ChatGPT can compose poetry, tell a story and write code. For the first time, artificial intelligence becomes a consumer.

We come from troubled times for those who chase the future and try to anticipate its moves: the pandemic, the demise of social networks as we know them, fake news, and the rise of TikTok have filled the technological pages of a future that seemed to be over, or on its way out. Then the first consumer artificial intelligence arrived among us. Its name is ChatGPT. It was developed by the OpenAI association, founded in 2015 also by Elon Musk (yes, him again) to solve one of Silicon Valley’s many problems: ‘let’s see how far an AI can go if we let it if we allow the public to use it as they wish’.

Microsoft and Google had already tried this in the past, with mixed fortunes. Tay, Redmond’s bot, had ended up posting racist tweets on Twitter, while Lamda, Mountain View’s AI, seems to have grown too fast, so much so that it led to the sacking of an engineer who had said so months ago. 

OpenAI’s latest magic, for now, is called ChatGPT and is what it sounds like from its name: a chat. The (human) user can interact with the machine, asking it for information to compose poetry or tell it the story of the October Revolution. But also to write the code for a working WordPress plug-in, for example, adding the programming profession to the long list of jobs endangered by AI. It is unclear what will happen, nor what the reactions of the public and big companies will be.

Some suspect that Google will choose to penalize web content produced by AIs, something it already announced last April. Will it work? More importantly, will it be enough? Such a tide cannot be predicted, let alone contained by traditional measures. Admittedly, Google is extremely powerful, but it is the same company that has been investing in the AI sector for years and has been overtaken on the right by a lean and rather unscrupulous entity like OpenAI.

Great power, greater responsibility

In addition to producing pseudo-Dantesque verses and credible paragraphs, a service like ChatGPT functions above all as a machine that can give answers to any question. Or almost. It may not always be correct, of course, but the impression is of talking to a magic alien box from the future, and it is the same effect Google must have had when it first came on the market: that strange feeling you get when you are at the beginning of something big. A tomorrow where you can open an app on your phone and ask it if anything is near, and that will make people like Google obsolete and old. Is this ‘tomorrow’ just around the corner? Not really, but it is certain, written in stone, an almost obligatory day for the development of technological innovation. Big tech is warned.

The aspect that allows ChatGpt to stand out is its ability to answer a naturally formulated question using a new variant of Gpt-3, called Gpt-3.5. This modification has unlocked a series of functions that allow any question to be answered, giving Ai’s powerful model a new interface that anyone can use. OpenAi’s decision to make the service available free of charge and that ChatGpt’s shortcomings could prove amusing helped the chatbot go viral.

Christopher Potts, a professor at Stanford University, explains that the method used to help ChatGpt answer questions, which OpenAI had already presented in the past, represents a significant step forward in assisting artificial intelligence handle language in a more comprehensible way. “It’s very impressive,” says Potts, despite the fact that the technique could make his work more complicated. “It got me thinking about what I will have to do in my courses that have assignments with short answers. Despite its great potential, ChatGpt also exhibits the flaws that haunt text generation tools.

Although ostensibly designed to prevent users from making them say unpleasant things or recommend illegal or unpleasant actions, ChatGpt can still reproduce horrible biases. A chatbot that, on the surface, presents itself as an eloquent and well-informed system but generates falsehoods with conviction could make these unresolved problems even more problematic. Users have also shown that its control systems can be circumvented: asking the programme, for instance, to generate a film script about how to conquer the world is a way to circumvent its refusal to answer a direct request.