Teaching systems free teachers' hands for more exciting work

Author: David Povolný for em.muni.cz

From the headline, it might seem that Radek Pelánek is mainly a teacher, but his profession is informatics. He has always been interested in asking interesting questions and finding solutions, so at the Faculty of Informatics, MU, he focuses on adaptive learning and the development of learning systems based on artificial intelligence. Their goal is to change the way some types of knowledge are taught today and give teachers space to do things the computer can't do.

What brought you to computer science?

I grew up in the '80s, so I experienced the first home computers that didn't do much, and there were only a few stupid games on them. And so my friends and I started making our own games, by the way, in the now almost forgotten and quite bad Basic programming language, in which I had a lot of bad habits.

So clearly through computer games?

Actually, yes, I was always close to games, because I was in a tourist club since I was a child, and there you always thought of something for yourself or others to have fun with. The fact that the computers didn't know much at the time was, in retrospect, quite a significant advantage over today because it forced us to try to program something very soon. There are a lot of things ready today, there are many more options, and it doesn't force you so much into programming.

So have you been interested in computer science since childhood?

I hesitated for a while whether to study mathematics or computer science.

Is it possible to do computer science well without mathematics?

Informatics is a very broad field today, which includes many professions. Suppose someone makes websites and focuses mainly on the design part. In that case, they probably won't come into much contact with mathematics. Still, in general, it can be said that whatever you do in computer science, you will do better if you have a solid mathematical foundation. That's why we pay a lot of attention to mathematics at the Faculty of Informatics. From people who say they don't need math for anything, I sometimes feel that it's because they don't understand it well enough, so they don't even realize they miss it. For example, statistics, data processing and evaluation are beginning to play a crucial role in informatics today. Without a good mathematical foundation, you can happily evaluate, make fundamental mistakes and not know about it at all.

You focus on what is called adaptive learning, or more precisely, on computer learning systems. What are they?

The basis of these systems is that we try to imitate what every good teacher does, who consciously and subconsciously adapts to the student or class of students. Adaptation can take place on many levels. Our group at the faculty is mainly concerned with adaptation to knowledge, i.e. how to present tasks and problems that are just as difficult for the student.

How do you do it?

You need to interact with the student first, and you try to guess what he already knows. According to that, you will offer him something that is not boring for him but at the same time not too difficult.

And that is the statistic mentioned.

Exactly.

What else can be adapted?

For example, an affective state - whether the student is bored or excited, for example. Even a good teacher can recognize and adapt the teaching to it. The teacher can also take the metacognitive abilities of a given student, preconditions for a particular way of learning and so on into account. This is currently being addressed by research, but it is a considerable challenge to respond appropriately to these aspects and detect them at all. In particular, we are not going in this direction. The commercial products based on adaptive learning that are available today are primarily about adapting to knowledge.

What is in the gut of these systems?

The basis is a system that produces numbers from clicks or written answers, which serve as a model of knowledge of the student. The problem is that this is relatively strict data from which you will learn little. Another difficulty we face is the different types of knowledge and skills that require a different approach. It is quite different to teach factual knowledge that requires regular repetition of the same thing and teach rules, such as counting fractions, where you need to practice changing tasks in different contexts. In addition, we have cases where these two aspects are combined. A simple case is the words that have "y" after b. On the one hand, it is necessary to learn these words by heart, but at the same time, it is essential to be able to apply the rules of word derivation and distinguish between conquering and recharge. So making a learning system that teaches you to write "y" and "i" even after b is not easy at all, even if you think at first glance that it's just a matter of drilling.

But you do not only work with the data of a specific person. You also evaluate the data as a whole.

Yes, we aggregate data from all users in our systems and try to infer from this, for example, the general difficulty of tasks. In some cases, statistics show you interesting things. For example, in testing how to write "kdo s koho". When we let people choose whether to "s" or "z", they have a worse success rate than monkeys.

Why?

It is a thing that makes sense in terms of language development, but today people no longer understand it and write mostly "z" monkeys have 50 to 50. When people have such a frequent error rate, the question arises whether the error is more rules than in people.

In the context of adaptive learning, there is talk of artificial intelligence. Where is the limit today for what artificial intelligence is? From what you say, it doesn't quite sound like the system learns on its own.

This is the problem of the definition of artificial intelligence, which is unclear and also changes over time because the moment computers do something, people will say that it is not artificial intelligence yet. Today, the term Good Old Fashioned Artificial Intelligence is also used to refer to what you have suggested, a system that simulates how a person behaves. But a lot of today's artificial intelligence is not about this but about intelligent behaviour. It can be, for example, playing chess, in which the computer behaves quite differently than a human. Today, artificial intelligence is also not so much based on some super-complex algorithms as on working with data. So intelligence is not in the system as such, but in having access to a vast amount of information. The result can be things like recommending relevant books on Amazon. The goal is usually not to replace a completely human factor but rather to supplement it appropriately.

Does this also apply to your learning systems?

Yes. I always try to emphasize that our goal is not to replace the teacher but to make his work easier. Our applications at the moment include practising basic things, such as the difference between "y" and "i", the multiplication table, the basics of programming, some facts. And it can be more efficient on your computer. Either because the computer remembers better that the student is struggling with a word in English, or because he has more patience and will practice with you all the time, even at night. It is also easier to get game elements there. This will give the teacher more time. We free his hands to do more exciting things, and instead of drilling the grammar, he devotes himself to making his students able to write meaningful text. So it certainly won't take teachers' work.

What feedback do you have from teachers?

The feedback we receive from teachers is generally positive. However, this is certainly also due to the fact that our systems are mostly used by teachers who are close to technology. If teachers were given the use ordered above, it would certainly not be so clear-cut.

Do such systems get into schools on a large scale?

They are definitely used, but I dare not comment on how large the scale is. There are significant differences between schools. I know many of them that our systems usually use in the classroom because they support it and have an excellent technical background. In others, we see that the systems are used for homework and out-of-school practice. And there are also schools where they don't want to hear much about computers.

You have been doing this for many years, and everything is changing rapidly in informatics. Has adaptive learning shifted a lot?

Developments in the field of adaptive learning are by no means rapid because we do not have a clear goal in our country. This is the case, for example, with colleagues who focus on more traditional areas of artificial intelligence. There has been tremendous progress there over the course of a few years. Tasks such as defeating a man in a go game or recognizing a cat in a picture that was still out of reach a few years ago are now solved. We do not have a simple milestone. We are trying to "improve teaching", which includes a number of sub-tasks. Many of the partial problems that our group solves are methodical. How to measure that a student has learned something in the web system? What does it actually mean that a task is "difficult" or that "two tasks are similar"?

What is meta for you personally?

Having a practice system that works from a people's point of view gives relevant recommendations on what to focus on next, offers a tailor-made learning process that is smooth and nicely linked. My ambition is not that users will say "the system is smart", but that they will not feel that it is stupid. For example: When you finish a small multiplier, the system should not recommend continuing in logarithms. This may sound naive, but we're working on systems that try to cover a wide range of different knowledge. We do not have the ambition to do something terribly complex for one type of knowledge, as is commonly done by colleagues in English-speaking countries, where it can be attractive due to the size of the English market. Instead, we want one easy-to-use system that will work for many things, from math to Czech and geography to programming. For that, however, we need versatile algorithms, which is not easy at all.

The systems that came from your research group are commonly used in schools. What does your collaboration with schools look like?

The most developed system is Umíme to, which is done by my former doctoral student Petr Jarušek. For a long time, it was something that was created in our free time. We developed it in the evenings and on weekends. Students learned different things about it, wrote diplomas about it, and started to get more users. It seems that it would even feed itself. Now we figure out what to do next because it has already crossed the boundaries of the student project.

You are very close to pedagogy and have long been active in the group Instructors Brno, which is behind the now legendary encryption game Through The Darkness. Does this hobby still last for you?

Since I have children, there is not much time and energy, so I only organize an encryption game Sandwich, which runs on the Internet and, moreover, has a righteous overlap because the earnings from it go to the non-profit sector. And Through The Darkness, who was now in her 20th year, is already a participant. But I make games for children, in the summer we go to camps, and I prepare various treasure chests and fighting games.

Participating in Through The Darkness as a player must be easy for you as a former organizer.

Not at all. Having the ability to create cyphers and solve them is quite different, so I'm definitely not one of the elite cypher players. My friends and I always hope to decrypt something and not end at the beginning.

Do cyphers have anything to do with what you do at university?

Rather at a general level. I like to come up with problems for people to solve. I enjoy preparing an exciting assignment and observing how someone manages to solve it, where it gets stuck and what is simple. This is exactly what is interesting for me in our teaching systems, I really enjoy inventing assignments and even thinking about what a good assignment is. But even those cyphers have, I believe, a significant educational benefit. One learns the general skills needed to solve problems and also work together in a group because they are usually team games where you need to put your heads together.

As I listen to you, have you ever considered teaching children in primary or secondary school?

Not exactly. I enjoy being in the background more, I'm more of an introvert. Not that I can't stand in front of the class, I also teach Introduction to Programming at the Faculty of Informatics, where I have hundreds of people in the lecture hall. Still, I don't consider contact teaching to be my strength, and I wouldn't be the type for managing class discipline. Even though I go to camps, I enjoy inventing the games rather than organizing the children. That is why, for example, I was interested in organizing Through The Darkness, where one hardly came into contact with the players, because everything was prepared for them in the field, and we often waited until the finish line, where almost no one reached.

I think this is also one of the misfortunes of the current school system, that teachers are wanted to master all components of the pedagogical process. To figure out for themselves what and how they will teach, and then even put it into practice, provide it in an organizational way and support the children. Each of these requires completely different skills. That's why I see a lot of sense in what we do, that we prepare teachers with tools for some ways of learning and give them space to focus on things where they are irreplaceable. I think there are a lot of people who have a knack for working with children, but they are slowed down as teachers by having to think about what and how instead of doing it. As a result, they are overloaded, and the work with children also suffers. Even with that, perhaps our systems should be able to help more and more.