SDG 4: Quality Education
2. Project Details
Company or Institution
General description of the AI solution
We propose a mobile application embedding a machine learning model that we called: Djehuty (better known by his Greek name, Thoth). For the ancient Egyptians, Djehuty was the god of wisdom, writing, hieroglyphics, science, and the arts.
The machine learning model can learn how to recognize any drawing characters like letters or numbers, and correct children trying to reproduce it in a game.
From the teachers' point of view, it is a teaching assistant allowing them to craft the right content progression, and allow Djehuty to provide personalized and immediate feedback to the child.
From the child's point of view, it is a surprising game with endless variations. Check a more detailed presentation on our website: https://preschooliteracy.com/application-mobile/
Excellence and Scientific Quality: Please detail the improvements made by the nominee or the nominees’ team or yourself if your applying for the award, and why they have been a success.
We present a gamified Intelligent Tutoring System (ITS) to help children learn handwriting with an Android mobile application. The prototype was published as a peer-reviewed paper (Sarr et al., 2020). The ITS was able to animate a drawing model to teach, and record the reproduction of a child to evaluate it, it introduced:
* a teaching module using a variational autoencoder to generate variations of many letters to teach to children
* a student module to measure the distance between the child input and the model drawing.
* a mixed-initiative component to adapt to the exercise to the child's previous input to guide her toward more legible and more fluent handwriting.
The system was deployed in a preschool class member of the Senegalese Red Cross in February and March 2020 before the Covid-19 pandemic hit. We used design-based research (Amiel et al,. 2008) to evaluate our technology. As a result, we changed our research agenda and worked to empower teachers. By letting teachers focus on elaborating the right content progression and letting the AI focus on giving individualized feedback to children.
Our learnings resulted in the current version that was deployed in the classroom for 6 months (January-June 2021) with cheap mobile phones for 50 €. We have used meta-learning to train a neural network teaching assistant. An article documents our learning on our blog: https://preschooliteracy.com/djehuty-dans-la-salle-de-classe/ . The technology is currently trade secret.
Sarr, J. M. A., Yannakakis, G. N., Liapis, A., Bah, A., & Cambier, C. (2020, September). Djehuty: A Mixed-Initiative Handwriting Game for Preschoolers. In International Conference on the Foundations of Digital Games (pp. 1-4).
Amiel, T., & Reeves, T. C. (2008). Design-based research and educational technology: Rethinking technology and the research agenda. Journal of educational technology & society, 11(4), 29-40.
Scaling of impact to SDGs: Please detail how many citizens/communities and/or researchers/businesses this has had or can have a positive impact on, including particular groups where applicable and to what extent.
According to the World Bank (2020), sub-Saharan countries have the highest population growth rate in the world (currently 2.7% annually). The African population is expected to double by 2050 to 2.5bn people according to a UN forecast. However, today only one out of four children in these countries are likely to enroll in preschool (UNICEF, 2019). We believe this is the main reason explaining why the literacy rate in sub-Saharan Africa is currently the lowest in the world (World Bank, 2020). Indicatively, in 2018 only 64% of adults could read and write. We think that technology can be leveraged to massively empower the next generations for greater autonomy through literacy programs. Thus harnessing the African demographic dividend.
During deployment, we have seen that Dhehuty can help easily detect children with writing difficulties, furthermore, it can arouse the curiosity of children with a bad attitude towards writing, and quench the thirst for learning of children who are already motivated.
We envision using Djehuty in rural areas with low access to preschool in collaboration with public authorities and education partners. The goal would be to scale the pedagogy of a handful set of educators to teach children who can access cheap mobile phones. We think that this solution can help sub-Saharan countries leapfrog their way toward better literacy rates before 2030.
* World Bank. 2020. Population growth (annual %)
* UNICEF. 2019. A world ready to learn prioritizing quality early childhood education. United Nations Children’s Fund, New York.
* World Bank. 2020. Literacy rate, adult total (% of people ages 15 and above)
* Nations, U. (2017). World population prospects: The 2017 revision, key findings and advance tables. _United Nations, New york_.
Scaling of AI solution: Please detail what proof of concept or implementations can you show now in terms of its efficacy and how the solution can be scaled to provide a global impact ad how realistic that scaling is.
We have already provided a public version of Djehuty on our website at https://preschooliteracy.com/application-mobile/, and we have conducted a pilot study with a private preschool in Dakar for 6 months, a blog post summarizes our learning and is available at https://preschooliteracy.com/djehuty-dans-la-salle-de-classe/.
Although the application can already provide benefits to end-users, its development is not finished. As Djehuty is purely software, its only barrier to scale up is to find product market fit for sustainability and to develop an efficient channel to find more partners and users.
Our approach toward sustainability is to profoundly understand how the technology could help families. Our market study told us that there are two types of parents we could target:
* The parent who want their children to go directly from preschool to the preparatory course, and jump the initiation course. (This is a specificity of the Senegalese school system, for instance in the French program children go from the last year of preschool straight to the preparatory course). We have one paying client from this segment since June 2021.
* The parents who have children with attention deficit hyperactivity disorder.
Our plans to amplify our reach are:
1. networking with education institutions open to leverage technology for impact as the UNESCO IRCAI is positioned to be.
2. leveraging content creation platform starting with blogging in 2021, and diversifying to podcasting or Youtube videos in 2022.
Ethical aspect: Please detail the way the solution addresses any of the main ethical aspects, including trustworthiness, bias, gender issues, etc.
As we have already discussed Djehuty can help systemize finding children with difficulty. We see it as a tool to detect children with learning disabilities at a large scale.
Furthermore legible and fluent handwriting has been seen as a good predictor of academic performance in children's careers (Simner, 1986 ; Graham et al., 1996). Good handwriting is also correlated to emergent literacy skills, which are essential for autonomy in our knowledge-based economy.
The literature shows that the return on human capital is highest in early childhood and diminishes rapidly as the child grows older (Heckman et al., 2007. See Fig. 12). Comfortable families tend to invest heavily in their children while poor families live in adverse environments. This plays a crucial role in maintaining inequalities. Moreover, wealthy families can afford a personal tutor which dramatically improves their academic achievements. Indeed Bloom (1984) has shown that the average tutored student benefiting from individual tutoring (1:1 teacher:student ratio) performed above 98% of students attending uniquely classroom instruction (1:30 teacher:student ratio).
Finally, we think that using AI and cheap mobile phone technologies can empower teachers to have more impact by giving personalized early literacy instruction at a continent-scale and position Africa to rip the benefits from the demographic dividend by 2050.
* Bloom, B. S. (1984). The 2 sigma problem: The search for methods of group instruction as effective as one-to-one tutoring. _Educational researcher_
* Graham, S., & Weintraub, N. (1996). A review of handwriting research: Progress and prospects from 1980 to 1994. _Educational psychology review_
* Heckman, J. J., & Masterov, D. V. (2007). The productivity argument for investing in young children. _Applied Economic Perspectives and Policy_
* Simner, M. L. (1986). Further evidence on the relationship between form errors in preschool printing and early school achievement. In _Advances in Psychology_