Che cosa resta della teoria degli strumenti cognitivi?

Incipit

Da qualche anno sembra che l’intelligenza artificiale abbia cambiato completamente il rapporto tra tecnologie e apprendimento. In realtà, molte delle domande fondamentali erano già state poste trent’anni fa da studiosi come David Jonassen. Vale la pena tornare a quelle riflessioni, non per nostalgia, ma perché possono aiutarci a capire che cosa, oggi, è davvero cambiato e che cosa, invece, continua a rappresentare un punto fermo.

Credo sia più che appropriato accostare le tante concettualizzazioni di David Jonassen, scienziato della mente e psicologo cognitivista, all’intelligenza artificiale, non tanto perché ne abbia anticipato esplicitamente e consapevolmente l’avvento — ci ha lasciati nel 2012, quando l’IA era ancora materia per pochi addetti ai lavori — quanto perché già decenni fa aveva lucidamente individuato le questioni fondamentali che la mente umana si trova ad affrontare quando le “macchine” sono in grado di svolgere attività abitualmente affidate agli esseri umani, alleggerendoci della fatica e, talvolta, svolgendo quei compiti meglio e più velocemente di noi.

Già all’inizio dell’era delle tecnologie digitali, le menti più avvedute avevano guardato alle loro potenzialità non tanto con il timore di perdere spazi propriamente umani, quanto con l’obiettivo di individuare una distribuzione intelligente dei compiti tra persona e macchina, capace di valorizzare — e persino ampliare — il contributo umano senza disperdere l’umanità tra gli ingranaggi della tecnologia.

È opportuno far fare alle tecnologie le attività che possono svolgere meglio dell’essere umano e riservare all’essere umano ciò che gli riesce meglio.

Da tempo circola lo slogan Stay human quando si affrontano questi temi. Jonassen, con il quale ho avuto il privilegio e la fortuna di lavorare per alcuni anni in importanti e pionieristici progetti dedicati al rapporto tra didattica e tecnologie, non parlava di software specifici. Parlava di una teoria del rapporto fra mente e tecnologia. Ed è proprio questa teoria che oggi torna di straordinaria attualità e che, a mio avviso, merita di essere riletta e attualizzata.

La tesi di Jonassen

Se dovessi riassumere il pensiero di Jonassen in una sola frase, sceglierei questa:

Non impariamo dalla tecnologia, ma con la tecnologia.

È probabilmente la sua affermazione più nota:

Learning with technology, not from technology.

Negli anni Novanta questa era una critica molto forte al software didattico che pretendeva di “insegnare” allo studente. Qualcuno parlava addirittura di “macchine per insegnare”; ancora oggi non è raro imbattersi nella pubblicità di robot presentati come sostituti degli insegnanti.

Secondo Jonassen il computer non doveva essere un tutor. Doveva diventare un partner cognitivo, un alleato del pensiero. Non uno strumento per pensare con meno fatica, ma uno strumento per pensare in modo più impegnativo, più rigoroso, più profondo.

Il concetto di Mindtools

L’idea più importante elaborata da Jonassen è quella dei Mindtools.

Non strumenti che fanno il lavoro al posto dello studente, ma strumenti che lo costringono a pensare.

Tra gli esempi ricorrono mappe concettuali, database, fogli elettronici, sistemi esperti, simulazioni, ipertesti costruiti dagli studenti.

Perché proprio questi strumenti?

Perché costruire una mappa concettuale significa distinguere concetti, stabilire relazioni, costruire gerarchie, esplicitare significati. È l’attività cognitiva richiesta da questo processo che produce apprendimento, non il software in sé.

Eppure, già nei primi anni della diffusione delle tecnologie digitali nella scuola, si parlava di courseware, quasi a testimoniare quel vizio antico di informatizzare perfino il lessico della didattica, attribuendo al software un ruolo educativo che appartiene invece all’attività mentale dello studente.

Jonassen definiva i Mindtools come strumenti con cui gli studenti costruiscono conoscenza, non strumenti dai quali la ricevono.

Una frase che oggi acquista un significato completamente nuovo.

Una frase che oggi assume un nuovo significato

In un articolo del 1998 Jonassen scrive, in sostanza, che la tecnologia dovrebbe essere utilizzata come amplificatore cognitivo, non come sostituto del pensiero.

Nel 1998 parlava di fogli elettronici, database e sistemi di rappresentazione.

Oggi potremmo sostituire quei termini con una sola espressione:

Large Language Model.

Andare oltre Jonassen

Jonassen dava quasi per scontato un presupposto.

Che gli strumenti cognitivi costringessero comunque lo studente a pensare.

Una mappa concettuale non si costruisce da sola. Un database non si progetta da solo. Un sistema esperto richiede un enorme lavoro cognitivo.

L’intelligenza artificiale rompe questo presupposto.

Per la prima volta compare uno strumento capace di produrre direttamente il risultato finale.

È questo l’elemento che modifica radicalmente il quadro teorico.

Il principio di Jonassen va riformulato

Oggi riscriverei così quel principio.

1995

Le tecnologie devono sostenere il pensiero.

2026

Le tecnologie devono sostenere il pensiero senza sostituire l’attività cognitiva che genera apprendimento.

Quella seconda parte, trent’anni fa, non era necessaria. Non perché mancasse una riflessione pedagogica, ma perché semplicemente nessuna macchina era in grado di “pensare” al posto nostro.

Oggi, invece, diventa indispensabile ribadire la centralità dell’attività cognitiva umana e, nello stesso tempo, chiarire che cosa significhi davvero pensare come esseri umani quando anche le macchine producono linguaggio, ragionamenti, immagini, codice e altri artefatti cognitivi.

Occorre rendere trasparenti le caratteristiche del pensiero umano, distinguendole da quelle del “pensiero” delle macchine.

Una nuova categoria

Credo che oggi sia possibile introdurre un concetto che Jonassen non avrebbe potuto formulare.

Lo chiamerei, almeno provvisoriamente, trasparenza cognitiva.

Una tecnologia è educativa quando rende visibile il processo mentale.

È anti-educativa quando rende visibile soltanto il prodotto.

L’intelligenza artificiale rischia continuamente questa seconda deriva.

Jonassen distingueva nettamente due modalità di utilizzo della tecnologia:

• tecnologia come insegnante;
• tecnologia come strumento cognitivo.

Oggi, forse, dobbiamo introdurne una terza.

Tecnologia come interlocutore cognitivo.

È una categoria nuova.

ChatGPT e gli altri sistemi di intelligenza artificiale non sono soltanto strumenti. Sono interlocutori con cui si dialoga.

Questo modifica profondamente il tipo di attività mentale richiesta.

Non si tratta più soltanto di costruire una rappresentazione della conoscenza. Si tratta di formulare domande, criticare risposte, negoziare significati, verificare informazioni, argomentare, confrontare ipotesi.

È una forma di pensiero dialogico che Jonassen aveva soltanto sfiorato quando parlava degli strumenti di conversazione, ma che oggi assume una centralità completamente diversa.

Con questa premessa non credo sia azzardato mettere, almeno sul piano concettuale, Jonassen in dialogo con l’intelligenza artificiale.

Non per sostenere che Jonassen “avesse già previsto tutto”, ma per porre una domanda molto più interessante:

che cosa del modello dei Mindtools resiste, che cosa deve essere modificato e che cosa va completamente ripensato quando il Mindtool diventa un modello linguistico capace di generare testi, idee, immagini, codice e molti altri artefatti cognitivi?

Una possibile archeologia delle idee

Come primo filone teorico di questo progetto mi piacerebbe costruire una sorta di archeologia delle idee, individuando gli autori che, ben prima dell’avvento dell’intelligenza artificiale, hanno preparato il terreno per una riflessione educativa oggi più attuale che mai.

• Lev Vygotskij, con la teoria degli strumenti culturali come mediatori del pensiero.
• Roy Pea, con il concetto di cognitive technologies e di cognizione distribuita.
• David Perkins, con l’idea della person-plus, la persona insieme al proprio ambiente cognitivo.
• David Jonassen, con le tecnologie intese come Mindtools, strumenti con cui si apprende e non dai quali si apprende.
• Edwin Hutchins, con la teoria della cognizione distribuita nei sistemi socio-tecnici.

E infine la domanda del 2026:

che cosa cambia quando il partner cognitivo non è più uno strumento da manipolare, ma un interlocutore linguistico generativo?

Ho l’impressione che proprio qui si apra un territorio ancora poco esplorato.

Mi piacerebbe provare a prolungare una tradizione teorica iniziata molto prima dell’intelligenza artificiale, verificando se sia ancora capace di interpretarne l’avvento oppure se richieda una revisione profonda.

Più che partire dall’intelligenza artificiale — come mi pare stia accadendo in molta della riflessione attuale — e cercare una teoria educativa che la giustifichi, si potrebbe percorrere la strada inversa.

Partire da una teoria dell’apprendimento costruita in quarant’anni di studi e domandarsi che cosa dell’intelligenza artificiale confermi quella teoria, che cosa la metta in crisi e che cosa costringa a rivedere.

Mi sembra un’impostazione epistemologica profondamente diversa.

Ed è forse proprio questa inversione di prospettiva che può consentirci di comprendere meglio non soltanto l’intelligenza artificiale, ma anche il significato che oggi attribuiamo all’apprendere e al pensare.

English version (ChatGPT)

From Mindtools to Large Language Models

What Remains of Jonassen’s Theory of Cognitive Tools?

Introduction

Over the past few years, the emergence of generative artificial intelligence has led many to believe that the relationship between technology and learning has fundamentally changed. Yet many of the questions we are asking today had already been raised more than thirty years ago by scholars such as David Jonassen. Revisiting those ideas is therefore not an exercise in nostalgia. It is an opportunity to understand what has genuinely changed—and what remains a fundamental principle for thinking about education in the age of AI.

In my view, it is both appropriate and intellectually fruitful to place David Jonassen’s work in dialogue with contemporary artificial intelligence. Not because he explicitly anticipated the emergence of Large Language Models—he passed away in 2012, when AI was still a concern for a relatively small community of researchers—but because he had already identified the central cognitive issues that arise whenever machines become capable of performing tasks traditionally carried out by human beings.

His fundamental question was never whether technology would replace people. Rather, it was how human cognition changes when technologies assume part of the intellectual work that was once entirely ours, often performing it more efficiently, more quickly, and sometimes more accurately than we can.

Even at the beginning of the digital era, the most insightful researchers did not approach educational technologies primarily with the fear that machines would erode what makes us human. Instead, they sought an intelligent distribution of cognitive work between people and technology, one that could extend rather than diminish human intellectual activity.

One principle, in particular, has lost none of its relevance:

Technology should perform the tasks it can do better than humans, while humans should devote their efforts to what they do best.

Today this idea is often summarized by the slogan Stay human. Long before that expression became popular, however, Jonassen had already developed a theoretical framework grounded in precisely this concern.

I had the privilege of working with David Jonassen for several years on pioneering projects exploring the relationship between educational technology and learning. What impressed me most was that he was never primarily interested in software itself. His real concern was understanding the relationship between technology and human thinking.

That is precisely why I believe his work deserves to be revisited today. Not because it offers ready-made answers to generative AI, but because it provides a conceptual framework that may help us ask better questions about what it now means to think, learn, and teach when our cognitive partners are no longer merely digital tools, but conversational systems capable of generating language, images, code, and increasingly sophisticated intellectual artifacts.

---

Jonassen’s Central Thesis

If I were asked to summarize Jonassen’s educational philosophy in a single sentence, I would choose the statement for which he is probably best known:

We do not learn from technology; we learn with technology.

Or, in Jonassen’s own words:

Learning with technology, not from technology.

When this idea was introduced in the 1990s, it represented a strong criticism of the dominant view of educational software. Much of the software developed at that time was designed to teach students directly, as if computers could assume the role traditionally played by teachers. In some educational circles, the expression teaching machines was still part of the vocabulary. Even today, advertisements occasionally promise AI-powered tutors or robots capable of replacing teachers in the classroom.

Jonassen challenged that entire perspective.

For him, computers were never supposed to become tutors. They were meant to become cognitive partners—tools that help learners think rather than systems that think on their behalf.

This distinction is subtle, yet fundamental.

The educational value of a technology does not lie in the amount of information it delivers, nor in its ability to automate instruction. Its value lies in the kind of thinking it requires from the learner.

Technology, in other words, should not enable learners to avoid intellectual effort. It should engage them in what Jonassen called thinking hard: thinking more deeply, more rigorously, and with greater cognitive commitment.

The purpose of a cognitive tool was not to make thinking easier, but to involve learners in forms of reasoning that were demanding, deliberate, and meaningful.

Learning is generated by the learner’s cognitive activity, not by the sophistication of the software.

This idea remains remarkably relevant today.

However, the emergence of generative artificial intelligence forces us to ask a new question—one that Jonassen himself never had to confront:

What happens when a cognitive partner becomes capable of producing the intellectual artifact that learners were previously expected to construct themselves?

That question, I believe, marks the transition from Jonassen’s world to ours.

The Concept of Mindtools

Perhaps Jonassen’s most influential contribution was his concept of Mindtools.

Mindtools are not technologies that perform intellectual work for learners. They are technologies that engage learners in intellectual work.

Jonassen described them as cognitive tools through which students construct knowledge rather than receive it.

The distinction is essential.

A Mindtool does not produce learning because of what it does. It produces learning because of what learners are required to do while using it.

Typical examples included concept maps, databases, spreadsheets, expert systems, simulations, and hypermedia environments designed and constructed by students themselves.

Why these tools?

Because building a concept map requires learners to identify concepts, establish relationships, organize hierarchies, make distinctions, and explain meanings.

Designing a database requires them to classify information, define categories, identify attributes, and make explicit decisions about how knowledge should be represented.

The software itself does not generate learning.

The cognitive activity does.

Ironically, even during the early years of educational computing, much of the discourse surrounding digital technologies continued to describe software as if it were the active agent of instruction. Expressions such as courseware reflected an enduring tendency to attribute educational power to technology rather than to the learner’s intellectual activity.

Jonassen argued exactly the opposite.

Knowledge is not transmitted by cognitive tools.

It is constructed through the thinking they require.

Looking back today, this distinction appears even more significant than it did thirty years ago.

In 1998 Jonassen wrote, in essence, that technology should function as a cognitive amplifier, not as a substitute for human thought.

At the time he was referring to spreadsheets, databases, simulations, and other forms of educational software.

Today, one could replace that entire list with three words:

Large Language Models.

The principle, however, remains the same.

Or does it?

That question, I believe, leads us to the real challenge posed by generative artificial intelligence.

Beyond Jonassen

There is, however, one assumption that Jonassen could reasonably take for granted.

Cognitive tools, by their very nature, required learners to think.

A concept map could not generate itself.

A database did not design its own structure.

An expert system demanded extensive knowledge engineering before it could function.

In every case, the intellectual artifact was the visible outcome of a cognitive process carried out by the learner.

Generative artificial intelligence changes this assumption.

For the first time, we are confronted with technologies capable of producing the intellectual artifact directly.

A concept map can be generated in seconds.

An essay can be written almost instantly.

A computer program can be produced from a brief textual description.

Images, lesson plans, summaries, explanations, even research proposals can all be generated with astonishing speed.

This is not simply another technological innovation.

It represents a qualitative shift in the relationship between technology and cognition.

The central educational question is therefore no longer whether technology supports thinking.

The question is whether thinking still takes place when technology is capable of producing the very artifact through which thinking was previously made visible.

This, I believe, is where Jonassen’s framework must be extended.

Not because it has become obsolete.

On the contrary.

Its fundamental insight—that learning depends on the learner’s cognitive activity—has never been more relevant.

What has changed is the nature of the cognitive partner.

For the first time, the partner is no longer merely a tool manipulated by the learner.

It is a generative system capable of producing meaningful intellectual outputs on its own.

Consequently, Jonassen’s original principle may need to be reformulated.

In the 1990s it was sufficient to argue that educational technologies should support thinking.

Today we must add an essential qualification.

Educational technologies should support thinking without replacing the cognitive activity through which learning is constructed.

Thirty years ago this clarification was unnecessary.

Not because educational theory was incomplete, but because no technology was capable of producing, autonomously, the intellectual work that learners were expected to perform themselves.

Today that condition no longer holds.

For this reason, preserving the centrality of human cognitive activity becomes not simply an educational preference, but a fundamental pedagogical principle.

At the same time, we are challenged to define more precisely what it now means to think as human beings when machines themselves are capable of generating language, arguments, images, software, and many other forms of intellectual production.

The educational problem is therefore no longer whether machines can “think.”

It is understanding which forms of human thinking remain indispensable for learning when intelligent machines can perform an increasing proportion of our cognitive work.

A New Concept: Cognitive Transparency

If Jonassen were writing today, I suspect he would still distinguish between technology used as a teacher and technology used as a cognitive tool.

Yet I believe that generative artificial intelligence requires us to introduce an additional conceptual category.

I would tentatively call it cognitive transparency.

A technology becomes educational when it makes the learner’s thinking visible.

It becomes anti-educational when it reveals only the final product while concealing the cognitive process that generated it.

This distinction has always mattered.

With generative AI, however, it becomes decisive.

A learner may submit a beautifully written essay, a sophisticated concept map, or an elegant computer program without having engaged in the intellectual work those artifacts traditionally represented.

The visible product may no longer provide reliable evidence of the underlying cognitive activity.

This is, perhaps, the most profound educational implication of generative artificial intelligence.

Assessment has always relied, at least implicitly, on the assumption that there is a meaningful relationship between cognitive effort and intellectual production.

Generative AI weakens that assumption.

Consequently, educators can no longer evaluate learning simply by examining what students produce.

They must increasingly ask how that product came into being.

The educational focus therefore shifts from the artifact itself to the process through which the artifact is constructed, discussed, revised, questioned, and ultimately understood.

From this perspective, the central issue is not whether students use artificial intelligence.

The real question is whether artificial intelligence makes their thinking more visible—or hides it.

This is what I mean by cognitive transparency.

A cognitively transparent use of AI enables learners to articulate their reasoning, examine alternatives, justify decisions, evaluate generated responses, and reflect on the dialogue that led to the final outcome.

An opaque use of AI does the opposite.

It delivers answers while concealing the intellectual path through which genuine learning develops.

If this distinction proves useful, cognitive transparency may become one of the criteria by which educational uses of generative AI can be evaluated.

Reformulating Jonassen’s Principle

If Jonassen were writing today, I believe one of his central principles would require a careful reformulation.

In the mid-1990s, it was sufficient to state that educational technologies should support thinking.

Today, I would reformulate that principle as follows.

1995

Educational technologies should support thinking.

2026

Educational technologies should support thinking without replacing the cognitive activity through which learning is constructed.

Thirty years ago, this additional qualification was unnecessary.

Not because educational theory was incomplete, but because no technology was capable of performing the intellectual work that learners themselves were expected to carry out.

Today, that assumption no longer holds.

Generative artificial intelligence compels us to reaffirm the centrality of human cognitive activity while, at the same time, asking a new question:

What does it mean to think as human beings when machines are also capable of generating language, arguments, images, software, and other intellectual artifacts?

This question is not merely technological.

It is educational.

More fundamentally, it is epistemological.

It requires us to distinguish more clearly between human thinking and machine-generated outputs, and to identify those forms of cognitive activity that remain indispensable for meaningful learning.

From this perspective, generative AI does not invalidate Jonassen’s theory.

Rather, it reveals the need to make explicit an assumption that his generation of cognitive tools could safely leave implicit:

Learning depends not simply on the production of intellectual artifacts, but on the cognitive activity through which those artifacts are constructed.

A New Category: Technology as a Cognitive Interlocutor

Jonassen drew a clear distinction between two ways of using technology in education.

Technology could function as:

• a teacher;
• a cognitive tool.

Today, I believe we may need to introduce a third category.

Technology as a cognitive interlocutor.

This is something genuinely new.

Systems such as ChatGPT and other generative AI applications are no longer simply tools that learners manipulate. They are conversational systems with which learners engage in dialogue.

This changes the nature of the cognitive activity involved.

The task is no longer limited to constructing an external representation of knowledge.

It increasingly involves formulating questions, critically examining responses, negotiating meanings, verifying information, evaluating arguments, and justifying conclusions.

In other words, learning becomes, to a significant extent, a dialogical activity.

Jonassen had already recognized the educational importance of conversation technologies, but generative AI gives this dimension an entirely new significance.

With this premise, I do not believe it is unreasonable to place Jonassen’s theoretical framework in dialogue with generative artificial intelligence.

Not to claim that Jonassen anticipated today’s technologies.

Rather, to ask a different question:

What remains of the Mindtools model? What needs to be revised? And what must be fundamentally rethought when the “Mindtool” becomes a language model capable of generating text, ideas, images, software, and many other forms of intellectual artifacts?

Those, it seems to me, are the questions that deserve our attention today.

An Archaeology of Ideas

As a first step in this line of inquiry, I would like to undertake what might be called an archaeology of ideas.

Rather than beginning with artificial intelligence itself, I propose revisiting those scholars who, long before the emergence of generative AI, laid the conceptual foundations for understanding technology as a mediator of human cognition.

Among them are:

• Lev Vygotsky, who conceived cultural tools as mediators of thought.
• Roy Pea, with his work on cognitive technologies and distributed intelligence.
• David Perkins, through the concept of the person-plus, in which thinking extends beyond the individual to include the surrounding cognitive environment.
• David Jonassen, who argued that technologies should function as Mindtools—tools with which people learn, rather than tools from which they learn.
• Edwin Hutchins, whose theory of distributed cognition demonstrated how cognitive processes emerge across people, artifacts, and social systems.

These perspectives all emerged well before the advent of generative artificial intelligence.

Yet they may still provide the most appropriate conceptual framework for understanding its educational implications.

This ultimately leads to what I believe is the central question of our time:

What changes when our cognitive partner is no longer a tool that we manipulate, but a generative linguistic interlocutor?

My impression is that this remains a largely unexplored territory.

I would like to explore whether a theoretical tradition that began decades before the emergence of AI is still capable of interpreting this new reality—or whether it now requires a profound conceptual revision.

This, in my opinion, is a more promising path than the one currently followed in much of the contemporary debate.

Too often we begin with artificial intelligence and then search for an educational theory capable of accommodating it.

Perhaps we should reverse the direction of inquiry.

We should begin with theories of learning that have been developed over decades of cognitive and educational research, and then ask:

What does generative artificial intelligence confirm? What does it challenge? And what does it force us to rethink?

This is not simply a different way of approaching educational technology.

It is, I believe, a different epistemological stance.

Artificial intelligence should not become the starting point from which educational theory is derived.

Rather, it should become the phenomenon through which existing theories of learning are questioned, refined, and, where necessary, transformed.

Only then can we understand not only what artificial intelligence is capable of doing, but also what it now means for human beings to think, to learn, and to teach.