The Four Capacities of Cognitive Autonomy: Induction, Generalization, Judgment and Self-awareness

Cognitive autonomy requires that people, while making use of external knowledge and intelligent systems, remain in charge of their own understanding and action. Yet merely asking students to think independently and judge rationally is not enough to produce a clear educational plan. Teachers need to know what kind of learning to organize, students need to know what they should practise, and the course needs outcomes that can be observed and assessed.

To this end the Redbook proposes the four-capacity model of induction, generalization, judgment and self-awareness, turning the educational goal of cognitive autonomy into a set of interconnected capacity requirements.

1. Why Understand Cognitive Autonomy through Four Capacities

Human cognitive activity can be understood in terms of several interconnected aspects: induction, generalization, judgment and self-awareness.

First, people need to form understanding from the materials and experiences they encounter. Seeing a number of phenomena, finding the relationships among them, establishing categories and proposing explanations: this is the work of induction.

The understanding thus formed must also be applied to new objects and problems. Does a method that worked in the past still apply today? Can a regularity found in certain samples be extended to other objects? Can knowledge already learned help one solve an unfamiliar problem? All of these involve generalization.

When understanding enters real life, people also have to make choices. Which explanation to believe, which plan to adopt, which tools to use, what risks to bear: all of these call for judgment.

Any of these processes may go astray at any time. People need to notice whether their understanding is sufficient, whether their confidence is warranted, and whether their judgment is being swayed by irrelevant factors, and then adjust in light of new evidence. This is the work of self-awareness.

The four capacities thus concern, respectively, the formation, application, selection and correction of understanding. In real cognitive activity they work together, enabling people to keep managing their own process of understanding.

It should be stressed that what is discussed here are human capacities, and they have nothing to do with AI as such. Whether or not AI exists, these are basic human capacities that need to be cultivated. It is only that the rapid development of AI has made the cognitive environment more complex, so that these capacities become ever more important and form the foundation of cognitive autonomy.

The AI general education course can cultivate these four capacities because AI is the science of simulating human intelligent behaviour: learning, generalization and judgment in AI models are precisely computational expressions of these human capacities, and so provide a point of reference for understanding human cognition.

Through such observation, comparison and analysis, learners not only come to know the characteristics of their own cognition but also understand the basic principles of AI, and thereby arrive at a clear-sighted sense of where they stand in their environment and of their own cognitive state, that is, "self-awareness".

This is the basic logic by which the AI general education course is able to cultivate cognitive autonomy.

It should be made clear that the four-capacity model is an educational model designed for curriculum development and learning assessment. Capacities such as memory, attention, deductive reasoning and creativity also take part in cognitive activity. The Redbook chose these four in order to make clear the main training content of AI general education; this does not mean that human cognition can, psychologically, be divided into only four parts.

2. Induction: Forming Evidence-based Understanding from Limited Experience

People rarely form an understanding only after grasping all the facts. We always come to understand the world around us gradually, within limited time and on the basis of limited material.

A student compares the results of several experiments and discovers the relationship between variables; reads several articles with differing views and forms a preliminary explanation of a question; observes the common features of several animals and establishes a classification. All of these processes involve induction.

Induction enables people to form, from concrete material, an understanding that goes beyond the material itself. It is an important way of acquiring knowledge, and also an important basis for discovering problems and generating new ideas. Faced with large amounts of scattered information, only by recognizing relationships and forming explanations can the material be turned into one's own understanding.

But induction is also inherently uncertain. A few successful cases do not necessarily represent the general effectiveness of a method; a phenomenon that recurs does not necessarily have a stable causal relationship behind it; the people and things one often encounters do not necessarily represent a wider population.

In an intelligent environment this problem becomes more complex. Platforms can repeatedly recommend similar content according to personal preferences, making a certain phenomenon seem especially widespread; several AI answers may rely on similar sources, so that a claim that keeps reappearing looks as if it has been confirmed from many sides. The number of times a piece of information appears is not the same as the number of independent pieces of evidence.

Inductive capacity therefore has two mutually supporting aspects: actively forming understanding, and checking the grounds of that understanding.

A person with inductive capacity can find patterns in material, and can also explain which materials they have looked at and which cases they have missed, how strong a conclusion the existing evidence can support, and what further testing is needed. They can put forward their own explanation while leaving room for counterexamples and revision.

This also shows that induction cannot be reduced to organizing information. Compressing several articles into a one-paragraph summary may merely rearrange existing statements. Forming one's own understanding further requires comparing the relationships among different materials, identifying disagreements, proposing explanations, and explaining why those explanations hold.

For cognitive autonomy, the significance of induction is that it lets people take part in, and monitor, the process by which conclusions are formed. Even when using AI to organize material, find leads and propose hypotheses, learners can still check where a conclusion comes from and decide what they have reason to accept.

3. Generalization: Applying What Is Learned to New Situations while Grasping Its Boundaries

Induction helps people form understanding; generalization allows that understanding to be applied to new problems.

A student who has mastered a mathematical method can keep using it after the form of a problem changes; a person who understands sampling bias can use that knowledge to analyse social surveys and online polls; a team that has summed up its project experience can use it to handle new tasks under different conditions. All of these reflect the capacity for generalization.

Generalization gives learning lasting value. People cannot prepare answers in advance for every problem they will meet in the future; they need to distil relationships and methods from what they already know, find structures in new situations that they can draw on, and then make the necessary adjustments.

But the application of knowledge is always subject to conditions.

A learning method that works in one class may depend on the students' prior knowledge, the teacher's organization and the time invested; a model that performs well in testing may depend on particular inputs and tasks; a management practice that succeeded may also have had to do with the resources and institutions of the time. When conditions change, the validity of the original conclusion needs to be tested again.

The capacity for generalization therefore includes both applying knowledge and recognizing its boundaries. The former enables people to tackle new problems actively; the latter keeps that application reasonable.

In practice, learners need to compare the old and new situations: which conditions are the same, which have changed, and which differences will affect the outcome. They also need to try applying what they know, observe the feedback, and adjust their original method accordingly. Being able to explain when knowledge applies and how to adjust it is likewise a sign of understanding that knowledge in depth.

Induction and generalization can be briefly distinguished: induction concerns how understanding is formed, while generalization concerns how existing understanding enters new situations. The two are also often intertwined. Successes and failures in new situations in turn help people revise their original understanding.

For cognitive autonomy, generalization enables people to adopt external advice conditionally. Faced with expert experience, successful cases or solutions offered by AI, learners can judge their applicability in light of their own circumstances, neither copying them mechanically nor abandoning them as a reference simply because differences exist.

This capacity also needs to be cultivated deliberately. Knowing that a model may fail when its data change does not mean that a student will naturally apply this insight to analysing real life. The course needs to arrange comparisons across different situations, helping students apply the ways of understanding formed in learning about AI to a wider range of problems.

4. Judgment: Making Reasoned Choices under Uncertainty

Once people have formed an understanding of a problem, they still need to decide how to act. Understanding provides the grounds; judgment links those grounds to goals and consequences and produces a concrete choice.

Judgment includes telling fact from falsehood, but its scope goes far beyond that. Whether information is reliable, whether a tool is suitable, whether a goal is reasonable, whether a risk is bearable, how the interests of different people can be balanced, and who is responsible for a decision: all of these are questions judgment has to deal with.

Such questions often cannot be settled by a single indicator.

For example, an AI summary may be enough to help a reader grasp the gist of an article, but when its content is to be written into a public report, the original text, data and citations need to be checked further. The same output must undergo different degrees of scrutiny for different uses, because the purpose of use and the consequences of error have changed.

Judgment also often takes place when information is incomplete. People cannot always wait until the evidence is complete before deciding; they need to choose, on the basis of the grounds at hand and the requirements of the task, whether to accept, modify, defer or reject a proposal. The capacity for judgment shows itself in being able to explain the reasons for these choices, and in knowing what new information might lead one to change one's decision.

AI has expanded the analytical power people can call on, and has also increased the complexity of choice. A task can be done by a person alone, handed to AI, or carried out through various forms of collaboration. Learners need to assign a degree of trust to suggestions, conclusions and options, and make decisions in light of the actual situation.

The capacity for judgment itself plays an active, creative role. People can use AI to find more material, compare alternative explanations and generate multiple solutions, and then choose on the basis of goals and evidence. Remaining in charge of judgment does not require performing every operational step oneself.

For cognitive autonomy, judgment enables people to organize external capacities into their own cognitive process, while understanding the reasons for and consequences of using them.

5. Self-awareness: Seeing One's Own Cognitive State and Adjusting in Time

Induction, generalization and judgment can all go astray. And when people go astray, they often do not know that they have.

Following a fluent explanation can lead people to overestimate their own understanding; seeing a precise score can lead them to overestimate the reliability of an evaluation; receiving an answer that agrees with their own position can lead them to lower their demands for evidence.

What self-awareness is concerned with is precisely whether people can notice such situations and make adjustments.

The first is noticing the degree of one's understanding. A person may be able to restate a conclusion and still be unable to explain its grounds. Self-awareness requires learners to discover the gaps in their knowledge, knowing which parts they have understood and which still need to be learned.

The second is noticing changes in trust. A person's confidence may come from new evidence, or from the way something is expressed, from authority, from group opinion or from personal emotion. Self-awareness enables people to check whether their trust is adequately grounded, and to raise or lower their confidence when necessary.

The third is noticing the role external systems play in one's understanding. When using AI to complete a task, learners need to be clear about which materials were provided by the system, which explanations were inspired by it, which judgments have been verified by themselves, and which important steps have not yet been fully understood.

This awareness should also include recognizing AI's actual contribution. A system may help people discover omissions, break out of their existing lines of thought, and encounter different viewpoints. Being able to recognize these gains and improve one's way of learning accordingly is also part of self-awareness.

Finally, awareness needs to be carried through into adjustment. Gathering more material after finding the evidence insufficient, changing plans after finding a method unsuitable, re-comparing explanations after finding oneself influenced by preconceptions: only then does self-awareness enter into the actual management of cognition.

In using the term "self-awareness" (自醒), the Redbook emphasizes an active and timely noticing and correcting. It includes reflection after the fact, but it also happens in the very moment of understanding and acting. What it mainly addresses is the state of one's understanding, trust and judgment; it is not simply self-criticism.

Self-awareness runs through the other three capacities. Checking whether material is sufficient when inducing, whether conditions match when generalizing, and whether reasons are sound when judging all require the participation of self-awareness. Cognitive autonomy thereby gains the possibility of continual correction.

6. How the Four Capacities Together Constitute Cognitive Autonomy

To illustrate the relationship among the four capacities, imagine a concrete task: a school plans to use an AI essay-scoring system and needs to decide what work it is suited to do.

First, teachers and students, working from clear assessment criteria, compare the AI system's scores and feedback on a set of essays with the results of human marking. They may find that the system identifies some language problems fairly consistently but handles certain features of expression poorly. These observations then need to be tested against more material. This process mainly reflects induction.

Next, the system is applied to essays from different grades, in different genres and different writing styles, to see whether the original understanding still holds. If the system performs differently under new conditions, the reasons for the change need to be analysed and its scope of application redefined. This process mainly reflects generalization.

On this basis, the school needs to decide how to use the system. Using it for feedback on everyday practice and using it for formal grading lead to different consequences and requirements. Whether human review is needed, whether students are told the basis of assessment, and how disputes are handled all need clear arrangements. This process mainly reflects judgment.

Self-awareness runs through the whole task. Teachers and students need to watch whether their own assessments are being pulled along by the AI scores, and whether they are trusting the system's conclusions too much because it offers detailed explanations; they also need to recognize whether the system has helped them notice problems they had previously overlooked.

The role of the four capacities in this task can be summarized as follows.

These activities should be practised repeatedly. New essays, new conditions of use and real feedback may all prompt participants to form new explanations, test boundaries again and adjust their plans. Self-awareness can also step in at any point.

The four capacities are therefore linked to one another, but they are not a procedure that can only be carried out in a fixed order. Together they form a cognitive process with feedback, enabling people to keep command of their own understanding and action under ever-changing conditions.

7. What Is Distinctive about Training the Four Capacities in the AI General Education Course

It is worth noting that mathematics, Chinese, science, history and other subjects all train these capacities to some degree, and some of that training is already quite systematic.

However, the training offered by the AI general education course is more systematic and more closely fitted to the task.

First, training in these four capacities has to address the changes in the cognitive environment brought about by AI. The specialized knowledge provided by the AI general education course allows existing cognitive capacities to work effectively in processes of understanding in which AI takes part. This knowledge cannot be replaced by the general thinking training of other subjects alone.

More crucially, AI offers a prototypical way of training through which intelligent processes can be analysed directly. AI uses models to simulate human intelligent behaviour, and shares common ground with humans in induction, generalization and reasoning, thereby providing a reference for analysing, understanding and training the corresponding human capacities.

Conclusion

Induction, generalization, judgment and self-awareness give cognitive autonomy a structure of capacities that can be understood and trained. Induction supports the formation of understanding, generalization its application, judgment reasoned choice, and self-awareness continual noticing and correction.

These capacities have nothing to do with AI as such; they are basic human capacities, and it is only the arrival of the intelligent age that has made them more important.

The AI general education course offers, in its own distinctive way, a method of training these capacities: by mirroring induction, generalization and judgment as capacities of models, it lets learners understand themselves while understanding AI, and so achieve "self-awareness" and "autonomy".