PSI’s vision and commitments

Abstract: the essentials in three minutes

Connect knowledge. Bring ideas to life. In the age of artificial intelligence, let us cultivate our ability to understand, reason by analogy and imagine what connecting knowledge makes possible.

PSI grew from a conviction: access to information should support thinking that develops, confronts reality and remains able to choose. An immediately available answer can be useful. It does not replace the work through which a person understands a situation, recognises a limitation, connects experiences and decides what deserves attention. We want to make that work more accessible.

Reasoning by analogy brings together relationships from different situations. It can suggest a direction, provided that differences and limits are examined. Cognitive research has proposed models of this process and investigated some conditions of transfer. It establishes neither an infallible human faculty nor a permanent human monopoly: AI models also succeed at analogy tasks. PSI’s vision is to exercise this capacity, with machines when they help, without handing over our questions. (Gentner, 1983) (Webb et al., 2023)

This ambition calls for broad general knowledge and an understanding of systems: their elements, relationships, timescales and dependencies. It also requires room for personal experience and cultural diversity. In our interpretative framework, a word evokes a concept, sociocultural associations and a personal resonance. This teaching device brings several traditions of reflection on meaning into dialogue; it does not claim to reproduce a single theory belonging to Barthes or Todorov. (Barthes, 1964) (Todorov, 1966) (Kecskes, 2010)

Thema is central to this commitment. This multilingual classification provides common reference points for finding content beyond titles and languages. PSI adapts it to a video catalogue. A category identifies a subject; it cannot exhaust a document’s meanings or the uses a reader may make of it. Classification is a starting point, open to discussion. (EDItEUR, evolving standard)

In practical terms, PSI classifies information, produces summaries, highlights main points and facilitates access to chapters or key moments in a source. People choose what to explore further. Links to videos allow them to recover context, check a formulation and pursue their own route. Summaries and assessments remain liable to error: they are reading aids.

Ecological and technological transformations can provoke real concern. Studies of climate anxiety document, among other things, young people’s worries in several countries; they do not measure any effect of PSI. Our ambition is to make some uncertainty examinable and support the ability to understand and act, without promising to eliminate anxiety or guarantee the future. (Hickman et al., 2021)

Understand systems. Keep a say in our choices. A person’s value does not depend on beating a machine. PSI supports a human place grounded in curiosity, judgement, creation and shared responsibility.

1. Why PSI exists

Finding a way through information

A reader can have an Internet connection, know several search engines and still find it practically impossible to understand a question. Sometimes the vocabulary needed to formulate it is missing. Sometimes there is no accessible entry into a field. A two-hour video contains an essential passage, but its title does not identify it. Two speakers use the same term differently. A third presents a conclusion without recalling the conditions under which it applies. The obstacle is then more than the quantity of information: it concerns how information becomes discoverable and usable for a person.

PSI starts from this concrete problem. Its ambition is to help turn available content into opportunities for understanding. This calls for subject organisation, sufficiently explicit reading aids and a route back to sources. It also calls for respect for uncertainty. An intellectual map becomes misleading when it suggests that all its boundaries are natural, that every important question appears on it and that every item placed in a category is equivalent. A summary becomes misleading when it conceals the choices made in shortening its source.

We want to offer an infrastructure for reading that lets people begin somewhere, change direction and reconsider an interpretation. That formulation commits PSI to the quality of its reference points and the clarity of its limitations. It does not make the website a universal authority. We do not ask readers to hand their judgement over to a classification. We offer material with which to exercise it: an identifiable subject, a summary, points to examine, moments in the source to recover and routes towards other content.

What we know, what we think and what we propose

This essay connects three levels. The first concerns research findings: observations or experiments conducted with particular populations, arrangements and criteria. The second concerns interpretations and hypotheses: what those studies suggest, what they leave open and how we bring them together. The third concerns PSI’s commitments: the design choices and values we wish to uphold. Moving from one level to another requires an argument; inserting a citation is not enough.

An experiment showing that an activity helps students in a given situation does not establish that every use of a documentary website will produce the same effect. A neuroimaging study does not turn a creativity slogan into a biological finding. A philosophical discussion of autonomy does not measure a reduction in anxiety. Such distinctions make reading more precise. They allow an ambitious vision to be supported without claiming validation that it does not possess.

This text is an original editorial synthesis based on selected sources. It is not a systematic review of all intelligence research, a clinical assessment or a prediction of the date of the singularity. Classic references sit alongside recent work because a current question can have a long history. Historical theoretical texts and the Thema standard are distinguished from empirical research. Bibliographic titles retain their original language; PSI’s French and English explanations should not be mistaken for official translations of those publications.

2. Reasoning by analogy

What reasoning by analogy means

Saying that two things resemble each other is not enough to produce useful reasoning. Similarity of colour, name or outline may draw attention without explaining how a situation works. An analogy becomes more productive when it concerns relationships: how elements interact, how a constraint travels through a system, how a change produces an effect. Gentner’s structure-mapping theory distinguishes relationships between elements from isolated properties. This framework describes a possible analytical mechanism; it does not guarantee that every comparison is valid. (Gentner, 1983)

In the practice PSI wishes to encourage, an analogy begins with an explicit question. What are we trying to explain? Why might the source domain help us understand the target domain? Which relationships are being mapped? Which properties remain different? The analogy can then suggest a hypothesis or a strategy. It does not remove the need to test that suggestion against the target situation. Here, induction means moving towards a possibility to examine, with a revisable conclusion; it does not mean a necessary demonstration.

Imagine comparing a warehouse with a public service to examine queues. Both may involve arrivals, limited processing capacity and decisions about priorities. That relationship can help formulate questions about delays. But people are not interchangeable parcels, and the purpose of a public service cannot be reduced to processing speed. The analogy becomes useful when it helps distinguish these features. It becomes harmful when an apparent structural similarity is used to erase the differences that matter.

This example is an illustration proposed by PSI, rather than a result reported by an experiment. Its purpose is to show how a comparison can be worked on. A reader can retain the relationship concerning capacity while rejecting a conclusion concerning the value of the people affected. Reasoning by analogy involves both connection and restraint: finding a relationship and deciding how far it can responsibly be carried.

Transfer is not automatic

Gick and Holyoak studied conditions under which people could draw on an earlier example to solve another problem. Their work makes it possible to examine the difference between encountering a relevant example and recognising that it can be transferred. A useful relationship can remain unnoticed; guidance and comparison can matter in the situations studied. (Gick & Holyoak, 1983)

For PSI, this suggests a design question rather than an established effect: how can content be made available in a way that invites comparison? A subject label can help locate a document. A summary can clarify its main mechanism. Returning to a passage can make a relationship more precise. None of these steps automatically ensures transfer. They create conditions in which readers can choose to investigate it.

A broad knowledge base matters here because a comparison needs material. But breadth alone is insufficient. It is possible to recognise familiar words without understanding the relevant relationships. It is possible to remember a striking story and miss the constraint that made its solution work. Our ambition is therefore to combine accessibility with opportunities to examine structure, not merely to place many topics next to one another.

Machines can also make analogies

Webb and colleagues report successes on particular analogy tasks using a language model. These results concern specified models, instructions and tests. They neither establish universal machine understanding nor support the claim that analogy is an exclusive human capability. (Webb et al., 2023)

PSI’s vision does not need such an exclusive claim. We want people to retain opportunities to formulate questions, compare relationships and evaluate what a proposed analogy leaves out. An AI-generated comparison may be useful material for this activity. It still has to be assessed in relation to the problem, the source and the consequences of accepting it.

The emphasis is therefore on exercising a capacity and maintaining responsibility for its use. We can value human reasoning without declaring that machines will never perform a particular operation. A durable educational ambition should remain meaningful when technology changes. Connecting knowledge, examining differences and choosing what deserves attention are activities we want people to be able to undertake in that changing environment.

3. Knowledge for understanding the world

A foundation for asking better questions

General knowledge does not mean possessing every fact. It means having enough reference points to recognise a domain, identify a missing concept and find a more appropriate question. A reader may not know the details of a chemical process, but knowing that it involves material constraints can prevent an overly simple account of an energy technology. Familiarity with institutions can likewise reveal that a technical possibility is not yet a collective decision.

PSI wants to make such entry points accessible. A summary can introduce vocabulary. Thema can identify a subject and provide a route to related content. A source can deepen an explanation. This sequence should allow a reader to move from a broad question towards a more precise one, without being required to master a whole discipline before beginning.

The broad foundation we envisage is not a collection of fashionable claims. It includes definitions, mechanisms, methods, uncertainties and contexts. It should make the reader better able to recognise when a conclusion depends on a condition. It should also make it possible to distinguish a question of fact from a question about objectives or values. Those distinctions are useful when fields meet.

Elements, relationships and timescales

Understanding a system involves more than naming its components. It requires attention to relationships, behaviour, constraints and change over time. Hmelo-Silver and Pfeffer studied differences in how experts and novices represented complex systems. Their work helps frame the distinction between listing parts and understanding how a system operates, in the educational contexts they examined. (Hmelo-Silver & Pfeffer, 2004)

Consider a battery. A description can focus on its materials, its reactions or its performance in an application. A wider inquiry can ask where the materials come from, how electricity is supplied, how demand develops and what happens when the battery is replaced. These are related questions, but they are not all answered by the same source or the same measure. A system-oriented reading makes the relationships visible while preserving the need for specialised knowledge.

Such a reading also asks about timescales. A benefit observed at one moment may depend on maintenance, replacement or changes in use. A local improvement may transfer a constraint elsewhere. A collective decision may alter demand; a material limit may make another trajectory more relevant. PSI’s role is to help readers find and connect the information needed to examine these possibilities, rather than to declare a single account sufficient.

Breadth and depth need each other

PSI seeks to avoid two shortcuts. The first is accumulating brief overviews without checking what has been understood. The second is treating a field as a completely isolated territory. A reading route can alternate breadth and depth: discover a concept, examine a source, compare explanations, return to a definition and then follow a link into another subject. The quality of this route depends on more than the number of items consulted.

This alternation also needs room for incomprehension. Encountering an unfamiliar term is not a reader’s failure. A question becoming more precise can count as progress even if an answer comes less quickly. The website should help situate that question rather than create an immediate impression of mastery. We want a summary to indicate what it clarifies and provide access to what it cannot develop.

In practice, a reader may stop at an orientation-level understanding or decide to investigate a mechanism in detail. Both uses are legitimate. A useful documentary tool should make the difference visible and let people choose their next step. That freedom is part of the foundation for cross-domain reasoning: it allows a person to deepen the particular relationship on which a comparison depends.

4. How ideas emerge and develop

An idea needs work

In PSI’s vision, creativity is a practice of connecting, proposing, evaluating and revising. It can occur in a scientific investigation, an artwork, a professional project or an everyday decision. We do not reserve it for people presumed naturally creative. Nor do we promise that a documentary tool will produce it on its own. Making knowledge accessible can contribute to a creative environment; it is a working condition PSI seeks to improve.

The distinction between generation and evaluation matters. Finding many possibilities does not establish which ones fit. Choosing a single direction too early can prevent others from appearing. In both cases, context matters: the criteria of an artistic project differ from those of an experiment, a technical object or a public decision. Readers should be able to move from an appealing idea towards a formulation precise enough to examine its usefulness, consequences and limits.

A use of PSI might start by discovering a solution in one domain and asking whether its principle could serve elsewhere. Another source may then show why transfer is difficult. That difficulty does not make the first encounter worthless. It helps transform the idea. We want to encourage this process of revision, in which readers can change their minds without losing their intellectual route.

Machines can help and shape choices

Doshi and Hauser’s experiment on short-story writing found improvements in some individual evaluations with generative assistance, alongside a reduction in collective diversity. The finding concerns a particular writing arrangement. It does not establish that every cooperation with AI necessarily homogenises creations, but it encourages examination of both the quality of a product and the diversity of products. (Doshi & Hauser, 2024)

Wang and colleagues’ large-scale comparison examines measures of divergent creativity in humans and language models. Such measures concern responses to specified tasks. They cannot on their own represent the creativity of a life, a community or a culture. Care over criteria matters whether a result favours machines or people. (Wang et al., 2026)

For PSI, productive cooperation can retain several entry points: a machine’s proposal, a reader’s objection, an external source and experience absent from the initial proposal. The website should not become a mechanism for reproducing a single plausible reading of every subject. Different authors, access to original formulations and movement between categories should make it possible to encounter a real diversity of approaches.

Human worth goes beyond creative performance

A work can be remarkable without its author scoring higher on a laboratory task. Someone can play an essential role in a group by helping others formulate a question, notice a contradiction or make a solution accessible. We do not want to reduce those contributions to a single indicator. Performance language can help compare a specified task; it becomes inadequate when it claims to establish a person’s value.

This distinction matters for the future envisaged by PSI. Even if artificial systems become highly competent in many domains, the question of our participation remains open. We can seek to widen opportunities to understand and create rather than make meaningful existence conditional on retaining a technical advantage. This is PSI’s commitment: support a human place that does not depend on permanently defeating machines.

Such a commitment also leaves room for collaboration and ordinary contributions. A person does not have to produce a revolutionary discovery to help a community think more clearly. Noticing an ambiguity, sharing experience or making a difficult explanation accessible can matter. These forms of contribution deserve opportunities for expression alongside more visible achievements.

5. What neuroscience tells us

The brain networks involved in creativity

Creativity research does not support a simple opposition between a rational brain and an imaginative brain. Beaty and colleagues discuss work in which creative thinking involves interactions between networks associated with internally generated thought and cognitive control, among other functions. Their review offers a framework for interpreting tasks; it does not describe a universal creativity switch. (Beaty et al., 2016)

A later study examines relationships between functional connectivity and individual differences in creativity measures. Statistical prediction should be distinguished from a complete explanation of a person. A relationship observed in those data does not establish that activity on PSI will change brain networks in a desired direction. (Beaty et al., 2018)

Shofty and colleagues use an intervention in a clinical context to investigate a causal relationship between the default network and a creative-thinking task. Its interest lies in going beyond correlation; its scope remains tied to the context, participants and task. We do not turn it into advice for brain training or evidence of the website’s effectiveness. (Shofty et al., 2022)

Meaning and the use of meaning

Lambon Ralph and colleagues’ review of semantic cognition distinguishes mechanisms for representing knowledge and controlling its use according to circumstances. It helps frame the difference between possessing knowledge and selecting knowledge relevant to a task. This distinction does not mean that a digital catalogue reproduces the organisation of the brain. (Lambon Ralph et al., 2017)

PSI’s design can take this question seriously without borrowing unwarranted neurological prestige. Content exists in a catalogue; readers still need to find it for the question they are asking. A summary exists; they still need to recognise when it is insufficient. A category exists; they still need to explore another route. The interface’s task is to make these actions possible. They are documentary functions that can be described directly without claiming that they trigger a particular biological transformation.

The same caution applies to claims about creativity. A research result can illuminate an aspect of a task without defining all creative activity. An illustration of brain networks can be informative without becoming a prescription for how to read. We want scientific explanations to sharpen the question, not merely make a promise appear more authoritative.

The limits of neuroscientific findings

Our editorial rule is straightforward: when discussing the brain, describe what research actually studied. Avoid claims that a feature automatically increases intelligence, activates a creative region or preserves a faculty against every technological change. Interesting findings can inform educational reflection, but they do not remove the need to evaluate the real uses of a tool.

This leaves room for ambition. We can want reading to involve active participation, navigation to permit comparison and sources to remain accessible. We can suggest reformulation exercises and invite readers to recognise differences between explanations. These commitments concern thinking practices. Their interest does not require a neuroscientific guarantee.

6. Artificial intelligence, world models and robotics

Rapid progress, an uncertain future

“Artificial intelligence” covers different systems and uses. A response generated from an instruction, a model trained for recognition and a system learning to act in an environment do not pose exactly the same questions. PSI’s vision should not depend on a frozen picture of technology. It should allow methods, results and limitations to be followed, and questions to change as capabilities change.

We do not treat the hypothesis that AI is already an absolute expert in every domain as a scientific finding. Nor do we set a date when it will become one. The prospect of very widely available powerful systems serves here as a scenario for thinking about education, autonomy and participation. A scenario can be useful even when its timetable and form remain uncertain, provided it is not presented as an observed fact.

Systems that learn consequences

Hafner and colleagues present DreamerV3, a method using world models to learn behaviour across many control tasks. Within that framework, a model represents possible developments of an environment and helps improve decisions through simulated scenarios. Success in the studied environments does not guarantee mastery of every real-world situation. (Hafner et al., 2025)

Work on RT-2 brings vision-and-language models together with robotic control. It examines how large-scale learning can contribute to generalisation in manipulation tasks. The research shows why discussing AI’s future solely as text generation would be too narrow; it does not establish unlimited robotic autonomy across all environments. (Zitkovich et al., 2023)

For PSI, these developments provide another reason to connect domains. A robotics question may lead towards models, sensors, materials, energy, work, uses and collective rules. A device cannot be understood solely through its most spectacular demonstration. We need to know what was learned, under which conditions, with which resources and how results are evaluated.

Human choices behind technical performance

Imagine a system able to carry out an operation efficiently. Efficiency still needs to be specified: speed, cost, safety, consumption, output quality or ease of control. A gain under one criterion can leave other consequences to examine. People affected by a system may give different weights to criteria. A decision to use it therefore also involves discussion of objectives and how effects are distributed.

This reflection does not rest on a claim that AI cannot discuss values. A model can produce arguments about them. PSI’s commitment concerns maintaining a real possibility for people to understand choices affecting them, challenge those choices and contribute to their formulation. Producing arguments and organising collective decisions are activities whose responsibilities need to be made explicit.

7. The human place in a changing society

Usefulness does not exhaust a person’s value

A feeling of becoming useless can arise when a human place is defined exclusively through a task a machine might perform. PSI wants to shift the central question: which opportunities to understand, participate, create and care do we want to maintain and widen? This does not deny the material effects of technological change. It asks us to consider both people’s capacities and the conditions that allow those capacities to be exercised.

Human dignity is a commitment of PSI. We do not deduce it from an experimental finding. We reject tying a person’s value to their ranking in a productivity contest. Curiosity, attention to a collective difficulty, transmission of experience and creation of relationships can matter without setting records. A society making those contributions invisible does not solve that invisibility merely by offering a new high-performing tool.

Work and automation

Acemoglu and Restrepo studied effects of exposure to industrial robots in American labour markets over a defined period. Their findings illuminate changes in employment and wages in that context. They cannot mechanically predict every effect of generative models or future robotics, but they recall that automation can alter living conditions and the distribution of benefits. (Acemoglu & Restrepo, 2020)

It would therefore be inadequate to answer every concern with “just learn”. Learning can increase the capacity to participate; it does not replace institutions, rights, resources and collective decisions. A person can understand a transformation without having the means to change it alone. PSI seeks to make useful information more accessible without assigning each individual full responsibility for changes beyond their action.

The conditions for autonomy

Ryan and Deci’s self-determination theory offers a framework in which autonomy, competence and relationships matter to motivation and development. We use it as an intellectual resource without attributing measured effects on those dimensions to the website. (Ryan & Deci, 2000)

Our documentary choices seek consistency with that orientation. Users should be able to choose their entry point, understand what a summary offers and recover a source. They should be able to consult content without adopting an intellectual profile fixed by the site. Financial support for PSI should not determine votes. Discussion of priorities should not turn access to knowledge into a group’s privilege.

These commitments do not create equal access by themselves. Differences in language, time, education and connectivity persist. Recognising them guides development: clearer explanations, more accessible navigation and intelligible references are concrete improvements. PSI’s human ambition is also judged in those details.

8. Thema: a map for exploring knowledge

A map for entering knowledge

A library becomes usable when people can search it without already knowing a book’s exact title. The same difficulty arises online. A video title may favour a striking expression, an event or a speaker’s name; the subject interesting a reader may appear only within the content. For PSI, thematic classification addresses a primary need: make questions discoverable and then allow movement between them. It is the foundation of our documentary commitment.

Thema is a multilingual subject classification developed for publishing. PSI uses it as a common reference system and adapts it to a video catalogue. This professional origin matters: we do not present Thema as a theory of the brain or a definitive representation of the world. Its practical value comes from an organised, shared vocabulary whose categories can be expressed in several languages. French-speaking and English-speaking readers can thus use corresponding reference points to explore subjects. (EDItEUR, evolving standard)

Correspondence does not mean that every text is identical across languages. In PSI, a video’s thematic classification provides a common reference point; summaries and assessments produced in each language may differ. A difference in formulation may draw attention to a nuance or reveal a translation difficulty. We want readers to retain the option of returning to the video and its context rather than assuming that one language’s short version necessarily contains every nuance of the other.

Classification also helps distinguish questions a title may combine. A talk about batteries may concern chemistry, energy use, industry or transport policy. Classification makes those entry points discoverable and discussable. It does not turn the video into separate objects detached from their context: the subjects remain doors into the same content. Readers can begin with an initial interest and discover that a question in another domain changes what they thought they understood.

Why the reader’s perspective matters

A document’s subject cannot simply be reduced to a label placed on an object. Hjørland’s inquiry into the concept of subject in information science points to the importance of conceptions of knowledge and uses in determining subject matter. We draw a requirement from that discussion: explain what classification does and acknowledge that content can become relevant to different questions. (Hjørland, 1992)

Consider a lecture about urban mobility. One person looks for information about public transport; another wants to understand older people’s needs; a third examines investment decisions. They consult the same source but select different passages and follow different connections. A common classification facilitates their encounter with the document. It should not impose a single interpretation of its interest for them.

Recognising these perspectives does not mean that every classification is equally defensible. Some categories are plainly inappropriate because they do not correspond to the content. Others are defensible but incomplete. Others become useful when a secondary question matters to a reader. Discussion benefits from distinguishing an error, a legitimate selection and a change in perspective. We want Thema to provide a structure stable enough to guide, explicit enough to challenge and open enough for several routes.

This openness needs to remain compatible with search. If everyone invents an entirely private vocabulary, bringing content together becomes difficult. If a nomenclature claims to dictate everything a document can mean, it hides productive uses. PSI seeks a balance: a common documentary language for finding our way, and freedom to ask what that language cannot fully describe. Analogy has a place in this space. It can connect problems assigned to different domains without erasing the distinctions that made their initial reading possible.

Classifying subjects and evaluating sources

It helps to distinguish what is discussed, how it is approached and how strong the claims are. A Thema category primarily answers a documentary question: what is this about? PSI’s phylum distinction concerns the approach or mode of inquiry. Critical examination seeks to illuminate arguments and limitations. None of these reference points replaces the others.

Two videos about the same subject can use very different approaches. An experiment, an interview, a philosophical proposition and a speculative fiction can share vocabulary. That does not give them the same evidential status. Foresight can help formulate a question without establishing that its scenario will occur. A theoretical analysis can clarify a concept without measuring an effect. An observation can be well described without supporting a universal generalisation. Classification should help readers situate content before deciding what can reasonably be drawn from it.

This distinction becomes especially important when domains meet. Scientific terminology in a discourse does not guarantee the validity of the whole. An analogy between a brain and a computer network can open a discussion, but the relationships being compared need examination. Thema helps locate the relevant knowledge; it does not remove that task. We therefore want classification to invite scrutiny rather than function as a seal of authority.

From classification to new questions

A possible use of PSI begins with a narrow question: how to conserve a resource, understand a decision or explain a mechanism. Navigation may then reveal a neighbouring subject, a disagreement or an unexpected consequence. The encounter need not follow a route imposed by the website. It may arise from a change of term, category or source.

We do not claim that putting two links close together produces a discovery. A useful relationship requires activity from readers: formulate the comparison, identify differences and decide whether it deserves investigation. Our responsibility is to make that work practicable. Multilingual classification, summaries and access to source passages should reduce the cost of orientation so more attention can be devoted to reasoning itself.

Thema organises entry points; readers build their routes. This expresses PSI’s orientation. It does not mean that routes are free of influence: categories, summaries and presentation order can shape attention. Making them visible and discussable is part of editorial responsibility. Readers should be able to understand why content appears relevant and change course when another perspective becomes useful.

9. A word, several dimensions of meaning

PSI’s proposed framework

To make this plurality accessible, we suggest a teaching formulation: a word evokes a concept, sociocultural content and subjective personal content. These are not three sealed compartments. They interact in a communication situation. Above all, they offer three straightforward questions: which concept is being used, which associations are shared within a setting, and which particular experience does a reader bring to interpretation?

“Autonomy”, for example, can refer to a capacity to decide, a property of a device or an institutional situation. In different contexts it may evoke emancipation, isolation, economic independence or a demand for performance. For a particular person it may also recall an experience of trust or abandonment. Defining the term improves understanding but does not remove the other dimensions. Recognising them helps explain why people can appear to agree on a word while discussing different things.

This framework is PSI’s proposal. We do not attribute it as an exact formulation to Roland Barthes and Tzvetan Todorov. Connecting it to their work requires preserving their objects of study and distinctions. We seek an intellectual relationship and insight for documentary practice; we are not reconstructing a shared historical theory they supposedly published under the name “three-dimensionality of language”.

Barthes: how presentation shapes reading

In Rhétorique de l’image, Barthes examines relationships between an image, its meanings and the linguistic message that can direct its reading. His discussion of polysemy and anchorage illuminates a current difficulty: a title, caption or short formulation can select certain meanings while leaving others in the background. The text concerns advertising images; its use here is an analytical comparison, rather than an experimental measurement of PSI’s effects. (Barthes, 1964)

A video title, summary and category also help direct reading. This comparison suggests a design question: which context does our presentation reveal, and which context might it conceal? A summary centred on a technical benefit may leave little room for the social conditions of that benefit. A title centred on conflict may obscure points of agreement. Such choices do not become neutral because they are automated.

A short source presentation is valuable because it provides an initial foothold. Its limitation is selection. PSI must therefore preserve a route to the longer content, where that selection can be examined. Readers can recover a speaker’s wording, observe the transition before it and understand the qualifications accompanying it. This ability to return is part of our response to the plurality of meanings.

Todorov: description and interpretation

In Les catégories du récit littéraire, Todorov distinguishes, among other things, the issue of meaning in a description of a work from interpretations that can vary with readers and their contexts. This distinction helps hold together a search for structure and recognition of situated readings. It does not amount to declaring every interpretation arbitrary. (Todorov, 1966)

In Synecdoques (1970, especially pp. 26–27), Todorov distinguishes the rules of language from those of different discourses and considers a word’s meanings in relation to context. This text complements our discussion of interpretative frames: a formulation need not mean the same thing in scientific, poetic or everyday discourse. Connecting this to the sociocultural and personal dimensions of our framework is PSI’s interpretation; the article does not itself formulate these three dimensions. (Todorov, 1970)

For PSI, this reflection encourages separating what a source contains, how we organise it and what a person may draw from it. A document may contain a precise claim. A summary may choose to highlight it. A reader may see an application to a personal or collective problem the speaker did not address. That final operation can be productive, but it should not be reassigned to the source as if the source had demonstrated it.

This separation also helps address disagreement. Two readers may recognise the same passage yet differ over its importance. They may also differ over its meaning because they give different weights to context. We want the website to facilitate discussion in which the material at issue can be recovered, rather than an exchange confined to impressions detached from a document.

Contributions from linguistics and pragmatics

Langacker’s work on cognitive grammar places conceptualisation at the centre of linguistic analysis. Kecskes’s framework brings individual and social dimensions together in a sociocognitive approach to pragmatics. Leclercq discusses relationships between linguistic knowledge and language use by connecting theoretical traditions. These approaches do not all make the same claims; they help move beyond a picture of words carrying meanings entirely fixed outside any situation. (Langacker, 1986) (Kecskes, 2010) (Leclercq, 2024)

From this theoretical neighbourhood we retain a practical precaution. Before evaluating a disagreement, it may help to examine what people mean by a term and what the situation allows them to assume is shared. A technical meaning can be relatively stable in one community and unfamiliar elsewhere. An ordinary expression can acquire a particular function in a discipline. A translation can be grammatically correct while changing the associations available to a reader.

Lupyan’s label-feedback hypothesis explores interactions between language and cognition. Majid and colleagues’ cross-cultural research on vocabularies of perception also shows the value of comparing languages rather than taking one language as a universal measure. These studies do not justify claiming that language permanently confines its speakers’ thought. (Lupyan, 2012) (Majid et al., 2018)

PSI draws a multilingual commitment from this discussion: make reference points accessible and retain sources without presuming that translation abolishes every difference of perspective. Readers can compare, question a term and acknowledge that a useful category does not resolve every interpretative difficulty. For a project seeking to connect knowledge, the plurality of languages also offers opportunities to see familiar things differently.

Boundaries between categories

Rosch and Mervis’s work on family resemblances illuminates properties of cognitive categories. It does not directly describe a professional nomenclature such as Thema. Its relevance here is to recall that a category can be recognised and used without all its members sharing a single simple property. (Rosch & Mervis, 1975)

A reader may therefore encounter content on the boundary between subjects. The difficulty is not necessarily a fault in the document or the reader. It can reveal that a problem has several dimensions. The website should avoid confusing the convenience of a category with the existence of an indisputable natural boundary. It should also avoid multiplying comparisons without reasons: a disputed boundary calls for an argument, not only an intuition.

In this practice, personal experience is both a resource and a limitation. It allows recognition of a situation a general description leaves abstract. It can also make a comparison seem too obvious and conceal important differences. Discussion benefits when experience is presented as a situated perspective. PSI wishes to accommodate a plurality of questions while maintaining requirements for clarity, return to sources and distinction between interpretation and evidence.

10. What PSI offers readers

Begin with videos

PSI begins with videos because their content can be rich and difficult to navigate quickly. A title and duration are not enough to establish whether a talk addresses a question. An important idea may appear in the middle of a demonstration, after a long introduction or in a response to an objection. We want to make such entry points accessible while preserving the connection to the source.

Our documentary pipeline involves several operations: identify content, obtain or produce a transcript when conditions permit, analyse, summarise, classify, translate, build pages and prepare publication. These operations serve different functions. A transcript represents what was said, with possible errors. A summary selects and reformulates. Classification organises subjects. Critical assessment makes a judgement based on material that should remain discussable.

The result offered to readers should facilitate a decision: does this content deserve further attention for my current question? It should also help them choose a passage, concept or neighbouring subject. That decision belongs to the reader. We do not wish to reduce navigation to accepting opaque recommendations or present a score as a universal certification of truth.

The choices behind a summary

Summarising requires choosing what matters, bringing some propositions together and leaving others out. A summary can improve orientation while losing an important nuance. Maynez and colleagues’ research on the faithfulness of automatic summaries shows why fluent wording or a good similarity measure is not sufficient to establish fidelity to a source. Their study concerns specified models and data; it does not provide an error rate applicable to every current generation system. (Maynez et al., 2020)

For PSI, this difficulty creates a responsibility in presentation. A summary should remain identifiable as a reformulation, and its source should remain recoverable. When a reader encounters a decisive claim, verification should not stop at another automatic reformulation. The reader should be able to reach the content from which the claim came and examine its surroundings.

Reducing length is not the only measure of a good summary. We also need to consider the question it helps clarify, the distinctions it preserves and the limitations it makes visible. A text can be very short and unhelpful because it erases the difference between a hypothesis and a result. A slightly fuller text can help someone choose a better next step. Our ambition is economy of attention compatible with precision.

Recover main points and context

Creator-provided chapters, when available, and key moments identified in a source can facilitate access to chosen passages. A link to a moment in a video allows readers to recover an explanation without starting from the beginning. This is especially useful for examining a term, qualification or demonstration. It also lets them continue from the passage instead of remaining confined to an excerpt.

A key moment does not represent the whole video. Its value depends on the question. Going back a little before the passage or continuing after it is part of verification. We want these movements to be straightforward: recover the source, understand the selection and decide whether more context is needed. Chapter access supports chosen reading; it does not make longer context irrelevant.

Liu and colleagues’ work on using information in long contexts shows that an element’s position can affect the studied models’ performance. Information being present in a context therefore does not guarantee its reliable use. Again, the result belongs to a specified protocol. It motivates attention to traceability without predicting the quality of every PSI page. (Liu et al., 2024)

Checking AI-generated answers

Petroni and colleagues investigated language models’ capacity to retrieve factual knowledge in defined tasks. In another register, Bender and Koller discussed relationships between linguistic form, meaning and understanding. The first concerns observed performance; the second offers a theoretical argument. Bringing them together helps distinguish plausible production, correct test answers and guaranteed understanding of every situation. (Petroni et al., 2019) (Bender & Koller, 2020)

Readers should be able to ask several questions: does the wording match the source? Does the source provide what its conclusion requires? Is that conclusion still relevant to the situation I care about? These questions follow one another and cannot be resolved through a single indicator. A faithful summary of a questionable discourse remains a summary of a questionable discourse.

The lateral reading studied by Wineburg and McGrew offers another insight: evaluating a source can require leaving its immediate presentation and examining other information about it. We take this as an invitation to compare and verify; we do not assume that an external link automatically creates critical reading. (Wineburg & McGrew, 2019)

Our commitments and their evaluation

PSI’s commitment concerns the availability of reference points, summaries, main ideas and routes back to content. It also concerns clear categories, respect for languages and the possibility of choosing what to read. These commitments can guide improvements and their public discussion.

We do not describe personalised tutoring, perfect analogy detection or a route guaranteeing competence as established functions. We do not claim to have measured reduced anxiety among our users. An ambition can guide development without being announced as an achieved result. This distinction makes a commitment credible and leaves evaluation questions to be investigated.

The common principle is straightforward: make information discoverable, understanding practicable and human choice possible. Implementing it requires concrete details. A clear title, relevant category, readable summary, accessible source and honest distinction between evidential levels give readers material on which to judge PSI.

11. Learning with AI tools

Access, understanding and retention

Finding information quickly, understanding an explanation and being able to use it later are connected but different objectives. The website may improve access without readers retaining everything. A summary may help decide whether to investigate a source without constituting durable learning on its own. PSI’s vision should recognise these differences rather than confuse ease of consultation with automatic development of capacities.

Dunlosky and colleagues’ review compares learning techniques and shows why usefulness depends on conditions of use. Research on retrieval practice, spacing and interleaving provides findings that invite us to move beyond simple rereading. Applying those findings to freely chosen use of PSI remains a question to investigate. (Dunlosky et al., 2013) (Roediger & Karpicke, 2006) (Cepeda et al., 2006) (Brunmair & Richter, 2019)

We can suggest a voluntary practice: after a summary, reformulate its idea without looking at it; after two sources, specify what they share and where they differ; after an analogy, seek a situation where it would not work. These suggestions exercise active reading. They are not an experimental programme already validated on the website. Above all, they bring out a question that a fluent explanation can hide: can I reconstruct what I believe I have understood?

A reading route need not become a test of memory at every step. Some information is better kept as a recoverable reference. Other relationships are worth understanding in enough depth to recognise them elsewhere. The point is to choose deliberately rather than assume that everything read has become available for future reasoning. PSI can help make the distinction visible, while readers decide how deeply they want to work with particular material.

Do something with an explanation

Chi and Wylie’s ICAP framework distinguishes forms of engagement in learning, including the cognitive activity situations make possible. We relate it to our documentary ambition: a user may consult a summary, produce a reformulation or discuss an interpretation with another person. These activities are not interchangeable. (Chi & Wylie, 2014)

Constructive reading need not resemble a school exercise. Someone can list questions, identify a contradiction or find an example in their own experience. They can also recognise that they do not yet have what is needed to conclude. This matters: identifying a precise gap helps choose the next source for better reasons.

The website should not impose the same intensity on every consultation. Wanting to recover practical information quickly is legitimate. Spending a long time on a distinction is legitimate too. Our commitment concerns the possibility of these uses, rather than an obligation to turn every visit into training. Free access and freedom of navigation should leave room for such variety.

Assistance can help or foster dependence

Recent findings about AI in learning encourage examination of the arrangement’s design. In a mathematics experiment studied by Bastani and colleagues, effects of AI access differed with the form of assistance and whether assessment occurred with or without it. In another context, Kestin and colleagues reported benefits from tutoring designed for learning compared with an active teaching situation. These studies concern different participants and activities. They prevent treating every form of assistance as a uniform benefit or threat. (Bastani et al., 2025) (Kestin et al., 2025)

For PSI, the issue is to preserve room for readers’ questions. Assistance that answers before someone has formulated what they seek can create an impression of immediate understanding. Assistance making distinctions and sources visible can facilitate investigation. The website is not a personalised tutor; this comparison clarifies a design orientation: provide footholds for thinking, rather than only formulations to accept.

That orientation remains revisable. If a presentation encourages excessive confidence, it should be possible to change it. If a summary is too abstract to support a return to the source, the clarity of the link needs improvement. A useful commitment offers criteria for discussing the arrangement. It does not ask users to believe that its mere existence will benefit them.

Limits of transfer

Practising one task does not necessarily improve every other task. Simons and colleagues’ review of cognitive training and Melby-Lervåg and colleagues’ work on working-memory training encourage careful examination of claims about far transfer. This caution concerns the programmes they study; it does not condemn learning or curiosity. (Simons et al., 2016) (Melby-Lervåg et al., 2016)

We therefore do not promise that thematic browsing produces a general increase in intelligence. Our ambition is more precise: make knowledge accessible, encourage connections and help examine their relevance. Successful transfer will depend on the problem, available foundations, practice and context. An interesting analogy may start worthwhile work even when it does not immediately provide a solution.

Measures of progress should match the objective. Discovering a field’s vocabulary is different from acquiring the ability to solve a problem within it. Recognising a relationship is different from applying it under constraints. Precision protects readers from overly broad promises and helps PSI choose concrete development objectives.

12. Memory and judgement in a digital world

What we entrust to external supports

A library, notebook or digital tool can allow us to perform tasks difficult to accomplish without support. Clark and Chalmers proposed a philosophical account of the extended mind. Risko and Gilbert reviewed cognitive offloading and the ways actions and external supports alter a task’s demands. These texts illuminate different questions; they do not show that every tool increases every capacity. (Clark & Chalmers, 1998) (Risko & Gilbert, 2016)

The problem is therefore not to choose between wholly solitary thinking and total delegation. It is to decide what to entrust to a support and what we want to reconstruct ourselves. An address, reference or passage can be kept for recovery. A relationship essential to reasoning may need sufficiently deep understanding to be used in a new situation. A link’s availability does not always replace such appropriation.

Sparrow and colleagues’ experiments on memory in a context of expected access to information illuminate this distribution between content and where to find it. They do not establish that the Internet destroys memory. They invite examination of what we learn when we know a support will remain available. (Sparrow et al., 2011)

For PSI, direct links are valuable because they reduce verification costs. They should not become an excuse for an incomprehensible summary. Readers should be able to grasp the main relationship and decide which information to keep within reach and which to investigate. Documentary support then becomes a chosen resource within broader intellectual work.

Check one’s degree of understanding

Metacognition concerns, among other things, how people assess their own cognitive activity. Fleming and Lau discuss methods for examining the accuracy of such assessments. Their work recalls the need to distinguish confidence from accuracy. Someone may feel certain about a fragile interpretation or hesitate despite good understanding. (Fleming & Lau, 2014)

PSI does not automatically measure users’ metacognition. We can nonetheless design explanations that make it harder to confuse fluency with evidence. Indicating a limitation, separating a hypothesis from a finding and offering a route to the source make examination possible. They do not guarantee that readers will use those opportunities, but avoid presenting confidence as the only sign of quality.

Scientific vocabulary also deserves attention. Weisberg and colleagues’ research on the allure of neuroscientific explanations, and Bennett and McLaughlin’s subsequent meta-analysis, examine effects of neuroscientific elements on judgements under particular conditions. A brain in an illustration or specialised terminology does not remove the need to examine an argument. (Weisberg et al., 2008) (Bennett & McLaughlin, 2024)

We want to apply this principle to our own text. Neuroscience references provide a research context; they are not decoration designed to make a vision unquestionable. Readers should be able to understand what a study observed and why we relate it to a PSI question. An explanation made more precise by science is preferable to one merely seeking to appear scientific.

Keep responsibility for one’s questions

Tool use can change habits of judgement. Vallor examines questions of moral deskilling and development in relation to automation. We draw an inquiry about practice from this discussion: which capacities does a device invite us to exercise, and which might it leave without occasions for use? This is a philosophical argument, not a measurement of PSI’s effects. (Vallor, 2015)

The answer cannot be to refuse every form of assistance. It can be to preserve occasions for asking a question, examining an objection and recognising responsibility. Someone can use a machine to explore arguments while retaining the decision about what to pursue. They can also compare an answer with experience or a source unavailable to the model. These practices need time, knowledge and favourable conditions; their value does not depend on universal human superiority.

This distribution of work clarifies our slogan. Connecting knowledge does not ask readers to memorise everything. Bringing ideas to life does not promise permanent originality. It means making possible activities in which information meets a question, experience and judgement. Tools contribute to that encounter; they should not conceal its choices.

13. Understanding the concerns of our time

What gives rise to concern

Anxiety can be fed by personal difficulties, collective transformations and uncertainty about the future. PSI’s vision occupies that last space without claiming to exhaust its complexity. Climate change, resource use and AI development are not merely wording problems. They concern environments, institutions and living conditions. Understanding them is insufficient to solve them, but lack of understanding can make useful action or discussion harder to choose.

Hickman and colleagues’ survey of young people in several countries documents concerns about climate and government responses. Pihkala offers analyses of eco-anxiety and ecological emotions, including in education. These approaches recall that concern about a collective threat should not immediately be reduced to individual failure. (Hickman et al., 2021) (Pihkala, 2020) (Pihkala, 2020)

Our purpose is not clinical. PSI is not a diagnostic or treatment arrangement. Supporting the power to understand and act is a documentary and civic orientation. We want people to identify questions, distinguish knowledge from scenarios and find material for participation in discussion. We do not promise that consulting the website will remove a difficult emotion.

Distinguish emotions and their contexts

Climate-anxiety research uses different instruments and populations. Mouguiama-Daouda and colleagues’ French validation of a scale, Wullenkord and colleagues’ findings in a German sample, and Stanley and colleagues’ work on several climate emotions allow examination of this diversity. Their conclusions should not become a single portrait of every user. (Mouguiama-Daouda et al., 2022) (Wullenkord et al., 2021) (Stanley et al., 2021)

Concern can direct attention to a problem; it can also make moving into activity difficult. Anger, sadness and hope are not necessarily interchangeable. Observed relationships between emotions and action depend on context and do not supply an individual recipe. The website should avoid a uniform demand to feel better, as if emotional state were the primary difficulty in a collective crisis.

Instead, we seek to make intellectual steps more accessible. An overly broad question can be decomposed. A scenario can be compared with another. A claim can recover its status: observation, hypothesis, extrapolation or political choice. Such organisation may help someone decide what to examine. Its emotional effects remain to be established and may vary between people and situations.

Resources, climate and interdependence

Non-renewable resources also raise questions about stocks, extraction rates and technological systems’ dependencies. Their finitude does not by itself provide a timetable for the end of civilisations. It calls for examining uses, possible substitutions, their limitations and collective choices. (Haberl et al., 2020)

Richardson and colleagues’ assessment of planetary boundaries describes an Earth-system state using a particular framework at a particular date. Haberl and colleagues’ review of decoupling examines a large body of work on relationships between economic activity, resource use and environmental pressures. These sources give reasons to take material limits seriously; they do not establish one inevitable date for the end of civilisations. (Richardson et al., 2023) (Haberl et al., 2020)

For PSI, the question needs to connect physical knowledge, uses and collective decisions. A technology can change efficiency without determining total demand on its own. A resource can be available somewhere and difficult to access for other reasons. A strategy can move a pressure rather than remove it. Systemic thinking invites examination of those relationships, their scales and their timescales.

The documentary purpose is not to make everything equally alarming. It is to enable proportionate reading: which mechanism is described, which data support it, which uncertainties remain and which decision is being discussed? Precision does not necessarily remove the gravity of a question. It can prevent distinct concerns from merging into an undifferentiated impression of catastrophe.

Agency extends beyond the individual

Ostrom’s work on socioecological systems emphasises examination of interactions among resources, actors, governance and situations. It does not offer one solution for every problem. It illuminates the importance of collective arrangements and concrete conditions under which action becomes possible. (Ostrom, 2009)

This corresponds to a PSI commitment: do not make a reader responsible for a whole system alone. Understanding can help formulate a demand, participate in a decision or choose a practice. It does not replace necessary material means, rights or institutions. The website’s commitment should respect that difference rather than add guilt to concern.

Ojala discusses emotions and anticipatory competences in education. This intellectual neighbourhood encourages us to seek hope compatible with recognising problems. We neither want to minimise threats for quick reassurance nor present powerlessness as an obligatory conclusion. It remains possible to choose what deserves investigation, recognise a contribution and construct a shared question. (Ojala, 2017)

Everyone’s place in relation to AI

Performance demonstrations can suggest that a person must constantly justify existence through a task machines cannot yet perform. Such justification becomes fragile as technical capacities evolve. PSI proposes another orientation: exercise human capacities because they allow understanding, participation and creation, without making dignity depend on the duration of a competitive advantage.

Analogy, general knowledge and systemic understanding matter in this orientation. They can help formulate a new question or see a previously secondary consequence. They are not defensive boundaries we can declare permanently impassable by machines. We want people to maintain an active place in defining problems and making decisions affecting them.

Tools can strengthen that place, but social conditions also matter. People need time, access and interlocutors. Changes in work can require collective responses. PSI’s documentary resources contribute to one part of that wider picture. Their usefulness does not require solving everything that makes the future difficult.

14. Challenges to our vision

AI can also reason by analogy

Mitchell’s review of abstraction and analogy in AI describes difficulties and methodological questions in the state of knowledge of 2021. Lake and colleagues’ proposals about machines learning more like people identify research directions. These texts help understand problems; their dates prevent treating them as definitive descriptions of future capabilities. (Mitchell, 2021) (Lake et al., 2017)

More recent findings already change some questions. Lake and Baroni report systematic generalisation in a compositional-learning arrangement. Webb and colleagues examine performance on counterfactual analogy tasks. Binz and Schulz use cognitive-psychology tasks to study GPT-3 and identify a performance profile requiring detailed description. These studies are not universal intelligence tests; they give reasons not to confuse a particular failure with an impossibility in principle. (Lake & Baroni, 2023) (Webb et al., 2025) (Binz & Schulz, 2023)

If models become able to make many cross-domain connections, PSI’s vision can remain relevant. It will then need to emphasise conditions for informed human use of those connections: which questions do we want to ask, how do we evaluate a relationship and which effects do we want to examine? Our project does not rest on defending a permanent machine deficit. It rests on maintaining people’s ability to understand and participate where technical capacities change.

Complementarity is not automatic

Vaccaro and colleagues’ meta-analysis examines human–AI combinations across many studies. It shows, among other things, that combinations do not systematically outperform the better partner and that tasks and conditions matter. The useful conclusion for PSI is a requirement to investigate: offering assistance is insufficient to establish synergy. (Vaccaro et al., 2024)

Someone may accept an incorrect proposal because it appears consistent with expectations. A system may offer an original comparison poorly suited to the question. Collaboration requires ways to recognise disagreement and verify decisive material. Sources, criteria and possibilities for changing direction then become components of the arrangement rather than decorations added after an answer.

This objection leads to a more precise commitment. PSI should make the material needed to discuss its productions visible. Their benefit cannot be concluded from the number of items processed alone. Errors, clarity and links’ ability to support verification require examination. Quantity is one dimension of a catalogue; it does not summarise documentary value.

The role of the body, experience and memory

Barsalou’s grounded-cognition review recalls a research tradition examining relationships among cognition, perception, action and situation. Schacter and colleagues discuss relationships between memory and imagining the future. These frameworks provide reasons not to portray the mind as a collection of definitions available independently of experience. (Barsalou, 2008) (Schacter et al., 2007)

This qualifies what a documentary website can do. A video may introduce a practice; it does not replace every experience needed to learn it. A summary can name a difficulty without automatically providing the capacity to resolve it. Our knowledge foundation should remain connected to uses, experiences and exchanges. The subjective personal content in our meaning framework is not simply a defect to remove from thinking.

Bartoli and colleagues’ work on electrophysiological dynamics and a network’s causal role in creative tasks offers a recent insight obtained in a specified clinical setting. In a methodological text, Poldrack recalls precautions needed when inferring cognitive processes from neuroimaging data. These references encourage presenting results within their scope and avoiding total explanations of creativity through a single region or measure. (Bartoli et al., 2024) (Poldrack, 2006)

PSI’s vision therefore remains open to several descriptions of human activity. Our general vocabulary—understand, reason, imagine—formulates an accessible ambition. It does not replace research distinctions used to construct questions. Such descriptive modesty allows new findings to be welcomed without defending a slogan as if it were an experimental theory.

15. An example of reading with PSI

Imagine someone seeking to understand robots in a public service. They begin with a demonstration video. Its summary helps identify the performed task, and Thema situates the main subjects. They recover a passage explaining the operation. This first step distinguishes what is shown from what is merely announced.

A second question appears: why use a robot for this task? The reader may look for content about work organisation, service constraints or the machine’s material requirements. A technical answer may illuminate execution speed. Another source may examine relationships with people affected. Categories offer several entries into a problem no longer reducible to the demonstration’s visible performance.

The reader then formulates an analogy with another arrangement they know. They may recognise a relationship between task standardisation and loss of local information. The hypothesis can be interesting. It needs precision: which information may be lost, in what situation and through which mechanism? Returning to sources may show that the analogy fits one aspect and misleads about another.

They may finally investigate a remaining question or recognise that a field observation is missing. The route need not produce a spectacular conclusion to be useful. It has transformed a general impression into examinable questions. The reader knows more clearly what is understood, what is assumed and what still needs finding.

This is an original PSI illustration, not a report of a measured website effect. It shows a possible use of our commitment: classify to orient, summarise to choose, recover relevant moments to verify, compare to reason and retain the right not to conclude prematurely. It also recalls that a route can leave the website for sources, exchanges and experiences.

16. How this synthesis was prepared

How the sources were selected

This first draft brings together 66 references around analogy and generalisation, creativity, learning, cognition, interpretation, classification, autonomy and ecological concerns. Selection combines foundational texts with more recent publications. It includes work supporting some PSI orientations and work limiting promises we might be tempted to make.

This is an editorial narrative review, not a systematic review. We have not established an exhaustive survey of all databases, a preregistered inclusion protocol or a new meta-analysis. Cross-domain connections belong to PSI’s argument. Cited meta-analyses remain their authors’ work and retain their own criteria. Inclusion does not mean an author endorses our project.

Bibliographic searches prioritised publication pages, institutional repositories, author-provided texts and public scientific archives. Depending on the source, consultation included records, abstracts and passages of full texts; it was not an exhaustive reading of every article. Links allow readers to recover references and continue examination. Some historical publications and monographs require extended reading to assess fully the framework with which we connect them.

Public access has different forms. An abstract gives access to a question and announced main findings; it does not replace complete methods. An author version can differ in layout from the published article. An open-access publication makes its text available without subscription. Where identified, the bibliography offers an additional public route alongside the publication link. No paywall is circumvented.

What peer review means

Peer review provides scientific and editorial scrutiny but does not make a result infallible. A study may have population, measurement or context limitations. Results can receive new interpretations or be reassessed. A theoretical review has a different status from an experiment; a survey does not necessarily establish causality; a professional standard is not empirical evidence of cognitive benefit.

We therefore distinguish source types. Empirical articles, reviews and conference publications belong to their scientific venues. Barthes and Todorov are used as historical contributions to the analysis of meaning and interpretation. The linguistic monograph presents a theoretical framework. Thema is a professional reference system. All can inform reflection, but their uses should not be conflated.

French references are particularly important for the history of our interpretative framework. Research involving French-speaking and German-speaking populations also broadens empirical contexts. Most selected recent publications are written in English. This distribution reflects the selected corpus rather than a ranking of languages’ value. PSI’s French and English versions retain the same bibliographic identifiers and original titles.

Originality, attribution and revision

This page is originally written. Reformulations of findings and frameworks carry references. Reading examples, commitments and connections proposed by PSI are identified as such. We avoid reproducing an author’s distinctive formulation without identification. Any direct quotation must be delimited, attributed and linked to its source; a translated quotation must be identified as a translation.

The requirement applies across website languages. Translating a reformulation does not permit removing its reference. A final bibliography does not remove the need to connect a claim to the source supporting it. Conversely, a citation should not imply an author stated our commitment. Separating observation, interpretation and promise is also an attribution rule.

This first draft can be revised following closer reading of decisive passages, discussion of selection and examination of objections. Knowledge about AI changes rapidly. Revision should identify what has changed and what that changes in the argument. We want to preserve a recognisable vision while accepting that its reasons and formulations can gain precision.

17. PSI’s commitments

PSI’s first responsibility is documentary: offer content that can be situated and recovered. The second is editorial: distinguish sources from reformulations, make limitations intelligible and cite work used. The third concerns readers’ freedom: enable several entry points, a return to context and a change of question.

Our access ambition should remain consistent with the website’s operation: free access, no advertising and no compulsory account for consultation. PSI excludes building personal user profiles and does not produce statistics about their behaviour. This orientation does not remove data needed for a transaction when someone chooses to donate. It means reading and navigation should not build an individual dossier of preferences.

Future evaluation must respect those commitments. We can examine documentary outputs, compare formulations and discuss reported errors. Studies involving people, if considered one day, would need their own explicit, voluntary framework. They should not be presumed from mere website use. At this stage no cognitive or emotional effect of PSI is presented as established.

Finally, we want the commitment to remain human in scale. Readers do not have to master every field to contribute. They can offer a question, experience, objection or particular attention to a relationship. A broad knowledge foundation does not mean knowing everything; it means having enough reference points to recognise what is missing, search discerningly and connect without confusing.

Connect knowledge. Bring ideas to life. Our place in a world of powerful machines cannot be reduced to the last task they fail to perform. It is also built through the questions we choose, relationships we explore and responsibilities we accept together. PSI wants to provide more footholds for that activity: a common map through Thema, summaries for orientation, sources for verification and freedom to imagine what their encounter makes possible.