[Translated from Italian to English by Google Translate]

History and future of ChatGPT and its sisters in the book just published by the Italian scientist who became famous at the Massachusetts Institute of Technology in Boston. «The genie is out of the bottle, but humans are more dangerous than robots»

by PIERO BIANUCCI

The authors know this. Some books ask to be written. They demand it. Even with a certain arrogance. It happens to those who write novels or poems, and then it is called inspiration. But it also happens to those who write essays: philosophy, history, science. This is the case of Tomaso Poggio, one of the most important experts in Artificial Intelligence, author, with Marco Magrini, of “Brains, minds, algorithms” (263 pages, 18.90 euros) just published by Sperling & Kupfer.

Long incubation
Six years ago Tomaso Poggio had the book ready: title, index of topics, a few sketched pages. But other works took precedence and the book ended up in a drawer. «By chance – Tomaso Poggio now writes in the Prologue – the decision to come out of the drawer was taken two months before artificial intelligence ended up on the pages of all the newspapers and in people's mouths...».

Media explosion
On June 11, 2020, still quietly, ChatGPT3 was made available to the public. He answered any question, translated from and into dozens of languages, invented film plots, helped (or replaced) doctors, lawyers, accountants, screenwriters, novelists, poets. It was “generative” Intelligence. With other similar algorithms, more backward but still developing, a competition was unleashed involving billions of investments. Of course, Elon Musk occupied center stage. Hollywood strike, everywhere debates on the ethics of Artificial Intelligence, alarm over the possible overtaking of natural human intelligence. Despite union conflicts, ethical doubts, confused fears and Elon Musk's anathemas, ChatGPT has made enormous progress in just a few months, and so have its "sisters". The most recent version was released on January 10, 2024. Among other things, it draws, paints and manipulates photos: a treat for the makers of fake news. Open AI, the parent company with a very detailed price list for its use.

Power and limits
And here we are. Poggio's book, conceived and written before the media explosion, arrives at just the right time, and brings a highly competent voice while many intellectuals improvise approximate opinions on the topic that he has studied all his life and of which he is universally considered a pioneer. The reader learns the evolutionary history of natural intelligence (which is not exclusive to Homo sapiens) and the vicissitudes of Artificial Intelligence, discovering its economic implications, power and limits. In summary: the Genie is out of the bottle and will never go back in, but “humans are more dangerous than robots”.

Caution with the "singularity"
Poggio does not believe in the danger of the catastrophic "singularity", the machine that subjugates man. But he doesn't rule it out a priori. Experience has taught him to be cautious. «After fifty years of life spent trying to understand the genius of natural neurons and artificial neurons, I look with pride, and with subtle amazement, at everything we have understood so far. And I can't wait to be even more amazed by what we will understand tomorrow."

What he said in 2007
In 2007 Tomaso Poggio came to Italy, to Sarzana, and inaugurated the fourth Festival of the Mind. It was August 31st. He was sixty years old, today he is 77. I interviewed him. Speaking with Poggio means leaving the ephemeral and walking on solid ground. The things he said 17 years ago are still interesting, indeed in some ways they are even more so than then because they help to understand where today's Artificial Intelligence comes from. Anyone who wants to have an unusual experience can. So, read this article-interview from a long time ago.

Retrospective reading
«Said in the words of Marvin Minsky, Artificial Intelligence is a set of disciplines (logic, mathematics, computer science, electronics, mechanics, biology, cognitive psychology, neuroscience, philosophy) which has the aim of "making machines do things that would require intelligence if they were made by man." Together with John McCarthy, creator of the expression "Artificial Intelligence", Minsky founded the Artificial Intelligence Laboratory at MIT (Boston) in 1958, now directed by Tomaso A. Poggio. Genoese, sixty years old, director of the Artificial Intelligence Laboratory of the Massachusetts Institute of Technology (MIT) in Boston for 26 years, Tomaso Poggio has a degree in physics but has always been fascinated by the mechanisms of our brain: one hundred billion neurons, each equipped of approximately 10 thousand connections. The computer, however powerful, is still very far from the capabilities of these 1400 grams of cells. But it offers a simplified model to those who want to explore how it works. Poggio chose this path and concentrated on the mechanisms of vision, first at the Max Planck Institute in Tübingen in Germany and then in the United States. Teaching a machine how to recognize objects and classify them is the aspect of Artificial Intelligence to which he has dedicated much of his research. In his work - 400 publications - the boundary between biology and computer science, cognitive psychology and Artificial Intelligence is so thin as to become elusive.

Neuroscience and AI
Professor Poggio, today studies on Artificial Intelligence serve us to illuminate problems of neuroscience or is it rather neuroscience that helps us achieve the objectives of Artificial Intelligence?
«I think we have reached a point where neuroscience is starting to take the lead and give ideas to those working in the field of Artificial Intelligence. I am particularly interested in this aspect. That neuroscience today can inspire those who work on Artificial Intelligence is a belief that also comes to me from the studies I am developing with my group on the functioning of vision. When I started doing research 35 years ago, in a race between Artificial Intelligence and neuroscience I would have bet on the former. Today I would do the opposite."

Exaggerated expectations
Compared to the expectations that arose after the historic Dartmouth Conference in 1956 during which John McCarthy coined the expression Artificial Intelligence, many enthusiasms have diminished: why was the challenge more difficult than McCarthy and Marvin Minsky might have thought, the another pioneer in this research?
«The initial expectations towards Artificial Intelligence were undoubtedly exaggerated. The problem of intelligence, of the mind, is the most difficult and profound in science today. It's a problem that humanity has been struggling with for a couple of millennia. Thinking of solving it in a few years was arrogant in 1956, and it still is today. Those hopes, however, are understandable when a new field of research opens up. Think about nuclear power: in the 1950s many were convinced that even cars would run on nuclear energy. There were also incorrect ratings. Back then it was thought that a computer with a terabyte of memory would be more intelligent than a person. Now there are terabytes but the human brain, or even that of an animal, is a completely different thing."

Autonomous systems
With the most advanced robotics, strong Artificial Intelligence seems to be back in vogue, according to the original spirit: I'm thinking of robots that play football, those designed for military actions or to assist the elderly...
«These things go in cycles. Lately, the prevailing idea was that we would not be able to create completely autonomous systems, capable of learning and making decisions. The idea was that even the most advanced systems would always need human assistance. Now, driven by robotics, we are once again believing in autonomous systems. There is and will be progress in this sense. But there has been no fundamental discovery that could radically change things. However, we shouldn't be too pessimistic either. We have "expert systems" that work very well: for example for medical diagnoses or for voice recognition."

Is there an insurmountable limit?
Won't there be an intrinsic insurmountable limit for Artificial Intelligence, a bit like Goedel's incompleteness theorem for mathematics?
«Maybe yes, and probably no. There is no other way to find out than to try
to create artificial intelligence. Arguments of this kind are philosophically interesting but matter little. They remind me of the scientists who at the beginning of the 1900s "proved" that heavier-than-air planes could never fly."
A problem that seems simple but has actually turned out to be complex is that of face recognition: what stage is the research at?
«The recognition of specific faces under controlled conditions (an image, lighting, frontal position) today is probably done better by computers than by a person! If conditions are not controlled it is a different matter, but our ability is overestimated. The fact that there is a photo of the face in the passport is a bit of a coincidence: a face does not contain many actual bits about the person's identity. However, today there are commercial passport control products that work better than the border police. For years, automatic systems have been used to read checks or check printed circuit boards and computer screens and discard the bad ones. These things have publicly traded companies worth hundreds of millions of dollars behind them. A more difficult problem is classifying objects in a scene: people, faces, cars, phones, and so on even if those particular objects (for example, that type of car) have never been seen before."

Doubts about the quantum brain
Do you believe in the hypothesis put forward by physicist Roger Penrose (later winner of the Nobel Prize for Physics in 2020, editor's note) that the human brain operates on a quantum basis?
"No. It is very unlikely that the brain, with cells operating at room temperature, can maintain coherent quantum states. Quantum coherence requires temperatures close to absolute zero. I believe that physics and biology are enough to explain the properties of the brain."

Will it be possible to create a quantum computer?
«Some already exist but with minimal capabilities and we don't see a good way to improve them substantially. Maybe, but for now I'm skeptical. I am more optimistic, however, about quantum systems for encrypting messages."
Can we talk about creativity when talking about a computer?
"Why not? A sequence of probabilistic choices is indistinguishable from a product of human creativity."

However, the machine is not aware of its creative act
«The question of awareness is very complicated. Nobody knows what this awareness means. It could be a program that controls other programs."

I'll turn the question back to you: the Deep Blue program beat the world champion Kasparov at chess. But he doesn't know he beat him. Kasparov, on the other hand, knows he has lost
«I believe that it would not be impossible to add a program to the software that gives rise to signs of satisfaction in case of victory and dejection in case of defeat. Of course it is one thing to show satisfaction, it is another to feel it. But an external observer may not distinguish the authenticity or otherwise of the feelings shown by the machine."

Types of creativity
Speaking of human creativity, do you see it more as a process of trial and error or as an enlightenment, a divergent thought that allows you to reverse your point of view, as Einstein did when he imagined chasing a ray of light?

«I believe that creativity is above all an ability to make associations. Einstein's example is interesting. His reasoning appears unconventional, but in reality he associated notions that were in the air (the contraction due to the speed of the Lorenz equations) and the experience of the trains of the time, which had on board clocks comparable to clocks on the ground . Furthermore, Einstein was outside the academy and could allow himself to think more freely."

Creativity therefore also requires a low level of inhibition, which is certainly not favored by the demands of a university career.

"Yes. Think of Craig Venter, biologist and computer scientist, who patented six thousand new bacteria discovered while sailing his boat halfway around the world. An academic would never do something like that."

Does the computer understand jokes?
«Doesn't it seem to you that creativity has mechanisms similar to those of humour, in the sense that humor also requires the ability to make associations in an unconventional way? If this were the case, it becomes more difficult to think of a creative machine precisely because it is unlikely that a machine can understand a joke... I agree with this. It's interesting. In fact, we are far from a computer capable of laughing at a joke."

What is your main research interest today?
«Understanding how the brain creates intelligence, and in particular
visual intelligence, our ability to perceive the environment and recognize objects. Clearly visual intelligence is not all intelligence, but it is an important part of it. I see it as a way to access intelligence using animal models, which would not be possible, for example, by studying linguistic intelligence. It is no coincidence that when you understand something in English you say “I see”.