egoTeach’s solution for global primary education is inspired by research in cognitive and educational science, combining the psychology of development, motivation, and learning. The impetus for our approach is that out of hundreds of studies, the strongest educational intervention found so far is to give each child a personal tutor (Bloom, 1984). Although 1-on-1 tutoring is impossible to scale worldwide, the primary benefits of personal tutoring can be mimicked and actually enhanced using tablet technology.
To most effectively use tables, we must abandon the classroom model of education, based on the Aristotelian idea of students’ minds as a tabula rasa waiting to be filled with knowledge. Instead, learners should be seen as budding scientists: actively creating hypotheses in their minds and putting them to test through their actions. Initially guided only by their intuitions, their hypotheses will often be wrong, but through immediate and interpretable feedback, they can be led to learn concepts of increasing complexity. In the field of human-computer interaction, this unfolding mental process is known as active sense-making (Renner, 2011). Learners generate possible hypotheses to explain their environment, and then gather new information via interaction to test the hypotheses. Upon seeing the outcome of their inquiry, they must understand and update their beliefs accordingly. As the cycle continues, learners achieve a better and better understanding of the world. Importantly, the process of sense-making mirrors the steps of the scientific method.
egoTeach introduces a suite of games, built on empirical foundations from the domains of cognitive and educational psychology. The software enables active learning, allowing the learner some choice in both the content and approach, by letting them choose and explore games presenting different material, thus motivating and engaging the learner. Using an adaptive AI engine shared across games, the software intelligently adapts the sequence of material given to the student based on continuous probing of the student’s knowledge, re-prioritizing content based on correct and incorrect responses, based on distributed practice effects in memory studies. Our games are designed to be learned via active exploration—trial-and-error, as in science—and are directly built upon the underlying concepts we are trying to teach—not just a gamified test. Combined with continuous positive and negative feedback that is shared across the games, our use of universal nature themes allows children to accurately induce and bootstrap their understanding of the concepts. egoTeach contains 9 primary games for teaching the core literacy and numeracy concepts, some of which unlock only after learning in earlier games.
In the ladybug game, learners hear a number and must identify the correct number from several options. Learners then tap the target ladybug the correct number of times, receiving intuitive audiovisual feedback as they learn the link between numerals and quantity. Incorrect actions slow the bugs’ movements, result in fewer options on the next round, and increase the probability of seeing that number again sooner. Successful interaction results in more options, faster speed, and a decrease in the number’s priority. Over time, the game proceeds to teaching sequences: counting by 2, 3, and so on. Star points reinforce the concept of counting.
In the strategic ant rescue game, learners hear a target number and attempt to move the correct number of ants to the large lilypad, where they find their food. They learn addition, including the standard notation, shown after each subproblem is completed. implicitly teaches the commutativity of addition: it clearly doesn’t matter which lilypad is the source, and which the destination. It is self-directed in that learners choose their own path, and there is no time pressure. Correct and incorrect responses adjust the AI’s estimates of their knowledge. Game dynamics adjust with responses: incorrect responses decrease the maximum number of distractor lilypads; correct responses increase them. Subtraction problems are introduced and appear more often based on correct responses.
In the nest building game, a quantity of eggs appear, and learners need to draw a nest to hold that number of eggs. It’s not that interesting for small or prime numbers, but they’re good for learning the mechanic, which teaches the geometric interpretation of multiplication. They intuitively see that multiplication is commutative: length and width are interchangeable dimensions. As always, responses are used to adjust the difficulty. Pilot testing in Brazil and Tanzania showed that although this game is initially a challenge to understand, children persevere because they love being able to tap all of the eggs when successful—which further reinforces the link between number and quantity.
To teach letter identification, a letter is named and the bee races them to the flower with the named letter. A straight path is good for initial learning, but in later stages the bee’s path wanders. The bee speeds up or slows down depending on response correctness, and more or fewer options are shown. Both upper and lowercase letters are practiced. A version of this game, unlocked after the alphabet is learned, is used to teach spelling: a word is spoken, and the bee is directed to pick up the correct letters—a spelling bee, so to speak. We start with very common short words, and gradually increase orthographic complexity. We’ve recorded the 1,000 most common words in Kiswahili and English.
This games teaches spelling and vocabulary. It’s similar to the common game ‘hangman’, but without the macabre theme, and with only limited availability of letters. At first, only the correct letters and one other are available, but the game introduces extra letters as they advance. If they guess an incorrect letter, the hidden object is partially revealed, giving them a chance to sound it out. The game starts with 50 animals we’ve illustrated to teach the concept, but proceeds to work with the other 950 common words we have recorded. Words are repeated over time increasingly-long intervals, in accord with a spaced repetition strategy suggested by memory experiments.
The classic memory matching game can be used for additional practice of several concepts, including upper- and lower-cased letters, as well as shape names, simple addition problems, and sight-reading of short words. The amount of time given to find all pairs, and the total number of pairs is adjusted as a function of performance.
For practice writing, we provide a simple sandbox in which they have the opportunity to draw whatever they want—but with a subtle hint to practice writing the alphabet at first, and later audio prompts to try writing animal names. Taking a constructivist approach, we will also leave a basic word processor, and are working on number line and physics sandboxes to enable children to use their new knowledge to create.
Finally, a separate app contains a library of open source illustrated children’s books that can be freely chosen from, but organized in order of reading level. When tapped, sentences will be highlighted and read aloud, providing assisted practice as required. Page indicators reinforce the concept of a number line. We are recording more books, and have written some science lessons introducing concepts of physics, chemistry, and astronomy that are now being illustrated. Our modular design makes it easy for others to create and drop in additional content, with or without recorded audio.
Our pilot testing in Tanzania verified that children can learn the games without instruction, that they like the games and will play daily for over an hour. We also saw that although some children prefer to work alone, others share and work together, and seem to enjoy teaching each other how the games work. They quickly figured out how to switch games, and eagerly tried them all, and continued to switch games over the five days of intervention. When we surveyed the children about their favorite games, at least a few children named each game, suggesting that variety and choice may be important for motivation, at least. We collected over 20,000 trials during the intervention, and have started to do in-depth analysis of usage and in-game learning progress.
We are excited to release egoTeach for Android next week, and will follow up soon with an iOS release thanks to our choice to be a web app. We are also eager to conduct scientific studies using the games as a research platform, for example to compare self-directed learning to partially-A.I.-directed learning, where games can be locked or unlocked depending on learner’s proficiency and recent use of them. Having piloted in Tanzania and Brazil, we are applying for grants to fund scientific testing in the US, the Netherlands, and abroad via national funding mechanisms, as well as NYU’s Global TIES international research center for children’s education. George is supervising several bachelor Artificial Intelligence students this Spring working on new A.I.-assisted education games, and on improving speech recognition and synthesis in Swahili with deep neural networks. Kat and George are currently organizing a special issue in an academic journal on the topic of adaptive, self-guided educational games.
In summary, egoTeach is a collection of games presenting the core literacy and numeracy concepts in a way that enables self-teaching, reinforced by a shared content engine with an adaptive A.I. back-end to re-prioritize content. The difficulty of the game dynamics adjust to the learner over time, keeping them engaged, while they still maintain the freedom to choose different approaches to learning in the various strategy and action games. As an official participant in the Global Learning XPRIZE, egoTeach is ready to ignite evidence-based improvement in primary education, worldwide!