EDUCATION FACING ARTIFICIAL INTELLIGENCE (AI)

CDD 71 English

EDUCATION FACING ARTIFICIAL INTELLIGENCE (AI).

The Commission on Artificial Intelligence (AI), launched in September 2023 by the Prime Minister and led by Anne Bouverot, President of the ENS Board and Philippe Aghion, Professor at the Collège de France, has just submitted to the President of the French Republic an important report on how to develop this new technology in France.

It is certainly surprising that the objective given by the Prime Minister to the Commission was, with the very French arrogance, to place France "at the forefront" of the sector from the outset, even though, obviously, a so important subject can only be dealt with effectively on an international scale, if only for weighty questions such as computing power or energy and financial costs. In this case, optimally, at the European Union level.

The report is nevertheless important and raises many questions, some of which are inevitable and weighty, such as the impact on employment, data and computing capacity, energy, and financial costs.

The reasons for a new craze for AI

The first interest of this report is to explain why this sudden emergence of a new interest in AI over the past year. As the report, and many experts before it, reminds us, the technology is not new. It is generally dated to the mid-fifties, with its name first appearing in 1956 with McCulloch's work. The power of this new science called computer science began to assert itself and the ambition to endow machines with faculties analogous to those of the human mind, such as reasoning, solving problems, using written and spoken language... To the initial symbolic approach based on classical computer instructions (if... so...), a different type of approach has gradually been replaced, sometimes called connectivist, because it is based on the analogy with the brain functioning, its neurons and the synapses that link them together. Mathematical functions simulating this operation make it possible to identify statistical relationships between data and are thus likely to "teach" a machine to recognize links on enough data to provide a solid basis for this "learning". However, this method, called "machine learning", presupposes that as much data as possible is collected and that the available computing power is able to process it. However, it is precisely these two elements that have developed strongly over the past generation under the dual effect of the digitization of data in all fields and the appearance of new high-powered calculation tools. From the conjunction of the two factors, under the impetus of a few pioneering companies and their founders, an impressively efficient AI was born, not only in the field of gaming (a field long favored by human intelligence confronted with artificial intelligence) but also in many other fields. Generative AI, i.e. capable of producing content using a multiplicity of sources as soon as the machine has the corresponding data. Chat GPT, for example, has demonstrated this spectacularly, but it is only the first example of a list that will soon multiply.

Can AI transform the pedagogical relationship?

The report provides an impressive overview of possible areas of application for AI and makes 25 recommendations. There can be no question of mentioning them all in this simple column. But those dealing with education are worth doing. Maybe it's just to make people understand what distinguishes a machine's learning from that of a child. And to take the opportunity to reflect on the possible changes in the teaching profession. Not its disappearance, as is sometimes feared, but its possible evolutions.

The recommendation reads as follows: "Encourage the individual use, large-scale experimentation and evaluation of AI tools to strengthen the public education service and improve the daily lives of teaching teams".

First, it is striking to note the extreme caution with which the question of education is approached. For a simple reason, which the committee has done well. Almost, as each student is different, "the standardization that digital technology requires to collect and enhance data" is ill-suited to the pedagogical act. The latter is made up of "human relations" through which the teacher must constantly adapt to the contexts and to his students based on the minimum standards constituted by the curricula and the internal regulations of the establishments. In addition, we immediately guess some obvious dangers, homework or exam fraud, increased addiction to screens...

But for all that, we also sense the potential that AI represents regarding students' individual learning. The report cites the main ones: support for the creation of courses, support for pupils in the form of tutoring, partial assessment by automatic correction, teacher training, etc. But all these potentialities raise serious questions. So, it is for tutoring. Many recent experiments have demonstrated the value of the so-called "flipped classroom" whereby the course in its lecture expression is mainly broadcast or distributed digitally (if all students have the means to access it) and the "face-to-face" periods in class are devoted to a good understanding and deepening of the discipline. On the contrary, what the commission suggests would tend to make tutoring a privileged use of generative AI. This presupposes not only the provision of such tools to all students, but above all their ability to fully understand the learning difficulties of each student. The hypothesis is daring, unless we assume that it can be accompanied by the presence of an adult behind each student and not substitute for it.

About teachers, the nature of their profession and their training, the report shows the same caution because of the sensitivity of the subject. Noting the need for the integration of AI in education through involvement and complementarity with teachers, it suggests that AI could "facilitate the evolution from a 'knowing' teacher (subject expert) to a teacher "accompanying the student, outside the historical paradigm of monodisciplinary transmission". Even admitting that "historical paradigm" (already often questioned, even in France, much more so in other countries), there is no doubt that many will question whether to read in it an inevitable evolution or on the contrary a strengthening of the teacher’s role, both "knowing" and "accompanying".

In any case, this implies a profound transformation, if not of the profession, at least of teacher training, not only by training them in the most relevant uses of AI but by considering their own creativity and ability to use such tools. It is hard to see why teachers should be denied "learning" abilities that we agree to confer on machines! In any case, it's a safe bet that these questions will quickly come to the forefront.

The costing of this recommendation is also interesting to consider, but unfortunately the details are hardly indicated in the public document. It is valued at €1 billion over five years. It can be assumed that it corresponds to the cost of training teachers and needed equipment for the recommended large-scale experimentation. But all this deserves a lot of clarification.

Are higher education and research essential resources?

At least two other recommendations relate directly to education, higher education, and research. The sixth and, to some extent, the nineteenth. The sixth proposes to "generalize the deployment of AI in all higher education courses and to acculturate students in secondary education to make specialized training accessible and attractive". In a sense, this makes sense, especially regarding the acculturation of secondary school pupils which is likely to attract them to the sector. However, it is still necessary to specify what is meant by "acculturation". To this end, the commission suggests that a reflection be carried out to "identify the contribution of each subject taught and allow all students to benefit from learning related to AI" regardless of the path chosen (general, technological, vocational). As the disciplines and curricula that currently include the issue are in the minority, one can imagine the immense work that should be done in this direction, especially since it presupposes teachers themselves who are likely to raise awareness, if not acculturate, the desired situation. All this therefore implies a necessarily very gradual process that involves research but also, once again, the initial and in-service training of teachers, which is little mentioned in the report except through the recommendation to deploy AI in all higher education courses. But does it make sense, even in the medium term?

While there is nothing to prevent higher education courses from making such a deployment if they can, it can easily be objected that the most important thing is to try to meet the needs of specialists as the committee considers them as a matter of priority, not without recourse to certain hypotheses that are certainly debatable, as is the case whenever there is a claim to want to adapt a training offer to a demand for jobs. This priority in no way requires the obviously illusory response, even by 2034, of generalizing this deployment of AI to all higher education courses. If we want to stay within a horizon of possibility, it would already be remarkable to succeed in tripling (i.e. to about 50,000 per year) the number of students in specialized computer science training at bac+3 for the development of AI. It will be difficult but not impossible. The same applies to the Commission's estimated target of only 1.5% of all students, the need for people combining disciplinary competence with AI competence for deployment in their discipline, within ten years.

But the question remains: "Who trains the future specialists?" It seems sensible, but only if the question "who trains the trainers" has been answered beforehand. It is difficult to see how a convincing answer can be given without a considerable effort on the part of companies already specialized in this field, and without higher education institutions being able to bear the corresponding costs.

An "AI exception"?

A final observation can be made on recommendation 19, which advocates "assuming the principle of an 'AI exception' in the form of an experiment in public research to strengthen its attractiveness". This experiment-exception comes down to a few simple principles, which have already been mentioned often: basically, for public researchers, a principle of "zero hindrance" at the administrative level, a facilitation of shared work with companies, a support for the installation of foreign researchers, and above all a significant increase in salaries. No one can balk at such a prospect. But we can see that it raises at least two questions: on the one hand, the state’s capacity to satisfy it, which is unlikely in the current conjuncture or even in the dominant state culture; and on the other hand why only AI would benefit from it. There is no doubt, even if we consider the development and deployment of AI to be a priority, that many other scientific fields could claim equivalent priorities.

At a time when the government is showing a desire for strong savings in public spending and seems to be choosing to do so in the areas of preparation for the future, nothing seems very realistic in this recommendation, at least in the short term.

In addition to the fascinating panorama offered by this report on the current state of AI development and deployment, the questions it raises and for which it ultimately offers few convincing solutions, at least in the short term, will be particularly noteworthy in the field of education. It is necessary to avoid the mistakes of the past, when brilliant minds believed that the introduction of computers into the classroom would substantially change the quality of learning. The pioneers of computer use in the 1980s and 1990s should not only be replaced by the pioneers of AI in the 2020s and 2030s. Nothing will be done without teachers, without their awareness of the AI potential, without training adapted to what teaching today means, and therefore profoundly transformed.

Jean-Paul de GAUDEMAR

April 21, 2024

You can find this column, in French, by subscribing for free to my platform "jeanpauldegaudemar.substack.com" or in English by subscribing for free to my platform "jpdegaudemar.substack.com"

.