The Big Bubble of Artificial Intelligence

In Simulacri digitali (Add Editore, 248 pages, 20 euros), Andrea Daniele Signorelli analyzes with lucidity and critical spirit the way in which technologies are redefining the boundaries between real and virtual. With a powerful and ancient tool: narration, which shapes the present and directs our future. The levers of marketing, storytelling and finance are used to create an illusion of the future, functional to transforming reality according to political, economic and speculative logics.

In the book, Signorelli questions the scenarios we are going through and how their interpretation is generated, and at the same time distorted, by the technologies we interact with. This is the case, especially in recent times, of artificial intelligence, to which this short excerpt from the book is dedicated.

In recent years, the world of new technologies has been marked by a long series of moments of euphoria, which have often generated enormous movements of money before deflating and in some cases disappearing without a trace. The end of these cycles has in some cases overwhelmed promising innovations that were still too immature to meet expectations, paradoxically ending up hindering, rather than supporting, their development. Are we sure that the case of artificial intelligence - a technology that has been changing our lives, for better or worse, for a good ten years - is so different?

Some of the most promising startups in the sector are actually facing a lot of difficulties. This is the case, for example, of OpenAI, which is expected to lose over 14 billion dollars in 2016 without having yet identified a sustainable business model. A similar situation, although on a smaller scale, for Stability AI: in the first quarter of 2024, the company that produces the image creation system Stable Diffusion recorded losses of 30 million dollars (against only 5 million in revenue). A situation that led to the resignation of CEO Emad Mostaque and the firing of 10 percent of employees.

If AI is indeed facing the various stages of a bubble, then we could be at the end of the euphoria phase, which in the classic subdivision theorized by economist Hyman P. Minsky precedes the “profit-taking” phase, when the most astute investors realize that growth is now at its end and cash in (the last phase is the panic phase, when everyone sells low and the last arrivals are left holding the baby).

But is AI really a bubble? Can we really compare the most transformative technology of our time – which should even give rise to a new industrial revolution – to the false promises of the metaverse or to an intrinsically speculative innovation like the cryptocurrency-based web3?

On the topic, experts are divided: in a post last March, Jeremy Grantham (co-founder of the GMO investment fund and famous for having predicted both the dot-com bubble and the 2008 real estate bubble) described artificial intelligence as “a bubble within a bubble”. According to Grantham, the advent of ChatGPT has in fact reversed the downward spiral that the markets had undertaken after the financial euphoria of the Covid years, giving rise to a “bubble within a bubble” that “can only begin to deflate”.

A diametrically opposed position is that expressed by Jamie Dimon, CEO of JPMorgan Chase, who, speaking to CNBC, stated: "When we lived through the first dot-com bubble it was because of the excessive hype generated. But artificial intelligence is not hype, artificial intelligence is a real thing".

But there are at least two flaws in this statement. First of all, the dot-coms were real. They were. From the ashes of that financial bubble emerged Amazon, Google, and many other Silicon Valley giants, while the Internet and the Web (which were at the center of that speculative wave) transformed the world before our eyes, helping to make artificial intelligence possible (by providing the data with which these systems are trained).

Despite its enormous and concrete potential, the Internet has nevertheless been the protagonist of an immense speculative bubble, due to excessive expectations, the fear of not getting on the moving train, and a technology that was not yet sufficiently mature.

Does this remind you of anything? Generative AI systems are also experiencing a similar situation. After having introduced ChatGPT and its siblings as omniscient oracles capable of making the entire human species obsolete, today we find ourselves dealing with chatbots that make countless mistakes and image generation tools that often crudely remix the images contained in the dataset they were trained on (unless they are guided by particularly expert professionals or creatives).

As much as they can help us write emails, transform an article into a post for Linkedin, edit some images, organize our day or create a Powerpoint presentation, are we sure that these systems are really capable of revolutionizing the economy?

Another supposed difference between the dot-com bubble and today’s expectations for AI is that the startups of the time—which had names like Pets.com, Priceline, eToys, and all of which failed within months—were on “shaky economic foundations.” Matt Cohen, founder of Ripple Ventures, thinks the situation today is not so different. Speaking to Pitchbook, he explained: “There’s a huge frenzy, and there’s probably a lot of people getting funding that shouldn’t have been given funding. Now, all I ask startups is to show me what makes them different from their competitors, and whether they’re using their own systems or systems that were designed by others. The market is completely saturated right now.”

The recent and terrible flops of the Humane AI Pin and the Rabbit R1, two devices that aimed to exploit artificial intelligence to replace smartphones with an all-purpose assistant, probably represent another signal to take into consideration; as well as the difficulty in generating significant economic returns even for well-known companies in the sector, such as StabilityAI, PerplexityAI and even OpenAI, which already in 2024 has amassed so many losses that it is feared an imminent bankruptcy.

Although some expectations may be exaggerated, there is no doubt that artificial intelligence (not only and not especially generative intelligence) is a technology capable of transforming the world, and that it has already partly transformed it. As the history of dot-coms teaches us, this does not mean that it cannot survive a speculative bubble. Indeed: in some ways, this seems to be a rite of passage even for truly revolutionary technologies.

All this does not mean, however, that artificial intelligence is immune to the perverse dynamics that we have described so far in relation to web3, metaverse, autonomous cars, etc. and that there are no actors who create narratives completely disconnected from reality in order to attract those companies that hope to be able to ride, or legitimately exploit, the promised magnificent potential of these new technologies.

According to data reported by the New York Times, the consulting firm Boston Consulting Group will generate 20% of its revenue from AI-related services in 2024 (up from 0% two years earlier). IBM Consulting has instead secured over a billion dollars in revenue thanks to consulting on AI systems. Accenture reached 300 million dollars in 2023, KPMG International reached 650 million dollars in the first half of 2024 and McKinsey predicts that in 2024 it will also obtain 40% of its revenue from generative AI.

How do consulting firms ensure such high revenues for a technology that – in its latest generative version – has not yet generated such a significant spin-off? To understand this, we probably need to start from the operational strategy of many of these consulting and market analysis firms. A strategy that we could summarize as follows: first of all, the potential of a technology surrounded by particular media hype is studied, then reports are produced in which its transformative potential at a corporate level and its economic impact are magnified, often in an improbable manner. And finally, very expensive consulting services are offered to companies that want to understand how to best exploit the potential that was first narrated by the consultants themselves.

The most recent case, in fact, is that of generative artificial intelligence, which by 2032 will give rise to a market worth 167 billion dollars (Future Market Insight). Or maybe it will reach 266 billion dollars, as Dimension Market Research claims? And what if Fortune Business Insight is right, estimating the value of this market in 2032 at 970 billion? And why not the 1.3 trillion of Bloomberg Intelligence?

Who offers more? The mere fact that the estimates of the various analysis and consulting firms have such macroscopic differences speaks volumes about their reliability. Just as the estimates relating to the added value brought to the global level by a technology that, in its “generative” version, is still taking its first steps, raise many doubts.

The mechanism is always the same: exploit the hype surrounding a technological innovation to create reports that magnify its incredible potential, and then earn money by explaining to companies how to adopt a technology that maybe they need or maybe they don't. That maybe has enormous potential or maybe not. That maybe will change the world and maybe (almost always) not.