The Dream of the Internet — Human-Centered AI

Originally published in Italian:

https://iltafano.it/articoli/sogno-di-internet-e-ai.html

We grew up believing that the internet was a wonderful place, a free place, where anyone could freely express their ideas and their passions. We convinced ourselves that the internet would unite humanity, would make it one great family. We believed the internet would tear down the borders between nations and erase the differences between peoples; that it would help people in hard places build themselves a future where even hope was running out. The dream of the internet is fading before our powerless eyes.

At the metro stop I see eyes lost in the grey glow of screens. I see fragility and torment growing, fed by the standards imposed by social media. I see hatred and frustration spat at one's neighbor without mercy. I see politics building trends to feed our fears. I see the powerful controlling dissent through digital oblivion, treating our identities, our entire lives, as numbers in a complicated algorithm. I see statistics, analytics, security and surveillance in every activity of our lives. I see phones becoming part of our limbs. I see intelligent computers becoming friends, therapists, lovers. I see humanity becoming a product to be sold to the highest bidder... on that internet we had welcomed with such jubilation.

To all of this we are opposed, because it is not the dream we grew up with. The internet was meant to connect us; instead it has disconnected us — from others, and from reality. We live shut inside our feeds, having lost the feel of everything that lies outside.

Perhaps we will not change things; perhaps these will remain words lost in the recesses of the web; perhaps history will forget us. But we have not lost faith, and we never will — continuing to dream of an internet that is free, safe, private, where each of us is the sole owner of our own identity and our own data.

A clarification: artificial intelligence is not a new form of technology, not a new discovery. It is the natural evolution of old inventions. The birth of a "general record of everything" (the internet) made it possible to acquire information without limit. This general knowledge of humanity, however, escapes human control, because the human mind is limited, able to absorb only a small fraction of the language with which reality can be described. Electronics, then, created machines so powerful and so capable that they could handle quantities of information unimaginable to the human mind.

How to organize all this information? Statistics answered that question. If you repeat to a child the phrase "green meadow," "green meadow," "green meadow" ten times and then ask "meadow?", the child will probably answer "green." But they might also answer "mommy," because ten sentences are not enough to teach a child the color of a meadow. This is how LLMs (large language models) work: they search sentences for recurring patterns and structures, they calculate the probability that after the word "meadow" comes "green," and so, when prompted with the question "meadow?", they answer "green" with a fair degree of reliability.

How can a machine "learn" the relationship between "meadow" and "green"? In a sense, by learning from human beings. Neuroscience has discovered that our brain organizes itself into recurring patterns, into pathways of neurons activated by certain stimuli: neural networks. These neural patterns evolve and change through a biological mechanism called synaptic plasticity: a dense network in which neurons (or nodes, in computational models) adjust the strength of their connections according to experience and the data they receive. The difference is that a machine, to find the right connections, must scan billions of sentences, consuming enormous amounts of energy — whereas a child's brain needs a single glance at reality to understand. Artificial stupidity!

The point is that AI can be broken into pieces and spread across the world: it does not reside in the skull of a single human being, but in enormous data centers working constantly in unison to compute and deliver accurate answers to the questions they are asked, as if hundreds of people were reasoning together to answer a single question. AI has a single purpose, and it spends its days keeping the cooling systems blowing, devouring billions of pieces of information. But while a child takes three years just to learn to speak, a language model needs only a few months to absorb all available knowledge — because it is optimized to do exactly that, and it does nothing else, all the time.

AI can learn everything and answer everyone. Humans cannot.

Artificial intelligence is redrawing the boundaries of the imaginable. Human thinking used to be bound to the question: am I capable of building this? Today the paradigm is reversed, and the question to ask is: what could I build? Imagining is all we have left to do — but imagination truly is all we have left. The intellectual professions are beginning to fade and lose the central role they once held; repetitive jobs inspire fear, because an AI might come along to replace them; even software engineers themselves could be supplanted by the AI they built. Does this mean machines will replace human beings on Earth? I don't believe so. But we need to understand clearly: what makes us human?

I don't have the answer to this question. I have an answer — my own. What makes us different from machines, and from every other living being, is the vision we hold of reality. It is the ability to look at, read, understand and communicate the strange system of spacetime we are immersed in. We can feel time passing, experience emotions, sense temperature, fear, happiness and everything else. We are the only beings on Earth capable of taking the measure of reality, encoding it into language and passing it on to other human beings. This way of living has guided the evolution of humanity up to now.

But things seem to be changing. If we begin to lose contact with the real — grounding our knowledge in information previously acquired by others and redistributed by AI — how will the universal knowledge of humanity keep growing? How will new AIs grow? And above all: how will we manage to remain human?

Artificial intelligence is a tool whose usefulness deserves to be appreciated. It optimizes our work, makes possible what was once unthinkable, and helps us make knowledge more accessible — but it creates no new knowledge, because that task, at least today, still belongs to us. This essay, for example, was translated into English with the help of an AI — a task that would otherwise have taken time, and perhaps the intercession of a translator. It translated it, certainly. But could a truly intelligent machine have grasped the meaning that these lines attempt to investigate?

The vision we pursue is that of a humanity that lives reality rather than reading it on a chatbot. Human beings have the one gift AI can never have: curiosity. To investigate our glimpse of reality — however distorted, blurred and incomplete — is our mission.

The pursuit of the Vision is, therefore, our purpose.

Giorgio Di Carlo — Founder of 11STUDIO