Educatec-Educatice exhibition: the Flowers team presents Kidlearn
The Educatec-Educatice exhibition is a key event bringing together education professionals. Kidlearn, a project run by the Flowers team from Inria’s Bordeaux–Sud-Ouest centre and ENSTA–Paris Tech, will be presented at the Inria stand in Paris from 20 to 22 November.
Digital technology serving schools is one of the key areas announced by the government as part of the restructuring of schools for the new school year in September 2013. Inria research teams are conducting work around digital technology and education, in conjunction with those involved in education (teaching staff, local education authorities, pupils, etc.).
At Inria’s Bordeaux–Sud-Ouest centre, the Flowers team is working on robotics, and in particular “developmental robotics”, by studying whether robots can learn in the same way as children, for example. It also includes artificial intelligence in its work, and is consequently developing innovative algorithms to aid learning. In this way, it is adding to research on computer-based learning environments, where intelligent tutoring systems designed to help pupils in their learning are found.
The Flowers team, and especially the researchers Pierre-Yves Oudeyer (team leader), Didier Roy, Manuel Lopes and Benjamin Clément , are working on learning curve optimisation. They have been running the Kidlearn project for two years, testing an effective algorithm in new learning software making it possible to automatically adapt educational strategies to each learner. A firm belief underpinning this project is that digital technology and artificial intelligence can deliver significant help to learners, and in so doing, help a great deal in combating school dropout rates.
Algorithms at the heart of an innovative adaptive method
There is in computing a family of algorithms known as “statistical inference”, including “bandit algorithms” which are highly effective and well known in the world of research. They are algorithms made for “prediction”. In practical terms, in the world of gambling (slot machines), these algorithms can deduce, from the results of turns played on each machine, the optimum choice of machine which will enable the player to obtain the most winnings. The Flowers team has developed an innovative and effective derivative of such algorithms, called RIARIT : “the RIght Activity at the RIght Time ”. By analogy, the learner’s exercise is the slot machine; the learner’s progress is the winnings. Finding the next exercise equates to finding the machine offering the best winnings. RIARIT’s special feature is to offer to the learner the most appropriate activity to produce the best possible progress.
© Inria / Photo M.S.
To test this algorithm, the team developed learning sequences on the use of money (and therefore mathematics) on a computer. The software analyses the learner’s actions and results in order to recommend a particular exercise. A learner can consequently move from level 1 to level 3 if the software deems it appropriate based on results.
The software’s choices are optimised because it uses the theory of intrinsic motivation to increase learners’ motivation, by keeping the exercises appropriate without the level being too difficult, so as not to discourage learners. The incorporation of a motivation module within Kidlearn reinforces the project’s benefit in combating school dropout rates. The software highlights the activity where the learner will make most progress. At the end of the sequence, a personalised assessment for the learner is obtained.
The incorporation of a motivation module within Kidlearn reinforces the project's benefit in combating school dropout rates.
The teacher remains at the heart of the learning process
Teacher expertise is key in designing software of this type. In fact, to supply the algorithm, significant parameters need to be built and an educational basis needs to be given to start the process.
The team set up an experiment for this software in Bordeaux primary schools, especially in classes of 7-8 year-olds. The goal was to evaluate its effectiveness in a real situation before targeting wider distribution. Teachers do not go to the laboratory; rather the researchers go into schools, and by the very nature of the project they give teachers the possibility of becoming stakeholders in a digital science research project. In addition, the team in charge of the project goes to schools with its own hardware, adapting to schools technology configurations.
The project’s next steps
After initial confirmation of the performance of the algorithm used, the money-learning software is being enhanced for the second phase of the experiment. Technology transfer is also one of the team’s objectives, being discussed with training companies and online extra-curricular tuition firms.
The team recently developed a more fun and enhanced learning schedule, and has refined the algorithm. It plans to extend its experiments. In particular, there are plans to test to the project on a larger sample of schools and children in Aquitaine at the start of next year. In addition, the team is also envisaging larger scale projects on online learning platforms with thousands of learners.
Lastly, the team is also taking into account recent digital tools as the software can be used on both tablets and desktops alike.