The 50,000 words you need to know to limit the risks of Artificial Intelligence
‘All the uses of words to everyone… not because everyone should be an artist but because no one should be a slave.’ This quote by Gianni Rodari, taken from ‘The Grammar of Fantasy’ (Einaudi, 1973), has become increasingly pertinent as the debate on AI continues to focus on its risks and the possibility that ChatGPT and other generative systems will independently create a distorted reality beyond human understanding. These systems could create their own vocabulary and language, in short, their own abstract, and therefore false and manipulative, version of the idea that nomen est consequentia rei (the name is the consequence of the thing) dramatically altering and overturning its values and meaning.
All the uses of words to everyone, then. Let’s start from that. From educating AI to the necessary skills and limitations to govern and steer its evolution, without fear. As early as 1966, Umberto Eco warned us against fictitiously dividing ourselves into ‘apocalyptic’ and ‘integrated’ groups in response to the unexpected evolution of television. He advised us to analyse, understand and incorporate any overwhelming technological innovation into the order of human knowledge. From Gutenberg’s moveable type to quantum physics, and from the internet to AI. Let’s be honest, however, that this is becoming increasingly difficult.
Which words are we talking about using that provide value? Let’s do some calculations. Italian is one of the richest and most articulate languages, with between 215,000 and 280,000 basic words (including technical, literary and regional terms), amounting to two million forms with all their variations. A good dictionary, such as Treccani or Zanichelli, includes between 100,000 and 160,000 words. The average educated person’s vocabulary is around 50,000 words. De Mauro speaks of 7,500 ‘basic words’, of which only 2,000 are used daily. This is a paltry number, providing a simplified and therefore misleading understanding.
These figures highlight the first requirement: the ability to understand, correctly use and distinguish at least several tens of thousands of words. This is not only to avoid falling into the ever-growing category of ‘functionally illiterate’ people (those unable to understand a medium-complexity text or perform elementary calculations, accounting for a third of the Italian population), but also to enable the formulation of precise and relevant questions about the functioning of AI (which advances through increasingly precise and probing questions).
In other words, we could update Rodari’s call for ‘democratic education’ (i.e. a broad and widespread education as referred to in the Constitution with regard to the responsibility for quality school education) when we consolidate ‘all the uses of words to everyone’ into a vocabulary of around 50,000 words. This is much more than the 2,000 words we use every day, such a low number that it makes us ‘slaves’ to manipulation, propaganda, instrumental simplifications, factoids and fake news. This makes responsible participation in ‘well-informed’ and therefore ‘critical’ public discourse impossible, as Jürgen Habermas demonstrated in Europe and Antonio Gramsci, Don Lorenzo Milani and Benedetto Croce demonstrated in Italy.
The question now is whether generative AI will ever be capable of such creativity that it can coin verses such as ‘I am illuminated by immensity’ or ‘to spend the afternoon, absorbed and pale’. To tell us that ‘we are like leaves in autumn on branches’ or remind us that ‘there’s rosemary, that’s for remembrance…and there is pansies, that’s for thoughts’. Could it fully set out the terms of the ‘wager on God’, describe being ‘in the grip of abstract furies’, delve into the sphere of doubt about the scientist’s responsibility in the face of the atomic weapon, quote ‘the winter of our discontent’ and inform us that ‘even the bramble had its bends of sweetness, even the plum tree its candour’, without drawing on its memory of Giuseppe Ungaretti, Eugenio Montale, William Shakespeare, Elio Vittorini, Lucio Piccolo, Blaise Pascal and the fathers of physics, Niels Bohr and Werner Karl Heisenberg. And quote ‘the Wine-Dark Sea’ to describe the murky darkness of its depths.
AI can recite poetry and combine different and similar verses, and it may even invent rhymes that sound like those from non-existent but plausible poems. However, it is not yet capable of generating original thoughts full of poetic and literary meaning.
Will it be able to do so in the future? The danger is there. Perhaps this is not even the greatest danger. We all know how important poetry and literature are in shaping the human soul, defining behaviour and inspiring great cultural and civic choices that have a radical impact on mass movements, the tastes of millions of people and their passions and aspirations (read or reread Elias Canetti‘s ‘The Holy Fire’ in ‘Crowds and Power’, as well as Alessandro Manzoni‘s ‘The Betrothed’ and Sigmund Freud‘s ‘Group Psychology and the Analysis of the Ego’ for a clearer idea).
Dario Amodei, CEO of Anthropic and one of the fathers of OpenAI alongside Sam Altman, says that the risks associated with AI are alarming. He parted ways with Altman precisely because of concerns about the distorted use of such a powerful and uncontrollable tool. In a recent piece in Il Sole 24 Ore (28 January), Amodei lists the risks, which include mass unemployment, fraud, dictatorial control over public opinion and market manipulation. He also calls for a ‘Constitution’ that establishes principles for the ethical development of AI. This is a minority position in a world of enthusiasts for all the opportunities offered by AI, especially for scientific and technological research and for the productivity of the most innovative companies.
However, the question of how to better understand and govern the evolution of AI remains open, as does the issue of how users can learn to interact with it.
A good example comes from Bocconi University in Milan, where Rector Francesco Billari has just announced a new course called ‘Cognitive Sciences and Human Behaviour’, which will be introduced in 2027 and will be compulsory for all Bocconi students. The aim is to ‘study human intelligence with AI and put people at the centre’ (QN/Il Giorno, 27 January). Billari adds, ‘We will open scientific research centres to develop criteria for the scientific evaluation of the impact of social policies and to verify the consequences of applying AI in the social sciences’.
This, too, is a good application of Rodari’s advice to build ‘a good education’ and work on conscious words, also recalling Robert Putnam‘s civil lesson that democracy requires deliberation and direct interaction between citizens. Without this, ‘public space is reduced to a technological convenience that makes us, paradoxically, more connected but profoundly alone and politically inert’ (Paolo Benanti, Il Sole24Ore, 28 January).
The silent erosion of institutions poses a risk to us. An AI system that has been fed all the vocabulary related to the Constitution could be a lifeline, albeit a fragile one.
(photo Getty Images)
‘All the uses of words to everyone… not because everyone should be an artist but because no one should be a slave.’ This quote by Gianni Rodari, taken from ‘The Grammar of Fantasy’ (Einaudi, 1973), has become increasingly pertinent as the debate on AI continues to focus on its risks and the possibility that ChatGPT and other generative systems will independently create a distorted reality beyond human understanding. These systems could create their own vocabulary and language, in short, their own abstract, and therefore false and manipulative, version of the idea that nomen est consequentia rei (the name is the consequence of the thing) dramatically altering and overturning its values and meaning.
All the uses of words to everyone, then. Let’s start from that. From educating AI to the necessary skills and limitations to govern and steer its evolution, without fear. As early as 1966, Umberto Eco warned us against fictitiously dividing ourselves into ‘apocalyptic’ and ‘integrated’ groups in response to the unexpected evolution of television. He advised us to analyse, understand and incorporate any overwhelming technological innovation into the order of human knowledge. From Gutenberg’s moveable type to quantum physics, and from the internet to AI. Let’s be honest, however, that this is becoming increasingly difficult.
Which words are we talking about using that provide value? Let’s do some calculations. Italian is one of the richest and most articulate languages, with between 215,000 and 280,000 basic words (including technical, literary and regional terms), amounting to two million forms with all their variations. A good dictionary, such as Treccani or Zanichelli, includes between 100,000 and 160,000 words. The average educated person’s vocabulary is around 50,000 words. De Mauro speaks of 7,500 ‘basic words’, of which only 2,000 are used daily. This is a paltry number, providing a simplified and therefore misleading understanding.
These figures highlight the first requirement: the ability to understand, correctly use and distinguish at least several tens of thousands of words. This is not only to avoid falling into the ever-growing category of ‘functionally illiterate’ people (those unable to understand a medium-complexity text or perform elementary calculations, accounting for a third of the Italian population), but also to enable the formulation of precise and relevant questions about the functioning of AI (which advances through increasingly precise and probing questions).
In other words, we could update Rodari’s call for ‘democratic education’ (i.e. a broad and widespread education as referred to in the Constitution with regard to the responsibility for quality school education) when we consolidate ‘all the uses of words to everyone’ into a vocabulary of around 50,000 words. This is much more than the 2,000 words we use every day, such a low number that it makes us ‘slaves’ to manipulation, propaganda, instrumental simplifications, factoids and fake news. This makes responsible participation in ‘well-informed’ and therefore ‘critical’ public discourse impossible, as Jürgen Habermas demonstrated in Europe and Antonio Gramsci, Don Lorenzo Milani and Benedetto Croce demonstrated in Italy.
The question now is whether generative AI will ever be capable of such creativity that it can coin verses such as ‘I am illuminated by immensity’ or ‘to spend the afternoon, absorbed and pale’. To tell us that ‘we are like leaves in autumn on branches’ or remind us that ‘there’s rosemary, that’s for remembrance…and there is pansies, that’s for thoughts’. Could it fully set out the terms of the ‘wager on God’, describe being ‘in the grip of abstract furies’, delve into the sphere of doubt about the scientist’s responsibility in the face of the atomic weapon, quote ‘the winter of our discontent’ and inform us that ‘even the bramble had its bends of sweetness, even the plum tree its candour’, without drawing on its memory of Giuseppe Ungaretti, Eugenio Montale, William Shakespeare, Elio Vittorini, Lucio Piccolo, Blaise Pascal and the fathers of physics, Niels Bohr and Werner Karl Heisenberg. And quote ‘the Wine-Dark Sea’ to describe the murky darkness of its depths.
AI can recite poetry and combine different and similar verses, and it may even invent rhymes that sound like those from non-existent but plausible poems. However, it is not yet capable of generating original thoughts full of poetic and literary meaning.
Will it be able to do so in the future? The danger is there. Perhaps this is not even the greatest danger. We all know how important poetry and literature are in shaping the human soul, defining behaviour and inspiring great cultural and civic choices that have a radical impact on mass movements, the tastes of millions of people and their passions and aspirations (read or reread Elias Canetti‘s ‘The Holy Fire’ in ‘Crowds and Power’, as well as Alessandro Manzoni‘s ‘The Betrothed’ and Sigmund Freud‘s ‘Group Psychology and the Analysis of the Ego’ for a clearer idea).
Dario Amodei, CEO of Anthropic and one of the fathers of OpenAI alongside Sam Altman, says that the risks associated with AI are alarming. He parted ways with Altman precisely because of concerns about the distorted use of such a powerful and uncontrollable tool. In a recent piece in Il Sole 24 Ore (28 January), Amodei lists the risks, which include mass unemployment, fraud, dictatorial control over public opinion and market manipulation. He also calls for a ‘Constitution’ that establishes principles for the ethical development of AI. This is a minority position in a world of enthusiasts for all the opportunities offered by AI, especially for scientific and technological research and for the productivity of the most innovative companies.
However, the question of how to better understand and govern the evolution of AI remains open, as does the issue of how users can learn to interact with it.
A good example comes from Bocconi University in Milan, where Rector Francesco Billari has just announced a new course called ‘Cognitive Sciences and Human Behaviour’, which will be introduced in 2027 and will be compulsory for all Bocconi students. The aim is to ‘study human intelligence with AI and put people at the centre’ (QN/Il Giorno, 27 January). Billari adds, ‘We will open scientific research centres to develop criteria for the scientific evaluation of the impact of social policies and to verify the consequences of applying AI in the social sciences’.
This, too, is a good application of Rodari’s advice to build ‘a good education’ and work on conscious words, also recalling Robert Putnam‘s civil lesson that democracy requires deliberation and direct interaction between citizens. Without this, ‘public space is reduced to a technological convenience that makes us, paradoxically, more connected but profoundly alone and politically inert’ (Paolo Benanti, Il Sole24Ore, 28 January).
The silent erosion of institutions poses a risk to us. An AI system that has been fed all the vocabulary related to the Constitution could be a lifeline, albeit a fragile one.
(photo Getty Images)