What connects the two is that after 1804, railways offered a work-around to a growing problem with the older canal network.

At just five miles per hour running down the Taff valley, Trevithick’s engine was hardly a competitor for canal barges pulled by heavy work horses at a similar speed. Indeed, Trevithick’s train had to be pulled back up the valley to Merthyr by a team of horses. But in 1804, and for the following eleven years, horses – particularly heavy working ones – were in short supply, having been drafted into the supply chains for the Napoleonic Wars. And, with a massive reserve of coal beneath the British Isles, steam locomotion began to be seen as a potential extension of the older transport networks.

By the 1840s, with competition heating up and with Parliament approving the construction of railways across Great Britain, the feeding frenzy began as Britain’s industrial nouveau riche queued alongside the old aristocracy to lose their fortunes on new railways with few customers and little chance of earning an income (if only someone had had a calculator… or at least knew how to operate an abacus). The bubble burst in 1847. But (like fracking and the DotCom bubble) the infrastructure had been built, and, at a slower pace, railways grew across the industrialised world.

The i-phone moment for steam locomotives came on 3 July 1938 when the over-engineered A4 Class Mallard set the steam train speed limit of 126mph. It had been a triumph of engineering, but it was an economic catastrophe because the service it was hauling was expensive and unprofitable. It was only the package of government subsidies, born out of the experience of industrialised slaughter between 1914 and 1918 which kept the service alive – essentially, Britain’s workers were paying for Britain’s rulers to enjoy a class of executive travel which they themselves couldn’t hope to enjoy.

If that sounds familiar, then you’ve probably heard of Concorde – the world’s first and last commercial supersonic aeroplane. Just as Mallard had come a long way from Trevithick’s 1804 locomotive, Concorde was several technological leaps beyond the 1903 Wright Flyer… and for much the same reason. It isn’t simply that new energy sources give rise to new technologies, but that those technologies follow an ‘S’ curve as they develop. Trevithick’s locomotive was not quite held together by string and canvass, although (except for its lightweight aluminium engine) the Wright Flyer largely was. But both, having proved their respective concepts were open to a series of relatively cheap and easy improvements.

Unfortunately, the progress narratives that we fall for time and again, are based on the erroneous belief that the process of technological improvement follows an upward slope. In the real world though, once the cheap and easy improvements have been made, further improvements provide fewer gains and come at an increasing cost. There are two big reasons why Mallard still holds that speed record – the freight businesses and passengers on Britain’s railways preferred to pay the lower cost of slower and older locomotives, while, in any case, steam was being overhauled by electric and diesel-electric locomotives. In a similar way, the development of jet engines paved the way for the brief (in historical terms) post-war period of mass air travel. But improvements were already getting difficult by the time the British and French governments decided to fund commercial supersonic flight.

In the media, Concorde was the wave of the future. It was, they promised, only a matter of time before you could fly from the UK to Australia in the same time as it would take for a rail journey from London to Glasgow. More than 100 provisional orders were placed by 17 international airlines. But then it turned out that supersonic commercial flight is a lot harder to achieve than was first thought. And with the onset of inflation from the 1970s, followed by the ban on supersonic flight over the USA, orders dried up. By the time Concorde made its first test flights in 1969, only the British and French were – mostly because of the sunk cost fallacy – still interested. And when the first “commercial” flights began in 1976, they had to be heavily subsidised by both governments, so that, just like Mallard they were a luxury service for movie stars and rich business owners subsidised by a mass of ordinary people who would never be able to fly on them. The miracle was that Concordes continued flying over the Atlantic for another 27 years before the plug was pulled.

By then, the need to keep fuel costs low, together with growing environmental concerns, meant that the new wave of aeroplanes would be slower but cheaper. And it turned out that nobody really needed to get from London to New York in three hours anyway. So that, with the exception of the occasional science fiction in one of the tech magazines, nobody seriously expects commercial supersonic flight to ever happen again. Concorde, it turned out, was at the end of its ‘S’ curve – already costing more than any benefits it was offering.

Which brings us to large language models (LLMs). These are to BBC Basic what Mallard was to Trevithick and what Concorde was to the Wright Flyer. Far from the wave of the future that the establishment media are desperately trying to make them, they are a very late stage of computer technologies which emerged publicly with the development of stable electricity grids in the 1970s. Famously, the early computers destroyed jobs while ushering in new jobs that nobody had even thought of back then. And nobody can deny the raft of incremental improvements that have made computers ubiquitous both at work and at home, and which allow us to carry out complex tasks which would have been impossible just a decade or so ago.

Nevertheless, like any technology, they have followed the ‘S’ curve of improvements, with all of the cheap and easy ones now far behind us in the rearview mirror. This is not a problem for we ordinary folk. But if you are the owner of a tech company or a godzillionaire investor desperately seeking yield, you need the technology to keep improving. And so, you might, for example, turn to virtual reality as a way forward.

In case you haven’t tried it, and despite a raft of tech magazine articles to the contrary, virtual reality is one of those branches of technology that was a dead end from the start. If you can afford the headset – which comes in at around £300 for the entry level – you soon discover that its two obvious drawbacks are motion sickness and the tendency to end up flat on your face chewing the carpet. But if you can get past these, you’ll discover that the two most popular uses are for playing a handful of not very good games and – inevitably – watching pornography. Despite these drawbacks, Mark Zuckerberg decided that there would be a multi-billion-dollar market for access to something he called The Metaverse… a kind of virtual mirror of the real world which we would queue up to pay Facebook – now rebranded as Meta – to allow us to access. Although anyone foolish enough to do so soon heard the sound of tumbleweed and found themselves wondering if watching paint dry might be more entertaining.

While there is some discussion of whether Zuckerberg’s Reality Labs lost all of the $80 billion thrown at the Metaverse, much depends on whether there is any interest in the various headsets still being developed, which might at least offset some of the loss. Nevertheless, $80 billion was spent and the Metaverse is no longer the wave of the future. Meanwhile, Zuckerberg has ridden off to join the other tech bros in pursuit of the holy grail of converting large language models into artificial general intelligence (AGI) or even artificial super intelligence (ASI).

This though, is where you need to reach for the calculator. Because, rather like the Metaverse, the tech bros – all of whom are a mix of P.T. Barnum, Elizabeth Holmes and Bernie Madoff – have been throwing hundreds of billions of dollars at a technology which promises to return a few billions of dollars at best… and nobody knows if it is even possible. LLMs are not intelligent. Nor, crucially, do they have a “self” which might at some future date become “aware.” At best they have been referred to as stochastic parrots – they are very competent at using statistical probability to tell you what you want to hear… a process which often leads to so-called ‘hallucinations’ which, in turn, have given rise to potentially costly legal precedent.

The big problem though, is debt, which will have to rise by more than $400 billion in 2027 and is projected to rise above $1.3 trillion by 2030. But as of 2026, total revenues are counted in mere tens of billions. The AI deficit though, cannot be closed through advertising and/or subscriptions. And so, some kind of miracle is needed. Thus, we are asked to put the calculator away and turn to narrative. In this, the AI companies are similar to the people peddling nuclear fusion – which is always 25-years in the future. Sometime soon, we are told, LLMs will develop the mythical ASI. This is then followed by two versions of the narrative. On the positive side, controlled ASI will cure cancer, solve climate change and launch humanity to the stars (solving several maths problems along the way). On the negative side (and here it would appear the AI’s PR people have been using their own LLMs to produce lame copies of 1980s sci-fi movies) uncontrolled ASI will take control of global communications, launch an army of humanoid robots against the humans, and ultimately wipe us out in a nuclear holocaust.

The latter narrative has become something of a moral panic over the last week following a spurious whistleblowing incident by self-publicist Jacob Coxon, claiming that:

“If we don’t slow down at the current rate of progress, there is a strong chance that we could all die in the immediate future.”

Far from the usual anger about corporate whistleblowing, the AI CEOs took to the airways to endorse Coxon’s claims – something that might be a little more plausible if they hadn’t been caught offering to pay science influencers like Sabine Hossenfelder to trumpet the same scare stories. And, of course, the proposed remedy is not a wholesale shutdown of the AI companies while a government-funded group of independent scientists investigate what the companies have been up to (hopefully with the tech bros themselves hauled up before a grand jury). Rather, they are merely calling for governments to intervene to slow them down.

Which, of course, brings us back to that massive debt overhang. Given that neither AGI or ASI are about to magically appear in the next six months, the AI companies are facing their real greatest fear… that they will be forced to open up their books and the world will get to witness some of the dodgiest accounting practices since Enron. If only there was some way of slowing everything down…

Encouragingly, neither the US congress nor the orange golem himself (apparently even a stopped clock is correct twice a day) seems keen to help the tech bros out. Which at least means that the AI bubble will burst sooner rather than later… after which, there may be little to gain from the wreckage. Because AI is more like Concorde than the Wright Flyer. It is an excessively expensive vanity project which is ultimately unaffordable. And it is mostly a dead end which cannot be saved by a raft of cheap and easy improvements. Unlike the railway boom of the 1840s or the DotCom boom in the 1990s, AI will not leave us with an infrastructure network to build on afterwards. Most likely, when the dust settles, less ambitious LLMs of the Deep Seek kind, running on private servers will offer an affordable way of continuing the handful of genuinely useful things that LLMs can do.

Meanwhile, the one legacy that you can bet the house on is that we will once again fail to learn the importance of using a calculator, and that the next time some Sam Altman/Elizabeth Holmes wannabee turns up selling an improbable miracle tech story, financial and tech journalists will be falling over each other to sell us the fraud.