AI-powered Adaptive Personalized Learning Platform

Traditional education needs a lot of work. Low proficiency levels threaten progress in poverty reduction, incoming demand far outweighs supply in human capital and infrastructure, and two-thirds of children and adolescents are not learning globally even though most attend school. Poverty and lack of education are intrinsically linked. Schooling in its current form isn't fitting the bill.

Through our platform, all learners are guided through content adapted to their specific needs and drives, optimizing for engagement and proficiency growth. Personalized recommended paths are offered so that the user can build and follow custom credentialing paths. 

The main potential impact of the platform is through mobile phones. There are 450 million unique mobile internet subscribers in Latin America. Without the need for additional infrastructure, mobile-based personalized learning pathways can have a broad impact through adaptive, flexible career modes.

Current schooling systems are overstretched and underperforming. Teachers are overworked and burnt-out. Educational systems and institutions are struggling to keep up with the fast pace of technological evolution. Despite considerable progress in coverage, proficiency levels remain critically low for over 600 million children and adolescents all over the world (UN, 2019). 262 million children and adolescents attend school but won’t be able to read by the end of primary and 78 million will drop out altogether. In Latin America, 35 million, over a third of school-age population, are not reaching minimum proficiency levels in reading and mathematics (UNESCO, 2017). 

In a recent MIT-Solve Talk, some reasons have been put forward: 

• Schooling looks like an assembly line, today’s tests don’t measure what we need when we need it (Arthur Levine, Woodrow Wilson Foundation).
• Teachers are pressured to teach bodies of content, being a student comes before learning (Tina Grotzer, Harvard Graduate School of Education).
• Standardized assessment and a focus on memorization impedes the development of scientific thought and inquiry (Chris Rogers, Tufts University).

“The consequences are grave: If all adults completed secondary school, the global poverty rate would be more than halved.” (UNESCO, 2017)

This solution is still at pre-development stages. At this time, three main potential use cases arise from the basic framework described in the solution section:

• Bridging formal and informal education and employment. In Mexico, over 60% of all jobs are informal. Our approach seeks to provide mobile-based education through employment pathways which merge vocational training and AI-powered self-directed micro-education.
• Educational institutions. Although restricted by technology and connectivity access, significant proficiency increases are within reach for schools able to implement this technology. Our platform seeks to address the issue of overworked and under-qualified teachers by providing them with evidence-backed best practices, guidance, training and world-class curricula. Previous solver Century Tech has already seen impressive and promising results using similar technology. Notably, the approach works as well for disadvantaged populations as for middle-class students, effectively reducing learning gaps and increasing opportunity.
• Upskilling and Reskilling. By 2030, one billion people will need to be retrained. AT&T is spending one billion dollars retraining half its workforce. Given the right vehicle, there is an opportunity for corporate spending to subsidize the education of the less fortunate; a platform such as the one proposed here might be such a vehicle. 

By 2030, there will be one billion new entrants coming into formal education and another billion (a third of the global workforce) in need to be retrained. To keep up with growing demand, the traditional education system would need one million new teachers every year and a new university built every four weeks (EdTechXEurope, 2018). The world is facing impossible scaling challenges; education in the 21st century is going digital. 

Fortunately, advances in Neuroscience, Deep Learning and Learning Science are coming to the rescue. It is now possible to leverage the same technology that recommender algorithms in platforms like YouTube and Facebook use to understand at a granular level how an individual student works, and then use that knowledge to maximize his/her potential. This is what a learning experience may look like very soon (and what we’re trying to accomplish):

1. AI-powered ongoing diagnostics. The platform measures engagement, response times, preferred subjects and weak points. All of a student’s preferences, needs and drives, every interaction, in real time.
2. Micro-modular, flexible curricula. Skills and subjects are translated into bite-sized knowledge points. Think curricular legos; From national decision-makers and workforce demand leaders to local communities and parents, everybody can participate in the construction of “knowledge landscapes”. Then let all learners explore, guided by a teacher as smart as Socrates.
3. Personalized learning paths. Prohibitively time-consuming for a teacher, natural for an AI tutoring system. Curricula are the terrain; the platform is the GPS. Incrementally better with use, the platform will adapt every aspect of content to what works best for each individual student, construct path choices and let the user “pull threads” according to curiosity and intrinsic motivations.
4. Meta-cognitive focus. Here’s what differentiates us from existing approaches. It’s not all about increased proficiency levels. Yes, we want better PISA and TIMSS scores, but what we are really after is adaptive problem solving and self-regulation, empathy and theory of mind, open inquiry and introspection. Learning how to learn. Not teaching creativity, but letting it flourish.  

Using this framework, all types of knowledge and skills can potentially be represented as user journeys. This presents opportunities for everybody to create learning experiences which may then be subject to a community-driven evolutionary process. The platform may drive optimizations to goal-directed group formation in the same way jobs apps match people together, for social benefit initiatives.

Mauritius Minister of Education, L. D. Dookun-Luchoomun, sums up the whole recipe: “Children are curious by nature. We need to ensure that the system does not dampen that curiosity. We need to ensure that when we bring them information, we do so in such a way that they ask for more. Very often, it’s by giving the right information that you trigger curiosity. It’s not by giving too much of information. You need to allow the child to develop his/her inquisitive mind, to ask for more, and to make him learn the answers for himself.”

Unfortunately, finding the right information for each learner in the current system is just not feasible. Teachers are already overworked as it is. By trying to teach everybody in the same way, the system does end up dampening curiosity. Fortunately, Deep Learning technologies can find just the right information for each individual student. The application of this technology is the main innovation, but differentiation will come from which parameters the underlying neural networks optimize for. That's what will ultimately determine the function and utility of implementations. While most similar educational technologies focus on efficiency, speed and proficiency levels, we believe learning how to learn, "developing an inquisitive mind", comes from finding and fueling curiosity triggers. And those rarely follow the curriculum, but we can guide them along. Learners will arrive at knowledge and skills from many directions, following intrinsic motivations. 

This is a pre-development solution. During the first year we will focus our efforts on initial prototyping and testing. After proof of concept, we will seek partnerships with Mexican schools to further study the effectiveness of our solution. Last year in Mexico, 12 million people used the Internet for work training, 22 million go online for school work, and a further 67 million (83% of internet users) searched for terms related to education* (INEGI, 2019). The Internet and education go hand in hand, and demand is on the rise. Initially, we will explore reaching out to these populations:

• There are 66 million smartphone users with internet access, 31 million in the lower-middle side of the income distribution. By targeting a prevalent pain point, poverty and access to employment, we are confident we can reach a significant percentage of this population. We will also seek government support in providing incentives and validation of future vocational training pathways.
• 37.7% of Educación básica schools (ISCED 1-2) and 50% of Media Superior (ISCED 3-4) have Internet access (DGPPyEE, 2019). We will seek to partner with education authorities and institutions for support and sponsorship of research and pilot programs.
• We will also offer creation of custom training and onboarding programs for companies and corporates. By targeting ongoing ESG alignment efforts, we will seek to create branded community improvement workshops, boosting PR and brand engagement. 

*Survey terms: "education, research, homework, online courses, training, tutorials".

Goals and key focus points for the first year

• Research. Conduct evidence review, develop design principles.
  • Interests
    • Technical
      • Knowledge Space Theory, Adaptive Learning Systems, Intelligent Tutoring Systems, Curriculum Content Mapping, Stealth Assessment, Brain Points, Teaching at the Right Level, Educational Data Mining, RCTs in education, Natural Language Processing, Deep Learning in education, Meta-cognition, Differential Privacy
    • Scope
      • EdTech in Latin America, Education2030, EduTechX global, Learning Technologies 2020, Financing Ecosystem for EdTech
  • Existing approaches (revenue models, content, use cases)
    • SkillsFuture
    • KnowledgeFox
    • CenturyTech
    • SquirrelAI
    • Sana Labs
  • What type of content works best now?
• Connect. Reach out to organizations in relevant fields. Explore avenues for collaboration, seek out guidance and input (at relevant stages of project maturity).
  • Observatorio de Innovación Educativa
  • Interaction Design Foundation 
  • C Minds
  • IA2030Mx
  • Aflatoun International
  • ANUIES
  • Education Endowment Foundation
  • What Works Network
  • J-PAL
  • IIEP-UNESCO
  • International Finance Facility for Education
  • Congreso Internacional de Innovación Educativa
  • Behavioral Insights Team
  • Some Key People 
    • Dra. Claudia Marina Vicario, Red LaTE (Educational technologies network)
    • Claire Field, HolonIQ (Education market intelligence)
    • Dr. Phil Lambert, Education expert
    • Cristina Martinez Pinto, C Minds, IA2030Mx
•  Make. Prototype basic interaction and minimum viable product
  • Explore possibilities for rapid prototyping and testing.
  • Seek collaborations with tech-enabled schools for randomized controlled trial studies in small representative sample groups.
  • Iterate minimum viable product through study collaborations.
  • Explore incentive structures for adoption
  • Explore alternative assessment methods

After achieving proof of concept, we will explore partnership potential for sponsored community initiatives: applying this technology to local improvement workshops as a way to gauge reception and encourage adoption.

Intangibles. Initially, we will draw inspiration from existing similar applications. However, our main focus, "Learning how to learn" and development of meta-cognitive abilities is not at all straight-forward. Existing solutions rely on (for instance) optimizing speed in arriving at "right answers" (Squirrel AI has been criticized for not addressing learning how to learn but standardizing quicker). We seek to reward smart strategy, adaptive problem-solving, right mindset and persistence (after O'Rourke). Implementing machine diagnosis of these less tangible behaviors, however, might prove to be the main challenge to our approach.

Natural Language. Strong engagement with Intelligent Tutoring Systems will require some degree of "mirroring" between the learner and a guiding agent. A conversational engine (chatbot) will be necessary to monitor (for instance) the evolution of strategy in problem-solving attempts. We envision the agent to be able to discuss (in as close to an emotionally-intelligent manner as possible) why the user is tackling a problem in a given way, so as to provide contextually relevant feedback. Enabling "soft mistakes", a way for the learner to switch strategy without becoming discouraged and quitting, will be another significant technical challenge in the development of our solution. 

Privacy. A fundamental consideration will be user data. "Anonymized data isn't" (Cynthia Dwork, Harvard). De-identified data has been shown to be easily re-identifiable. This makes solutions like aggregated anonymized datasets to fall short of the goal. We will conduct further research on implementing more robust alternatives such as Dwork's Differential Privacy.

The technical challenges outlined above are only a few of the many that will arise from the application of new and emerging technologies. Fortunately, Deep Learning is also the scientific field behind most AI-related technological innovation happening now. This means accelerated research and increased access to customizable solutions. Many services offer optimized, pre-trained, ready-to-implement neural networks. This translates to reduced time and costs of development. Additionally, the possibility of geographically distributed development teams provides an alternative to local talent recruitment (more research is necessary since both options have significant trade-offs).

We recognize the creation of a strongly aligned development team will be the core concern during initial stages. Overcoming emerging challenges (and all three barriers mentioned) will depend almost exclusively on this factor.

According to the GSMA, 4G penetration in Latin America will reach 67% by 2025. That's 484 million unique subscribers ready to benefit from internet access for education, up from 418 million in 2018. Mobile MicroLearning will not replace traditional education anytime soon, but it can support all stakeholders in significant ways. Educational Data Mining (EDM) presents a great opportunity for policy-makers to improve resource efficiency and allocation by bringing real-time data into the decision-making process. Teachers can see immediate workload reductions through automated administrative tasks and students get to learn in ways that work best for them. Community members will develop their own "learning how to learn"  methods and will be encouraged to share their progress through local improvement workshops. Finally, companies will be able to easily create their own training programs and leverage personalized education to boost productivity.

We intend to pursue outreach campaigns for all five of these groups. School subscriptions will include teacher training and curriculum access to world-class sources. We will explore ways to get one side of the market to pay for the other; custom training programs for companies will provide the revenue that will allow free access for low income populations. "Pre-trained" users in high demand fields will then be offered pathways into employment with partner organizations. Additionally, this will drive better accuracy in employee-employer fit. Learning goes hand in hand with productivity, by enabling distributed learning and vocational pathways, we seek to create alternative, flexible modes of income generation.

Sergio>

Currently it's just me. I've been researching this project for about a year as part of a Master's application project. The program that I believe would be the best fit is called Media, Arts and Sciences, at the MIT Media Lab. They have a very hands-on approach to project building and a strong impact focus. I would like to pursue this project as the main focus of my Master's dissertation under Professor Alex Pentland, who wrote two amazing books on human behavior (Honest Signals) and social dynamics (Social Physics); the latter proposes a quantitative model of tie-strength and influence in social networks, with strong predictive capabilities. Prof. Pentland is also involved in a movement called "The new deal on data", which proposes the creation of Data Trusts, a way for individuals to own their personal data and give out (and withdraw) permissions. He has also worked on modeling dynamic influence in human interaction through the use of wearables and mobile phones. One paper from his group, called "The Wisdom of the Network: How Adaptive Networks promote Collective Intelligence" really inspired me to pursue this project. One of the paper's coauthors, Alejandro Noriega Campero, who is developing a mobile camera -based solution to Diabetes diagnosis in Mexico, wrote his doctoral thesis on "Human and Artificial Intelligence in Decision Systems for Social Development". I would like to join these ideas in exploring scalable technologies for social well-being.

Sergio>

I'm currently taking 0.SolveX: Business and Impact Planning for Social Enterprises on edX. I'm also in an international mobility program sponsored by FUNED (Fundación Mexicana para la Educación).

We will pursue a subscription-based model. Our main value proposition, an enhanced learning experience, is the same for all users (There is also value added for the teacher segment which will come from reduced workloads). However, user journeys will differ depending on the client organization.

The main source of revenue is business programs: there are potential applications even in recruitment. Since one of the main features of the platform is getting to know individual learning styles, we will explore use cases in recruitment and onboarding. Further applications will be explored in upskilling and reskilling, compliance training, product training, etc. Knowledge Fox has already seen enormous success with clients such as Credit Suisse, Samsung, and the Vienna International Airport. The main purpose of targeting this market is to be able to subsidize low income and disadvantaged populations. As a platform business, we intend to bring market sides together by creating vocational training programs which link together business partners and individual learners seeking employment.

An additional source of revenue is educational institutions. Also a subscription-based model, the service will include importing and customizing curricula. The platform will also be able to diagnose teachers' learning styles and provide advice on the type of assistance that works best for each pupil, improving teacher-student relationships and boosting theory of mind development (of benefit to both parties).

Finally, we will explore a freemium model for individual users. In order to encourage widespread adoption and achieve scale, the smartphone application's main functionality needs to be free.

Some of the organizations we've mentioned on the first year plan are major international bodies. Support from Solve and TPrize sponsors would really help open a lot of these doors. Even though this is a really early stage solution, we believe it has a lot of potential, as seen in existing solutions. EdTech's relevance is being felt right now because of the pandemic. This is why we're confident enough in its impact potential to submit a relatively undercooked solution. 

The main interest in submitting this solution is in potentially connecting with the Media Lab in order to actually make this solution happen. [Sergio> I would like to add a personal note. A major short-term goal in my life plan is to study a Master's (the Media Lab's Media Arts and Sciences program) under Professor Alex Pentland and his Human Dynamics group. This solution would also be the main project. If the TPrize is the way to help make that happen, I'm making the wish.]

Other potentially catalytic connections would be the Abdul Latif Jameel Poverty Action Lab (and last year's Economics Nobel laureates Esther Duflo and Abhijit Banerjee, we're drawing inspiration from their RCT-driven programs), Aflatoun International, the What Works Network, IIEP, and locally, leveraging the connection with Tec de Monterrey's Observatorio to recruit the development team. Also relevant in the national scale would be connecting with the IA2030Mx Initiative and its research group on education. Finally, the International Finance Facility for Education and the Education Endowment Foundation.

Observatorio de Innovación Educativa. Their scope will be vital in understanding the EdTech and education landscape in Mexico and Latin America. We would also be interested in pursuing a partnership for developing and testing this solution.

IA2030Mx. The initiative exploring IA in Mexico. Will be important in understanding the landscape in all things AI-related at the national level.

Education Endowment Foundation. A work centre of the What Works Network under the Cabinet Office (UK), an initiative that aims to improve government through use of evidence. Also related to the Behavioral Insights Team (The Nudge Unit, directed by Dr. David Halpern--whose work is also related to Alex Pentland's). Without a doubt, behavioral nudging will be a main component of social policy in the near future.

ANUIES. Mexico's National Association of Higher Education Institutions. In their Visión y Acción 2030, they call for an integrated nation-wide system. We envision every university being able to create and make available new generation content through our platform. This would help institutions have a broader reach and individuals to have access without needing to relocate. Although this may be a significant challenge, nation-wide validation and support by this entity would be key in achieving something like what SkillsFuture has done in Singapore. 

UNOi. Already working on a Learning Management System and making a great impact through education innovation in Latin America. We would be interested in exploring avenues for collaboration and learning from their best practices.