ChatGPT, the artificial intelligence that writes and speaks like us (but shouldn’t scare us) – Corriere.it

hor seen things that you human beings could not even imagine …: this is the future that awaits us with the growing power of big tech and their object of study and sale, artificial intelligence (AI), which one day maybe will become sentient and govern a world with humans as extras?

The famous phrase pronounced by a Rutger Hauer in the part of a replicant in the film Blade Runner by Ridley Scott seems to be back in fashion these days where the new generation of machines designed and educated with the machine learningit shows us that it can do things beyond our imagination with the data made available by the industrial revolution we are experiencing.

A test? GPT-3 (Generative Pre-trained Transformer) la deep networks more advanced (3 stands for third generation) for natural language processing. composed of almost 200 billion parameters that are fixed during the learning done on all web tests. For those who do not believe in the scope of the product, you can either look at the example on the page where an input is given to the machine to produce an output autonomously or you can take a few minutes and do a similar exercise on the site https://openai.com/blog/chatgpt/.

a simple example of a reference letter, but we could also imagine making him write an article like the one you are reading (so we guarantee: we wrote it in our own hand!) or some school exercise or assignment academic (which is why the algorithm has already been banned since December due to the risk of plagiarism by students in schools in the New York district). GPT-3 result of the Open-AI research laboratory, which saw among the founders Elon Musk and other American tech operators including Amazon Web Service. Established as an institute not for profit in 2015, it shortly after changed into for profit (200 million dollars the estimated turnover in 2022) and was valued at a value of 29 billion dollars despite the recent crisis into which big tech has fallen.

The level jump

Impressed? Well, how not to be. Let’s say that, using the language of videogames, we have definitely changed the level, or rather the impact, compared to what we had intended up to now. To date, in fact, we had arrived at a Deep Blue, the IBM computer that beats the world champion at chess. Impressive operation and yet still understandable since, despite the level of complication, even the game of chess has a finite number of plays. We used to have a Siri and an Alexa executing our commands.

But here the result is different because it is not a question of programming a machine on game routines imagining its moves or answer routines with respect to simple questions, what time is it or looking for the way on Google Maps. It is about interpreting and manipulating a text. Something far more sophisticated. More human, one might say.

All of this is happening because machines have learned to learn: precisely machine learning. Thanks to the infinite ocean of data on the web, we no longer have to instruct the machine with commands to explain what a table is (an object made up of a top, four legs, a certain material, a certain color, etc.) or what a cat (a animal with four legs, furry, etc.), as we did with the first AI models.

We can teach her what to do by submitting her to millions of frames on the web, exactly like a human being learns as a child: by dint of seeing tables and cats she then knows how to recognize them over time. Not only is the result far more precise (in reality tables can only have three legs and cats for allergy sufferers are now hairless), but this process endows the machine with a knowledge routine. Then by expanding the processing capacity and the parameters it uses (GPT-3 as we wrote has in the order of billions) we can achieve the production of the reference letter you have read.

And it is no coincidence that Microsoft, one of the recent backers of OpenAI, is considering including the aforementioned chatbot service among the Word and Power Point services.

There are several criticisms that can be shared and for which it is essential to activate careful monitoring of the commercial developments of AI. For example, the risks of mass dissemination of fake news. The risks associated with an involution of our linguistic skills. Everything that needs to be discussed, monitored and regulated so that the diffusion of this precious technology does not lose its role of being a tool and not an end in itself. But the main criticism, the danger that AI machines will have the upper hand over men, seems to us really unfounded. First of all, the machine does not learn by itself, but behind its learning there is precisely the human being who drives the machine.

Criticisms and limits

Even in the case of GPT-3 and the resulting chatbots, it was the support assistants who gave indications on what was right or wrong in the operation of the machine, for example saying that a table, even if it has three legs, or commenting positively or negatively the letter that was written.

Then there are the users who improve the performance of the machine with their information, exactly as we do more or less voluntarily when we rate a service we receive from an app.

But there’s more. The way in which we acquire, process, store and manipulate information as human beings is in many ways still unknown today and the subject of study of new disciplines such as cognitive sciences and neurosciences, which will hopefully proceed with small advances in the coming years to help us to understand who we are and how we think. Basically, as we already know about the physical world we know just 5%; 95% is totally unknown and what is easy for AI is not easy for humanity but also vice versa. And then there is the human differential element linked to emotion, empathy, feelings, which are fundamental in the production of knowledge and which no machine is currently able to replicate.

In summary, both for the inability to read the context like a human being, and for the inability to express those emotions that will always distinguish us from technology and more intelligent, no machine will ever be able to see things that we humans are not able to. able to see.

Subscribe to the newsletter of The Economy

Whatever it Takes by Federico Fubini

The challenges for the economy and markets in an unstable world

Europe Matters by James Fontanella-Khan

Italy and Europe seen from America

And don’t forget the newsletters
The Economy Opinions and the Economy 6 pm