Presenting at the Solve For X event

The Moonshot Education Project, is a community that's developing a highly scalable adaptive learning platform that can cost-effectively be distributed to all children who need it. A platform which can work in a formal supervised mode as well as an informal self-directed mode, in order to reach those children who cannot get quality education and those who will never go to school. It is being designed for the more than 250 million children that cannot get a basic education and the more than 50 million that will never go to school.

I recently did a short presentation about this where I touched on the 6 key technologies we are using. I've described these in more detail by writing the below real world example of how I help my daughter learn, and how that translates to some of the things we are doing with the platform. 


A working example with my Daughter


1. A Natural Interface to learning

I have 'Work time' with my daughter, a one to two hour chunk of time every few days in which we play games together, read books, do projects, and many other things. We call it 'work time' however it's more accurately described as a time for learning and discovering new things.

At the end of the previous work time session —in preparation for the next session— I ask her questions about what she wants to learn about. She responds with something like: can I be a fairy, can I live in a treehouse? —She is 4 :). Sometimes these questions can easily be answered quickly, and I do so then and there. But what I am really looking for, are more detailed questions, that I can build content around. This Q&A interaction is quite natural, and if you have been around young, inquisitive children for any amount of time, you will know how well they respond to these sessions. They fire so many questions at you in rapid succession that sometimes it's hard to keep up, often asking the same questions many times over, but each time —given what was just spoken about before— would be answered slightly different than before.

How it relates to the platform:
In this part of the example I illustrate some of what we are doing. A key approach will be to provide the children a beautiful, natural interface to this platform, one that the child can communicate with, ask questions to, and over time, in sync with the recommendation engine —explained more fully later, offer increasingly sophisticated guidance. This 'chat bot' will use Wikipedia type information as its corpus of knowledge, and be session aware, remembering what was discussed before and using it for future conversations. Our friends at Princeton AI are helping us by developing this key part on the platform. 

2. Personalized, Contextual Learning

With these questions in mind I spend the next few days casually searching for appropriate content about those questions. Given that I am quite close to my kids, I know what her other interests are, I know what type of content she prefers, what level she is at, what style she best relates to. I also know what she is learning in pre-kindergarten and what lessons she is learning in dance class and many other things she is doing. I know for example she does not just like any fairies, she really likes tinkerbell but dislikes the barbie fairy. I also know that when she talks about tree houses, she is talking about those found in the books we are currently reading together. 

How it relates to the platform:
This part of the example is relating to the on-device data store of the child's personal data, her progress, interests, usage data and all the information learned from the child and how it relates to her learning. The work we are doing here is quite similar to that being done by many others, including Carney Labs and their MARI project. In our case this data never leaves the control of the child, and remains local to the device.

3. Finding the right content

With a deep understanding of my daughter and her questions in mind, I can now create a search and find what she is looking for. Over time I have come to find numerous repositories of content, from the various educational sites to the likes of Amazon and Google Play.  I end up with a few ebooks, a game or two, some digital coloring in pictures, a movie maybe. I also find a few articles and some information about those topics from wikipedia. 

How it relates to the platform:
This function in the platform will be conducted by the search engine that looks up the meta data associated with a piece of content. We will be using the Learning Resource Metadata Initiative LRMI or similar initiatives as the basis of this work. In addition, to ensure we get usage data from the content, our content framework standard will include the use of the Experience API (AKA Tin-Can API) to pull this session data and feed it back to the personal data store, for later analysis and use.

4. Social, Collaborative and Peer-to-Peer Learning

In addition, because I know my daughters friends, I can also find friends that have the same questions, share the same interests, or maybe can offer some advice. lets say for example her brother —my two year old son— liked treehouses, or if her friend Alice was just as interested in fairies as she is. Giving them a social task, project or group work could be arranged. They could for example build a mini treehouse out of blocks and the one that survives the longest with my two year old running around could win the challenge.

How it relates to the Platform:

The Mesh network, which in addition to providing a mechanism to connect the devices locally, is the backbone of much of the educational approach we are taking. We know that these children will likely not have formal teachers, but we also know that these children will, and do, teach each other. And so key to our approach is to provide the ability for the platform and child to search for peers to learn from or who the child should go teach. We are also working closely with Sugata Mitra and his School in the Cloud team to create engaging, dynamic content for these Self Organized Learning Environments (SOELs)


5. Non-prescriptive, personalized learning plans: 

By this stage I have already filtered out a lot of content that is definitely not appropriate, but I still have a fair deal. I put this in a list, weight the content according to parameters I know about her, and then whatever pedagogy logic I can to it and maybe even relate it to her current school curriculum —if appropriate. 

In the next session we then go through that list of content and activities. Some are individually listed and some are listed as part of a group that must be done as a unit. The reason for some groups is because some have prerequisite logic to it that I base on pedagogy or a game mechanic that I want to include to encourage her to do something she may not want to do. As we do that selection I learn much more about her and my assumptions about her are continually adjusted, especially if she picks to do something that I thought would not be at the top of the list.  

How It relates to the Platform:
The above filtering, sorting and structuring of content is not contained within one technology area but rather a combination of various areas. The search algorithm uses the Personal Data Store to search the meta data of all the content stored in the library. As part of our initial library we are focusing on partnerships and Open educational resources.

6. Pedagogy, Curriculum and data, balanced by Feedback

I approach the search for content in a pretty robotic way; given the information I know about my daughter, and what I happen to know about pedagogy best practice and her school's curriculum. I use those as inputs into the 'search algorithm'. However I am also keenly aware that I do not want to put her in a box by only providing her what I think she wants and needs. I also provide some curveball content and see if she works well with that and allow her to let me know when I get things wrong. Maybe I totally misunderstood what she was interested in, or maybe her interests have changed. I see this as the supervised learning loop, where together we continually evaluate the weights and parameters that I use within my searches.

How it relates to the platform: 
A child can choose to subscribe to one or many curriculums and pedagogy plans and have a long list of parameters that all influence how the recommendation engine searches and recommends the next course of action. However the child or educator will be able to review and adjust the data used by the engine locally, and thereby train it to adapt to the unique requirements of the child.


There is much else to the platform and this is explained in more detail on the site and technology page. You can also view a draft prototype on our prototype page

If you want to Help this project please contact me directly here, or view the Join Us page and let us know how you can help.