AI General Education: A New Form of Education

In the earlier commentaries we discussed the formation of the intelligent society and why people need to keep their cognitive autonomy within it. Cognitive autonomy, however, does not arise by itself; it needs the active involvement of education. What kind of education, then, can take on such a mission for our time? The Redbook's answer is AI general education. Here is the Redbook's definition:

Artificial Intelligence General Education is a form of education created through the integration of general education and AI education. It takes the knowledge, methods, systems, and social operation of artificial intelligence as its content and assumes general education's mission of cultivating the capacity for cognitive autonomy in an intelligent society.

This definition states where AI general education comes from, what it contains and what it aims at. To understand it in depth, we need to explain why "artificial intelligence" and "general education" are converging, and why this new form of education is able to fulfil the mission of our time, "cultivating the capacity for cognitive autonomy".

1. Two Kinds of Education Moving Toward Each Other

1. The mission of general education in our time

General education is concerned with the shared cognitive foundation that people should have beyond their professional specializations. People may work in different occupations and have different expertise, yet they still need to understand the world they live in together, and to be able to communicate, judge and take part in public affairs. Scientific, humanistic and social knowledge all bear closely on this educational goal. Learning this knowledge gives people basic ways of understanding the world, and gives people from different professional backgrounds the conditions for understanding one another.

As the intelligent society gradually takes shape, general education has acquired a new mission: when intelligent systems can offer explanations, organize reasons and recommend actions, general education must help learners keep an independent cognitive capacity while relying on those systems, that is, cognitive autonomy. How to fulfil this mission is the key question facing general education.

2. The trend toward general education within AI education

Seen from the other direction, education in the intelligent society is repositioning its own value. As technology develops, knowledge becomes ever easier to obtain, and with the help of AI the threshold for technical work keeps falling. One consequence is that traditional "specialized education", built mainly on knowledge learning and skills training, declines in function, while "general education", which attends to transferable abilities, becomes the main body. We call this trend "the generalization of education".

AI education must respond to this change too. It should of course teach the specialized knowledge of AI, but what matters more and more is how it finds its own role in public education and cultivates in citizens basic, transferable mental abilities of a certain kind. AI education is therefore also looking for its own place as general education.

3. The convergence of disciplines

General education needs a new approach that can cultivate the capacity for cognitive autonomy, and AI education needs to establish its own place as general education. The two thus move toward each other and gradually form a new, integrated form of education: "AI general education".

In this form of education, AI education supplies the material for education, and general education supplies the direction, namely "cultivating the capacity for cognitive autonomy". In this way general education fulfils its mission for our time, and AI education moves toward general education.

There is a key question here: can the educational material that AI supplies really fulfil the mission of cultivating the capacity for cognitive autonomy? If it cannot, the definition above is merely conceptual packaging.

The Redbook does not answer this question with assumptions and assertions. Instead it starts its analysis from the disciplinary character of AI and the concrete paths by which AI enters social activity, and gives both the theoretical feasibility and a concrete path of implementation, thereby laying the scholarly foundation for AI general education.

2. The Basic Logic of AI: The Computationalization of Intelligent Activity and of Social Activity

Artificial intelligence is the science of using computers to simulate human intelligent behaviour; in essence it reproduces human intelligent behaviour with computational models. This process can be called "the computationalization of intelligent activity".

The computationalization of intelligent activity gives activities such as recognition, learning, reasoning, planning and judgment a mathematical form that can be constructed, run and tested. Take machine learning as an example: how a system learns from data, what objective it adjusts its model against, and under what conditions it is evaluated can all be made concrete as observable computational processes.

People can change the training samples, compare how different models perform, and examine a system's errors in new situations. Although a model's internal mechanisms may be hard to explain, its inputs, outputs, operating conditions and differences in performance can become objects of analysis and experiment. This gives human intelligence an external form that can be explored in depth.

On the other hand, AI also brings social activity into the computational process. Real-world objects are represented as data, the goals of social activity are converted into training metrics, and historical experience enters models; model outputs in turn enter actual action through platforms, organizations and institutions. A recommender system, for example, has to turn user behaviour into data, organize the ranking of content according to some objective, and then influence users' choices through what the interface displays. Users' new behaviour may in turn be recorded and become the basis for subsequent computation.

These processes of intelligent systems can be instantiated as computational workflows that can be taken apart and analysed, and simulated with data of a suitable scale, thereby providing examples for study and learning. Through these examples, learners can gain a clear understanding of the paths by which AI enters social activity, and can analyse in depth the social impact of AI.

3. How AI Cultivates the Capacity for Cognitive Autonomy

"The computationalization of intelligent activity" and "the computationalization of social activity" together determine the basic content of AI general education. But to cultivate the capacity for cognitive autonomy, we need to analyse further the roles this content can play in general education. We sum them up in two aspects: a body of knowledge and training examples.

1. A body of knowledge

As a body of knowledge, AI supplies key concepts, methods and system structures. How data represent reality, how models form capabilities, how results are tested, and under what conditions a system is valid: all of these constitute substantive knowledge. Once AI enters society, data sources, platform objectives, organizational workflows and the allocation of responsibility also become basic content needed to understand how it actually operates.

This knowledge can help learners form a well-grounded understanding of AI. For example, only by understanding that training data may differ from the conditions of actual use can one explain concretely why a model fails in a new situation; only by understanding the objectives and feedback mechanisms a recommender system adopts can one analyse why the information one sees has passed through a particular selection. Without such a knowledge base, training in cognitive autonomy becomes empty talk.

This also means that AI general education must first of all provide a scientific and systematic structure of knowledge. It should help learners gradually build connections between concepts, be able to explain the basic mechanisms of systems, and understand the relationship between technology and the way society runs. The depth of the content can vary with age, experience and task, but understanding knowledge is always an important part of the educational process.

2. Training examples

As training examples, AI further turns many cognitive questions into concrete processes that can be observed, compared, intervened in and tested. What "turning into examples" stresses is using real systems, or systems designed for teaching, to present how conclusions are formed, how errors occur and how consequences arise in a concrete, operable form, such as code, models, operating workflows and environments. Learners can thereby understand concretely how a system runs, and explore the conditions and grounds on which results are produced.

For example, one can design a task of recognizing two kinds of animal in which the animal category in the training images is highly correlated with the background colour. Even if the model performs well on similar images, it may make frequent errors once the background is changed. By adding samples, controlling the background conditions and comparing test results, learners can check whether the model has relied on inappropriate cues. The representativeness of samples, the validity of grounds and the scope within which a conclusion applies all become concrete research questions.

The way society runs can likewise yield examples that can be analysed. For example, one can design a simplified content-recommendation experiment and compare how the ranking of information changes when the objective is click-through rate and when it is content diversity; one can go on to discuss whether, when some content gains more exposure because of its ranking, subsequent behavioural data will be affected. In such an example, how the algorithm affects ranking, how ranking affects action, and how action in turn forms feedback are all displayed concretely and given an operable training procedure.

4. Implications for AI General Education

Having understood the basic principles of AI and the ways in which it contributes to the capacity for cognitive autonomy, we still need to design sound curricula to achieve the overall goal of cognitive autonomy.

The same AI system can be used for teaching algorithms, for training in tools, and also for general education.

In AI general education, therefore, the selection and organization of knowledge content must take "cognitive autonomy" as the overall goal. When studying classification models, for example, one should understand how a model forms its ability to classify from data, and also analyse how the classification criteria are determined, which errors deserve particular attention, and under what conditions the results can be adopted. Knowledge itself is of course important; more important is cultivating, through the learning of knowledge, abilities directly related to induction, generalization, judgment, self-examination and cognitive autonomy.

Assessment of learning should be designed accordingly. A model that runs successfully is not enough to show that students have understood its principles; a completed project is not enough to show that students can make sound judgments in new situations. Education needs to attend to whether students can explain their own choices, whether they know the conditions under which a conclusion holds, and whether they can re-examine and correct things after finding a problem. These are all basic requirements of the capacity for cognitive autonomy.

5. Conclusion

The Redbook draws the conceptual boundary of AI general education clearly. It is a new form of education in which general education gives the study of AI the unifying goal of "cognitive autonomy", while AI supplies solid knowledge content and working examples that can be explored in depth for achieving that goal. Through the convergence of the two, learners both understand the systems they are drawing on and gradually learn to regulate the relationship between themselves and those systems: on what grounds to believe a result, under what conditions to adopt it, when to check again, and how to take responsibility for their own actions.

From now on, "cognitive autonomy" is no longer a slogan; it is realized in a concrete, systematic process of learning.

From now on, "AI general education" is also no longer a vague, cure-all concept to be used at will, but a brand-new form of education with a clear goal, content and path of implementation.

From now on, facing a complex intelligent society, we have an educational tool that can resist cognitive degradation, so that our descendants can keep their independent individuality and their whole character.