Are we sure that ChatGpt will revolutionize the world and our lives?
Within a couple of months ChatGpt has already become a global phenomenon, but all the worries about a world in which we will lose our jobs and our children will be rendered ignorant by AI seem, for the moment, decidedly exaggerated. Here because. Peter Kruger's article for Agenda Digitale
Released by OpenAI at the end of last November, ChatGpt has already become a global phenomenon, capturing the dystopian fantasies of many observers. Beyond the ongoing debate between skeptics and enthusiasts, however, few seem to be asking the fundamental question: is there a sustainable business model for generative AI?
The economics of generative AI
Accustomed to the triumphant stories of the digital economy, from Google to Facebook, we often forget that even the success of these large digital platforms was preceded by many failed attempts. Furthermore, the economics of a generative AI platform could be very different from those of more traditional digital applications.
The question is not merely rhetorical, since, in the absence of financial sustainability, it is difficult to see how a service like ChatGpt can have such a disruptive effect on the world we live in, let alone the more dystopian visions promoted by so many digital "gurus". All the more because, in terms of costs, few technologies suck up computational resources as much as AI.
For example, the first estimates suggest that the cost of ChatGpt is really very high. A single response from the platform would cost, on average, in the order of 1 euro cent, which would be between 1 and 2 orders of magnitude higher than the cost of today's most established digital platforms. And we're just talking about operating costs (OPEX). Then there would be investments (CAPEX) which, for AI, are among the highest in the entire digital industry.
Yes, of course, the infrastructural costs of digital services tend to decrease over time (even if the glories of Moore's Law are not destined to last forever), but the need to provide a more performing service could further cause costs to explode both level of OPEX and CAPEX.
The current ChatGpt service, for example, is based on the training of online content created before 2022 and this limits its topicality and relevance (for example, ask ChatGpt to comment on the war in Ukraine…). However, a service that needs to carry out continuous retraining operations in order to update itself could have a cost and also pose technological challenges that are anything but trivial.
What services at what price
And, so far, we have limited ourselves to talking about costs. Financial viability presupposes the existence of services capable of justifying, at the same time, an adequate price and sufficient volumes of customers willing to pay such pricing (two things that tend not to get along very well with each other) .
The fact that the operating costs of ChatGpt are 1 or 2 orders of magnitude greater than the costs of the major digital platforms suggests that the revenues generated for each individual interaction must be 10 to 100 times higher. All this at comparable volumes of use. Much has been said about the possibility of ChatGpt to replace Google Search. But with these economics, you don't see how this can be achieved with the classic advertiser-funded ad models.
Is it possible to imagine a model based on subscription payments? Social networks these days are full of professionals who declare their willingness to pay any amount to be able to replace their networks of collaborators with the ChatGpt service. However, the rhetoric of "any price" ends as soon as that price exceeds the cost of the resources it wants to replace. What is the market size of professionals or companies that could benefit from using ChatGpt? And is there a price acceptable to this market, capable of covering the huge investments and operating costs?
The simple answer: we don't know. And, even if there were, let's remember that it took about 10 years from the introduction of the first search engines in the early 90s to the arrival of a model that was able to adapt optimally to market demand (in startup parlance, the so-called product-to-market fit). Today development cycles have accelerated dramatically compared to 20-30 years ago, but finding a monetization model for generative AI could prove to be a much more complicated challenge.
No sustainable business model on the horizon
What we can say with almost certainty now is that ChatGpt has already practically brought OpenAI to the gas barrel in just over a month of operations. Microsoft's decision, just a few days ago, to invest 10 billion dollars in OpenAi, valuing the latter at 29 billion dollars, made headlines. Less has been said that the founders have seen their control shrink to about 2% of the company. Sign of an operation made on the verge of desperation, of the "take it or leave it" type. It also proves that, at the moment, there is nowhere near a sustainable business model on the horizon.
This is not to say at all that Microsoft has made a bad deal. This is a high-risk investment, but the potential return if successful is incalculable. In this sense, one could say that Microsoft has just made the most expensive seed investment in history ($10 billion) for a company that not only hasn't achieved product-to-market fit yet, but where even the product from only still has a long, really long way to go. And, although the inclusion of ChatGpt functions within Microsoft products such as Office is already foreseen, it is really difficult to see how this could in itself represent a sustainable business model.
Moreover, rumors from the industry suggest that LaMDA , the equivalent generative AI product developed internally by Google, is already more performing than ChatGpt. The same rumors argue that the reason why Google has not yet released this application is related to the reputational risk of adding a still unreliable and unreliable function to the search product, now recognized as an industry standard. I suspect, however, that the operational costs of serving a feature like LaMDA on Google's search volume scale (we're talking several trillion searches annually) are simply unmanageable. Just as it is probable that the recently announced "ChatGPT Professional", a first paid monetization model released by OpenAi in the same days of the Microsoft investment, could have more the function of introducing a cost barrier to contain costs by reducing consumption .
Conclusions
In conclusion, there is no doubt that generative AI represents an epochal innovation. And there is no doubt that AI, a discipline truly capable of generating value only in the last 15 years, is still in its infancy, holding great innovations in store for us in the future. But only if the accounts come back.
There is still a lot of development to be done on the product, a lot on market development. Meanwhile, all the worries about a world in which we will lose our jobs and our children will be rendered ignorant by AI seem, for the time being, greatly exaggerated.
This is a machine translation from Italian language of a post published on Start Magazine at the URL https://www.startmag.it/innovazione/siamo-sicuri-che-chatgpt-rivoluzionera-il-mondo-e-le-nostre-vite/ on Sun, 05 Feb 2023 06:23:43 +0000.