Last March, two of the most discussed startups in the generative artificial intelligence sector, Inflection AI and Stability AI, suffered heavy defections. In the first case, Inflection AI CEO Mustafa Suleyman left the company to lead the division that brings together all of Microsoft’s artificial intelligence projects. The news caused a sensation for at least two reasons: Suleyman is one of the best-known entrepreneurs in the sector, he was the co-founder of DeepMind, a British AI company acquired by Google in 2014, and in 2022 he founded Inflection AI, which last June had received $1.3 billion in investments, reaching a valuation of $4 billion. Suleyman himself seemed to have great ambitions for the company: in October he declared that he wanted to “create a 100 billion dollar business”. Last month, however, he admitted with Bloomberg that the company has failed to find a sustainable business model.
A few days later, Emad Mostaque, CEO of Stability AI, also left his position, the latest in a long series of prestigious resignations that have affected the startup, which since 2022 has been among the best known in the sector. Stability AI, in fact, is the developer of Stable Diffusion, a linguistic model capable of generating images starting from textual descriptions (called prompt) and among the services that have been at the center of the expansion of AI in recent years, together with ChatGPT and DALL-E.
These resignations occurred in a period in which the AI sector seems to be going through a transition phase: after about two years of continuous growth and enormous investments, in recent weeks signals of a different type have arrived, not exactly of a crisis but of a general rethinking. In February the Wall Street Journal told how some “early adopters” – companies that were the first to invest in generative artificial intelligence – are struggling to find applications useful enough to justify the expense.
– Read also: We need to understand what “artificial intelligence” is
In recent months, Microsoft has heavily promoted Copilot, a digital assistant that helps workers, at a price of $30 per month per user, to use programs from the Microsoft 365 suite, such as Word, Excel and PowerPoint. The Wall Street Journal underlined the success of some Copilot applications (of all the most appreciated is the one that allows you to transcribe and summarize what is said in video calls and emails), but also recorded some criticisms which mainly concern the poor reliability of AI in generating slides for presentations or other types of documents.
These problems are due to so-called “hallucinations”, factual errors in which the system produces results that have no connection with reality. At the basis of what we call generative artificial intelligence, in fact, there are large linguistic models, i.e. very sophisticated algorithms that carry out probabilistic calculations to generate theoutput considered more correct, the result that appears to the user. These calculations do not always give the right result, however: within an advertising agency, for example, it happened that the summary of a video meeting prepared by Copilot reported that this Bob had spoken about “product strategy”, even if It hadn’t been discussed at the meeting, and there was no one called Bob among the attendees.
The accumulation of these testimonies fuels the widespread suspicion among some analysts that AI are notable and potentially useful tools but that companies in the sector are exaggerating their capabilities today, especially to obtain investments. This is also supported by Gary Gensler, president of the Securities and Exchange Commission (SEC), the US federal body responsible for supervising stock exchanges, who coined the concept of “AI washing” to indicate the strategy used by companies that appoint intelligence artificial in their reports, often without any concrete basis.
According to a study by Goldman Sachs, 36% of companies in the S&P 500 index, the US stock market index that brings together the 500 largest capitalization US companies, mentioned AI in the fourth quarter 2023 report. Scott Kessler of research firm Third Bridge Group explained to the site BusinessInsider that, despite the “grandeur” of the promises made in this field, “we don’t know if or when some of these things will be possible”. According to Kessler, in fact, changes similar to those described by entrepreneurs such as Sam Atman of OpenAI, who has been talking about AI for years with a mixture of amazement and trepidation, “don’t happen overnight”.
– Read also: For now it is difficult to make money with artificial intelligence
Another signal comes from OpenAI, which in January presented GPT Store, an online store where you can download specific and customized versions of ChatGPT, the well-known chatbot produced by the company. At the time, the novelty inspired comparisons with Apple’s App Store, which helped popularize the iPhone and allowed the birth of many hugely successful services. A few months later, however, the launch of GPT Store was judged slow by the site The Informationalso because in recent months the company has been involved in many, perhaps too many, big projects: it is currently working on a search engine and has just developed Sora, an AI capable of generating videos from textual prompts, while the its co-founder Sam Altman wants to invest in the production of next-generation chips to make the development of these technologies less energy-intensive (and expensive).
The economic issue risks becoming the most pressing for the entire sector. The same cases of Inflection AI and Stability AI can be reduced to this: developing generative AI is very expensive and the services related to them are not yet profitable; for this reason, companies such as Microsoft (which has an alliance with OpenAI and has invested in many other startups), Google, Meta and Amazon, which have enormous capital to invest and can count on a vast ecosystem of services on which lean. In recent months, the centralization of power in a few companies has therefore consolidated, which has created a situation of enormous disadvantage for the smaller and ascending ones, as denounced by Gensler himself.
Despite this, even the largest companies are struggling to monetize AI, even those most appreciated by users, such as the one that summarizes video calls and emails received. Using technology like GPT-4 to summarize an email message is, moreover, an inefficient use of a very powerful service: “like taking a Lamborghini to deliver a pizza”, according to the Wall Street Journal.
Among the sectors in which the use of these tools has become more widespread is that of computer programming, where products such as GitHub Copilot, a virtual assistant capable of writing and reviewing code, have had great success despite remaining at a loss. According to a revelation from Financial Timeseven Google is thinking of charging for a search service powered by artificial intelligence, a choice that would be a complete break with Google’s traditional free approach.
The expansion of the sector occurred as technology companies abandoned the cautions and shared norms that they had followed for some years in the development of artificial intelligence. In an interview with Mira Murati, chief technology officer of OpenAI, also given at Wall Street Journal on the occasion of Sora’s presentation, journalist Joanna Stern addressed the issue of copyright and the use of protected materials to “train” and develop these AIs. Murati gave a very confusing response, specifying that the material used was “available to the public”. As for YouTube, the company responded by reminding us that the use of its videos for a product of this type would violate its terms of use. Last week the New York Times revealed that OpenAI already in 2021 used a program to transcribe videos on “one million hours of footage on YouTube”.
OpenAI wouldn’t be the only one. According to New York Times, Meta and Google have also discussed collecting copyrighted data from across the internet, “even at the risk of legal action.” Last year, some Meta managers went so far as to consider purchasing the American publishing house Simon & Schuster to “procure long works” on which to develop artificial intelligence. As for Google, last year it revised its terms of use to allow the company to “use documents available on Google Docs, restaurant reviews on Google Maps and other online materials for its AI products”.
In addition to disadvantaging startups, however, this approach also represents a possible legal risk that still needs to be addressed: there are currently several open lawsuits between authors, artists, publishers and newspapers (such as the New York Times, which denounced OpenAI) and companies in the sector. It is not easy to predict how they will end and what consequences they will have: according to Nilay Patel, director of the site The Verge, «These cases are like flipping a coin. (…) If the legal system itself is not predictable, copyright is intrinsically unpredictable.”
– Read also: For artificial intelligences to work, human intelligences need to do very boring jobs
Despite some risks and limitations still found in the technology, the AI sector continues to obtain interests and investments, immersed in “a feeling of inevitability”, as Rana Foroohar, journalist of the Financial Times: «Even if you think that AI is today’s equivalent of electricity or the internet, we are still in the early stages of a complex transformation that will take several decades to complete and which is in no way assured today».
Yet the enormous interest linked to the sector has made the fortune of companies such as Nvidia, a US company producing microprocessors, which last February surpassed Alphabet (the holding company that controls Google) in terms of market capitalisation. The video cards produced by Nvidia are the most used in the development and operation of artificial intelligence systems, and have made the company fundamental for the entire technological sector. However, such a large and sudden growth has also raised suspicions that it could turn out to be a speculative bubble, similar to those linked to the metaverse or cryptocurrencies. According to others, however, it is necessary to find a new term to indicate a type of investment which, although not a bubble, represents a much higher risk than usual, as in the case of artificial intelligence.
