Artificial intelligence and education

Choices to be made now

The subject of artificial intelligence in education has been the subject of much debate, and raises many questions. The aim of this article is to take a look at some of the key issues in education management. The disruptive new field of A.I. can be taught, accompanied by operational tools that touch on student and teacher management data.

What is artificial intelligence?

"Artificial Intelligence (AI) is a set of theories and techniques aimed at creating machines capable of simulating human intelligence.
Often classified in the group of mathematics and cognitive sciences, it draws on computational neurobiology (particularly neural networks) and mathematical logic (part of mathematics and philosophy). It uses methods for solving problems of high logical or algorithmic complexity. By extension, it includes, in everyday language, devices that imitate or replace human beings in some of their cognitive functions.

AI applications include search engines, recommendation systems, natural language understanding, autonomous cars, chatbots, image generation tools, automated decision-making tools and competitive programs in strategy games.

Since the concept first appeared, its aims and challenges, as well as its development, have given rise to numerous interpretations, fantasies and concerns, expressed as much in science fiction stories and films as in philosophical essays. While specialized or generative artificial intelligence tools have proved their worth, reality still seems to keep generalist artificial intelligence a long way from the performance of living beings in all their natural aptitudes".

Source : wikipedia https://fr.wikipedia.org/wiki/Intelligence_artificielle

At what point does an education manager have to make choices?

The word education manager is used here in a very general sense. It should be taken in the sense of a decision-maker in the choice, use... of an artificial intelligence that will be involved in his or her work.

The possible choices are as follows

choosing AIs to be taught or used by

students
teachers
other professionals (career guidance, nursing, etc.)
fellow managers

choose AIs to manage the school or university ecosystem by

administrative staff
logisticians
other interrelated organizations
...

What to look out for

Best practices

"Based on this observation, several good practices can be implemented to limit and prevent sampling and cognitive biases as much as possible. To this end, we have listed 10 golden rules to be observed during an AI project:

Select real data, from the same source as that which will be used in production: the more accurate the training data, the more accurate the results.

Use recent data and update them as much as possible, as past data cannot accurately reflect current data (changes in practices, ways of thinking, etc.).

Have a sufficient quantity of data: so as to identify objective trends.

Start from a business need and identify the target user with precision: it is indeed much more difficult to develop an AI system if its objective is not tangible and clearly defined. In addition, knowing the target user, who will therefore be an integral part of the AI's continuous improvement, enables us to take into account potential biases.

Communicate with the people in charge of training and train them in advance: similarly, the clearer the communication and the best practices to be adopted, the more doubts and errors will be avoided during training.

Use independent validation: an algorithm's performance must be validated on independent data, not used for training.

Form a multi-disciplinary team: it is essential to be able to take a critical look at the solution developed and its performance, and to always ensure that it answers the question posed, and that the results obtained make sense and do not risk being tainted by bias.

Ensure that the variables used for training are coherent and relevant to the targeted result: for a chatbot, for example, you need to prepare the algorithm for the questions it will have to face when it goes into production.

Use tools to identify biases: Lime, Open Scale, AI Fairness 360, etc.

Last but not least, avoid putting too much trust in AI, and keep a critical eye: especially for the target user, because the more the AI is corrected, the more effective it will be".

Source : AI biases: how to control them? - Romain Lamotte - 2021
https://kpmg.com/fr/fr/blogs/home/posts/2021/1/comment-maitriser-utilisation-ia.html

Intelligence cannot be left to run free. It's your responsibility to regularly check the above points.

What is a cognitive bias in Artificial Intelligence?

"Cognitive biases are repeated thought patterns that lead to inaccurate and subjective conclusions. Confirmation bias, for example, refers to the brain's tendency to seek out and focus on information that supports what someone already believes, while ignoring facts that run counter to those beliefs, despite their relevance. Attribution bias occurs when a person tries to attribute reasons or motivations to their actions or those of others, without these reasons or motivations necessarily reflecting reality.

Cognitive biases can help us make decisions more quickly, but sometimes at the expense of rationality. To date, a total of 180 biases have been identified as altering our judgment. It's an impressive number, and one that calls into question the fairness and impartiality of our day-to-day decision-making.
However, so-called cognitive biases are not the only definition of bias. In statistics, for example, the collection of data from a sample that is not representative of the general population constitutes a "sampling bias". The results produced cannot lead to conclusions that apply to the entire population.

We need to go back to the basic principles of Artificial Intelligence solutions to understand the implications of these biases. First of all, artificial intelligence is designed by a human. It can, in fact, become a magnifying glass for its own biases. Firstly, an AI is designed by a human being, so it can become a magnifying glass for its own biases, which include the fact that, in this field, you only find what you're looking for: an AI simply reproduces what it has been developed and trained to do, and only answers the question put to it.

Secondly, an AI is trained on the basis of known examples, selected by a human. It is therefore essential that these examples are reliable, in sufficient quantity, and themselves untainted by bias - sampling bias, for example".

Source idem : AI biases: how to control them?

Another no less significant risk is the use of data by other users or unidentified users.

This point touches on two distinct issues

cybersecurity
backdoors

Data managed by artificial intelligence is not protected by this very fact. And the more sensitive the data, the less it needs to be directly manipulated by internal or external third parties (human or technological). Every time there are gateways, open doors and bridges to other systems, the risks multiply.

The second point is more sensitive and difficult, as it involves risk assessment. As mentioned in the cognitive section, a computer program is created by a human being and then developed by other human beings. So there are already risks of defects in the model, which can multiply and intensify over time. For example, a humanitarian program can become a purely financial program over time, if an unwanted grain of DNA of a financial nature is introduced.

The real risk, however, is not that of error, but that of setting up internal processes that do not correspond to the customer's wishes. The problem may come from the company itself, or from an employee's personal initiative. For example, artificial intelligence needs to cross-reference as much data as possible in order to refine its operation, and could use your data to this end without your knowledge. That's why it's important to sign confidentiality protocols with these service providers, especially if they don't exist in the marketing protocols.

As for unknown programs that may have been installed without anyone's knowledge, they are sometimes impossible to detect, but always end up surfacing. That's why we need to be extremely rigorous with data use and security, so that you don't have to deal with some of the problems later on - another reason to train your staff effectively and ethically. And if the protocols put in place by your community or country don't seem efficient or ethical enough... then there's nothing to stop you creating your own complementary contractual documents with your wishes in mind.

And, if the intermediary or creator of the solution refuses to sign, then tell yourself that he can't be a good partner to take care of and manage your data. Don't forget that you are a customer, even of a turnkey system, and a customer should be served according to his or her wishes, not contorted into an existing product. If you can afford it, opt for custom-made or modular tools.

Artificial intelligence is changing the world

Artificial intelligence remains a tool that must remain at our service, there to help us in our work. Revolutionizing the world of education is complicated. To avoid making the wrong choices, resources, partnerships and working groups on the same themes are available. We're not the only ones facing these choices.

Source image : Pixabay - Hobim