At first glance, imagine that the huge expectations placed on ChatGPT and generative AI are disregarded seems madness. On the other hand, it has now been about 10 years since deep learning (i.e. the system that powers all AI) started changing the world before our eyes.
By algorithms that predict what we would like to buy on Amazon or see on Instagramto those that allow cars to move with increasing autonomy, up to the systems used in the medical or scientific fields (e.g do not talk about autonomous weapons): artificial intelligence is now so pervasive that it has been compared to electricity. Why shouldn’t their most recent and impressive evolution deliver what it promises?
Artificial Intelligence Why are actors afraid of AI? A concrete example of Emanuele Capone 22 July 2023
Indeed, the economic forecasts speak for themselves: according to Bloomberg Intelligence, generative artificial intelligence (i.e. able to create text, images, music, video and more) will create a 1.3 trillion dollar market by 2032. According to Goldman Sachs, ChatGPT and its ilk they could instead increase global gross domestic product by 7% (about 7 trillion dollars).
Expectations are so high that they have revived the fear that these algorithms turn out to be so powerful and advanced to represent an “existential risk” for the human species: Is it possible that all this could finally turn out to be a disappointment or even a hole in the water?
Se the recent lesson of the metaverse (That it has been a failure so far) has however taught us something, it is precisely that we should not overly trust the science fiction narratives coming from Silicon Valley or even the pharaonic assessments of consultancy companies, but that we should instead pay attention to concrete data and the considerations of the most cautious experts.
Let’s start with the data. Which show how already in June the number of users of ChatGPT has experienced a significant 10% drop compared to the previous month and as, more generally, also the adoption rate of other generative AI systems have been disappointing. In a nutshell, after the overwhelming initial curiosity, users seem to be gradually losing interest in systems that, perhaps, do not have significant uses for them in the professional field or otherwise.
However, there are some important exceptions: according to one study, 37% of professionals who operate in the field of marketing, advertising and communication have used ChatGPT as an assistant, while it is known that this tool has also proved to be very useful for computer programmers. “To date, most of the turnover (of OpenAI and other companies that produce similar systems, ed) seems to come from two sources: semi-automatic programming and text drafting”, reported computer scientist Gary Marcus.
These uses, Marcus added, however, guarantee very limited revenues: “The evaluations foresee a trillion-dollar market, but at the moment the turnover produced by generative artificial intelligence is estimated at a few hundred million dollars. It’s possible that these numbers will increase exponentially, but we shouldn’t take that for granted.”
All this is even more true if we consider how much it costs to operate these systems: it is estimated that OpenAI spends $700,000 a day managing ChatGPT and that his accounts are at a loss of over 500 million of dollars. Considering the decline in users, it is possible that Sam Altman’s company is having a much harder time than expected to become profitable. Several other companies, which cannot enjoy the colossal funding received by OpenAI, are instead having many difficulties to create their generative artificial intelligence systemsprecisely because of the prohibitive costs.
At the moment, the only industry that could in a short time produce important economic returns is that of search engines powered by ChatGPT and the like (as both Bing and Google are experimenting). The hope is that successfully integrating generative AI into search engines will further boost a sector that is already worth something like 225 billion dollars today.
To succeed in this endeavor, however, it is necessary to overcome obstacles that may prove insurmountable. All LLMs (simplifying, the AIs that produce texts) are afflicted by frequent hallucinations, a term which indicates the episodes in which these systems produce data or information invented from scratch. It is a problem that, at the moment, makes their use impossible in all fields that require accuracy and reliability (including, of course, that of search engines) and which according to many experts it may not be solvable.
The Case Does Generative AI Infringe Copyright? The 2 reasons why Getty sues Emanuele Capone’s Stable Diffusion January 16, 2023
Another problematic aspect is related to the use of copyrighted material. All generative artificial intelligence systems limit themselves to recombining (in a very complex and sometimes surprising way) the contents extracted from the Web present within their database. Content that, however, is often protected by copyright, to the point that the New York Times has decided to block ChatGPT from accessing its articles and is also considering suing OpenAI. A lawsuit that could potentially force OpenAI to reset ChatGPT (starting its training all over again) and to receive fines of 150,000 dollars for each content for which it has illegally used material from the journalism giant.
In short, the road to the success of generative artificial intelligence is much less downhill than expected. Seen them enormous potential, it would certainly be wrong to bury ChatGPT and the like in Silicon Valley’s graveyard of broken promises right now. At the same time, the lessons of the past should have taught us to evaluate the sci-fi (and economically interested) promises of investors and tech-billionaires much more cautiously.