Marni Hybrid Learning

Optimal learning environment for all - anywhere in the world:

Marni, ("Teach" or "Learn" in Kaurna, which is the language of the original custodians of the Adelaide plains), offers a solution to Bloom's 2 sigma problem; namely, that personal tutoring is highly effective, but too costly for widespread implementation and cannot easily be replicated in the classroom. Here, AI is harnessed to deal with routine student assistance, which leaves an experienced tutor to focus their attention on the high value-added aspects of their craft, such as detailed formative assessment and personalised feedback and forward planning with the student. Due to being delivered online, Marni gets the best tutors to the most needy students, no matter where either of them are, so long as they have access to a basic smartphone and on-demand internet connectivity. The intermediate AI assistant provides instant tutor-trained help and encouragement and is available 24/7, irrespective of student and tutors being based in different time zones. In future, Marni aims to hire local teachers and tutors, to help them earn extra income, as well as help with their professional development.

Designed to be engaging and effective for all types of learners:

Again, by delegating routine tasks to the AI engine, the most skilled and captivating tutors can focus on the areas that produce the greatest learning and personal gains for their students. Principally, this is engaging presentation, formative assessment, personalised feedback and forward planning with the student. Once each week-long project is started, the student submits their first draft or section of work to the tutor, who assesses the specific learning needs and provides feedback, advice and encouragement to the student, as well as guidance to the language model for ongoing support. The AI easily deals with technical questions, such as grammar and format conventions and follows support guidance, as programmed by the tutor. In early trials, the quality of AI support was remarkably good and adhered well to the initial input from the tutor, including specific support for neurodivergent learners and relatively inexperienced English users.

Provides opportunities to develop inter-personal and future-proof skills:

Embedding AI into the system teaches students to assess the inherent strengths and weaknesses of human and machine interactions respectively. Interactive AI systems are part of the future and Marni allows the tutor and student to discuss and learn how to make best use of AI tools. It also allows the tutor to design more complex or obtuse learning opportunities for the students, knowing that they have AI assistance to deal with complex and highly specific interactions efficiently. For example, using a story mountain, (story curve etc), to write a fictional recount of events in the style of a Victorian newspaper article.

Marni is an AI-assisted online tutoring system that amplifies learning outcomes by simulating 1:1 tutoring and provides the greatest number of students access to top quality educators, irrespective of their location and financial status. In the first two trials, a series of eight weekly sessions were themed around authors with a London connection and a juxtaposed scientific concept, which virtually necessitated original ideas for the writing project.

A - Each week, the tutor introduces a topic via a pre-recorded video, incorporating engaging true stories and props and links to background materials to read and watch. For example, Daniel Defoe's burial site inspired a discussion on Robinson Crusoe and the concept of human colonisation on Mars. An original writing project relating the two is outlined.

B - The following day, students receive a second video focusing on technical instructions tailored to the learning objective, such as genre features or the use of a specific planning templates.

C - Students begin the project and submit their first draft/section, which the tutor reviews, providing personalized feedback and prompts for subsequent AI-generated feedback.

D - Students can submit drafts and further sections as needed and request AI assistance. The system adapts to individual needs, while respecting the tutor's learning support guidance. Students who need help more often are free to do so and those who don't, are encouraged to extend themselves.

E - At the end of the week, the final product is reviewed by the tutor and a realistic assessment of the student's progress is shared, along with suggestions on how to make even better use of the AI-assisted workflow in the next project.

F - Student feedback on the trials shows that they tend to forget that the intermediate feedback is AI-generated, as it mirrors the tutor's manner and concentrates on their tutor-assessed needs. Additionally, the 24/7 availability of the model reduces anxiety about writing. Students appreciate the thought-provoking video introducing the concept, as well as the clear technical instructions, which they can watch as often as they like and the ability to pace the project steps to suit their ideal pace, allowing time for reflection and improvement as the work progresses.

G - Using a 50% work rate, (allowing for other in-person tutoring and making of the videos), it was possible to accommodate over 100 students, with staggered start days. This was maintained for eight consecutive weeks, in two separate trials. Once the videos have been created, they can be reused indefinitely and build up to a comprehensive library.

H - In smaller scale trials, other text based subjects, such as social and general science topics were successfully tackled. The current model does not do so well with mathematics, but with more specific training and general model improvements, it would be viable. The two issues it currently suffers from are that is often does not always use conventionally taught mathematical methods and it makes logic errors.

General

Marni allows tutors to reach the greatest number of students with learning quality similar to 1:1 tutoring. Because of its efficiency, it can be extremely cost effective, allowing students who would never otherwise be able to afford such attention from an experienced educator. The system was originally conceived to help me help children in the Philippines, during my spare time as a contracted private tutor. Below, I will discuss it in terms of that country, but it could be used anywhere that could benefit from affordable accelerated learning. I intend to run a further trial with indigenous American families later this year, in Wisconsin.

Pro bono after school tuition

Because the barrier to entry is so low, (a basic smartphone and occasional internet connection), Marni lends itself well to families in developing countries. The first two trials were based on typical western school students in the 8 to 12 age range. Although the children all had English as a first language, there was a range of specific learning requirements present and the bespoke nature of the system helped them make great progress. Currently, we are recruiting for a trial in the Philippines, which does use English as a language of instruction, but is not widely spoken outside of schools. Excellent English is considered a key gateway skill in many professions in the country and overseas. Even having just one well educated person can lift an entire extended family out of poverty. Because of the efficiency of the Marni system, any employed teacher could take on quite a number of students after hours and because of its 24/7 online nature, they could be anywhere in the world.

Non attending school-aged children

Unfortunately, the Philippines has a small, but persistent cohort of children who need to work to support their family. If they cannot afford to attend school, then it would be possible to teach them some key skills after hours, possibly with the aid of a barangay hall or office or non-government organisation. That might provide students with space and/or wifi for a parent's smartphone. Given that tailored tutoring is the fastest way to learn, it should be possible to keep the motivation of such students relatively buoyant. 

Extra income for teachers in developing countries

Teaching in the Philippines is a rather poorly paid profession. Local teachers could use the system to take on extra private students for a small fee to supplement their income. If this were administered as part of a tutor network, that could also allow for professional development opportunities, using the same system and more experienced tutors as their mentors.

Home schooling

Some younger writers in the previous trials were paired up with siblings and parents. So long as the writers identified themselves as a pair, the system had no difficulty accommodating them. An additional benefit for home schooling parents was that they saw first hand, a different manner of educational questioning and instruction. They claimed it helped them learn to be better home schoolers.

Background

I was a specialist freelance teacher in many challenging inner London schools over many years and learned how to maximise all types of accelerated learning programs to get classes back on track. For the last 15 years, I have been a full time international private tutor. Between those two vocations, I have a huge amount of experience with both ultra-high achievers, as well as low achievers. Both extremes have specific requirements, not to mention everyone in between. The art of good group instruction is offering a ladder/scaffold to those who need support to raise their level, as well as open-ended extension for those that are ready to take it to the next level. I learned those skills through my years of teaching and tutoring. 

I have made connections within the Australian indigenous community through my training there. I was a specialist trained Catholic primary school teacher in London from about 2000 to 2008 and made many long term connections in the Philippines from the community and the local church. I have fostered eight African children. I have many educational contacts in the US, from my many tutoring contracts there, including state and national leaders in education committees and a chair of the board of regents at a well known university.

2023 USA / Philippines

In 2022 / 2023, I developed a tech-based work solution for Filipinos and won the Network For Teaching Entrepreneurship World Series of Innovation with it.

2024 Philippines

Shortly, we will be commencing the first trial with non English mother tongue families. I have educational and local government contacts on each of the main islands, who will assist with any social difficulties that may arise, such as internet access. However, I intend to run the program from London, where I am continuing to make further video learning resources, using interesting historical locations and some of London's amazing museums.

2024 - 2025 USA

Later this year, I will be based in the USA, where I will trial the system with indigenous American families, where I plan to further train the GPT to meet the needs of the families I recruit there. I will have plenty of spare time to work on developing Marni.

2025 - 2026 Philippines / Australia?

After the USA tutoring contract is finished, I intend to devote all my time to Marni. As an Australian qualified and registered teacher, I can legally teach and tutor there. There is a huge need for additional bespoke learning in the Aboriginal community; especially in the remote regions. If the opportunity to continue developing Marni there arises, I will consider taking it. Otherwise, I would also consider basing myself in the Philippines, as I qualify for residency there. I can see a modest amount of development money going a long way in that country - and teachers tend to have better than average English skills. A number of wealthy parents of former students have expressed an interest in supporting the program and are already significant education philanthropists.

The model is up and running and has had over 200 users over an eight week program. A further pilot will be launching next month, in which the system will be tested with English second language users. It may need further training for that. I am also not satisfied with its performance with mathematics and  physics calculations. It makes too many illogical errors. I have tested it privately, (on myself), in German and Spanish, with good results, but it did not do well with Russian grammar.

I personally do not consider the product ready for launch, although it would not take a huge amount of work to get there.

The main areas of help that I need are Technical and marketing. Unfortunately, my programming is hopelessly out of date. Because of my background in computational fluid dynamics of meteorology and oceanography, I am stuck with Fortran as my language, which is not suitable. Also, I am well aware that, if and when the time comes to roll Tendi out to an open market, I would be out of my comfort zone. Perhaps it is something I could learn, but I'm not sure. If I can get those two pieces squared away, I am confident I can get backing from one of my many former tutoring clients. I am planning on being in Wisconsin for the next year, so should be close enough to the action to work with or alongside partners to strengthen those skills.

Marni is not an especially innovative idea. It just answers a question that not many people have been asking correctly. How can AI be used to simulate 1:1 tutoring? Instead, most people are asking AI how to do something. Just learning by asking a language model questions is not enough. What is missing from that solution is the performance and danger aspect. When a teacher or tutor sits with their student(s) and looks into their eyes and asks well pitched and thought provoking questions, it puts the student on edge. It presses them to reach above where they are at - and if it doesn't work out, they run the risk of feeling silly or falling flat on their face. Of course, a skilled educator judges the fall, like a gymnastics coach and breaks it. When that happens regularly, the student learns to trust their skill more and takes bigger apparent risks, in relative safety. That's how they improve.

Marni, (which means, "Learn"), is about forcing meaningful and original dialogue, by juxtaposing two unrelated concepts together, (an author and a science concept, in the example given elsewhere), and then assisting the tutor to cope with what could end up being a huge amount of work, by using AI to do some of the more straightforward support work with the students. The four most important points of contact are preserved: the big picture rhetorical question video that the featured author might ask if they were present, the distilled technical instruction video with as much or as little extra help for the student to use, the initial tutor assessment and advice - and the final conclusions to the success of the project and actionable strategy for improvement. The relatively incidental intermediate questions can be handled by the pre-trained language model.

Benjamin Bloom's 2 Sigma Problem can finally be solved, by judicious use of AI. I don't suggest that Marni is as good as an in-person 1:1 tutor, like for like, because whatever a tutor can do using this system, can be done even better face to face. However, it's Bloom's 2 sigma problem. The results that a highly experienced tutor can achieve in person is more like a 3 sigma improvement. That is how great classical musicians are trained. It is a regime designed to achieve exceptional standards way beyond what would ever normally be considered possible. Even if that is not possible, using Marni, I have been seeing a 2 sigma, or thereabouts improvement in the students who use it. That, for almost anybody is a life changing amount of improvement - especially when it is compounded over a length of time. Think how that would change the prospects of a Philippine squatter community, where few of the living generation learned to read. Marni could allow one tutor to work with 200 children, 5 minutes per week each x 2 sessions = 2,000 mins < 7 hours per day. 

Marni's model is reminiscent of a program I implemented about 20 years ago, targeting a community where spoken language was poorly used and undervalued. It was observed that the children were not responding to classroom teaching of reading and writing, indicating a missing link. By recording the children periodically and showing them their progress, speech became a source of prestige, driving out playground language.

With the first two Marni trials, parents reported that students enjoyed their apparently personal tutor, who asked searching questions and provided highly focused attention to them and their learning. They understood that their private tutor would review their work and their use of AI support. Here's the theory of change:

Step 1 - Student receives a fun challenge from an admired person. 

Step 2 - Tutor helpfully makes the learning objective easy to digest. 

Step 3 - Student aims to impress the tutor with originality. 

Step 4 - Tutor describes the strengths of the work & possible improvements. 

Step 5 - Student rehearses/improves with a safe AI wingman. 

Step 6 - Tutor and student celebrate the product and agree future strategies.

Each step leads directly to the next, minimising gaps. Students are always free to edit given examples of the genre of writing set, to make them their own. There's no upper limit to quality, and many students go above and beyond. With more challenging writing, they often overdo it and then moderate themselves in the next section or iteration.

The project has been in an experimental stage since last July. A lot of the early iteration was unmeasured and purely judged intuitively. However, the only suitable objective measure for an amalgamation of accelerated learning techniques and a claim to have a solution to Bloom's problem must be a two standard deviation improvement in scores. 

Impact Goals:

1. Enhance language proficiency and communication skills, particularly in underserved communities.
2. Increase student engagement and enthusiasm for learning.
3. Foster a sense of achievement and pride in academic work.
4. Reduce educational disparities and enhance learning outcomes for all students.

Measurement and Progress Indicators:

1. Language Proficiency:

   • Pre- and post-tutoring assessments to measure improvement in language proficiency.
   • Percentage increase in the quality and complexity of spoken and written language.
   • Performance on language proficiency tests aligned with the Common European Framework of Reference for Languages (CEFR).

2. Engagement and Enthusiasm for Learning:

   • Attendance and participation rates in Tendi sessions.
   • Student feedback on enjoyment and motivation levels in Tendi sessions, measured through surveys.
   • Frequency of interaction and completion of tasks within the Tendi platform.

3. Sense of Achievement and Pride:

   • Percentage of students reporting increased sense of achievement and pride in their academic work, measured through surveys.
   • Number of students voluntarily sharing their work with peers or in public forums.
   • Students' self-assessment of their progress and skills development.

4. Reduction of Educational Disparities:

   • Comparison of pre- and post-tutoring assessments to evaluate reduction of the achievement gap.
   • Percentage of students from underserved communities showing improvement in language proficiency and academic performance.
   • Percentage increase in literacy rates among students from disadvantaged backgrounds.

These indicators will be regularly assessed and analysed to measure progress towards the impact goals, ensuring that the solution makes a transformative impact on students' lives and effectively addresses educational disparities.

In the longer term, I would also like to see how school graduation rates are impacted in the Philippines, where the system typically does not automatically push students to graduation, unless their grades warrant it.

Marni uses a fine tuned GPT, that has been trained to use educational language and instructions in a manner directed by the tutor. There are currently a number of issues with the way it works, especially in terms of usability for different tutors and unreliability in mathematics, physics calculations and certain language grammar systems. It is also yet to be proven with students who are less experienced in English.

Full time: 1

Consultant: 3

Contractors: 4

Part time since July 2023, then full time through 2024.

At Marni, we are committed to fostering a diverse, inclusive, and supportive environment for all team members, regardless of their background or location. While the project is currently run mostly by myself, we have four Indonesian IT contractors, and we intend to engage local tutors as the project progresses. Indeed local hiring is a key strategy to make Marni affordable for local students. A corollary to local hiring is global professional development, so as provide the best opportunities for local tutors to progress professionally and learn more about current best practise. Here's how we will ensure diversity, minimise barriers to opportunity, and provide a welcoming environment:

1. Diverse Hiring Practises:

   • We actively seek out diverse talent during the hiring process, ensuring that our team represents a variety of backgrounds, perspectives, and cultures.
   • We prioritise hiring local tutors, providing economic opportunities and creating a sense of ownership within the community.

2. Cultural Sensitivity and Inclusion:

   • We foster an environment where every team member feels valued, respected, and included.
   • We provide cultural sensitivity training to ensure that everyone understands and appreciates the diverse perspectives within our team.

3. Language and Communication:

   • We ensure clear and effective communication by providing language support and resources to accommodate the diverse linguistic backgrounds of our team members.
   • We provide language training and translation services to ensure that every team member can effectively contribute and understand.

4. Equal Opportunities:

   • We provide equal opportunities for growth and development, regardless of background or location.
   • We offer training, mentorship, and resources to all team members to ensure they have the tools and support needed to succeed.

5. Inclusive Decision-Making:

   • We involve all suitably qualified team members in decision-making processes, encouraging diverse perspectives and ideas.
   • We create an open and inclusive culture where everyone's voice is heard and valued.

6. Supportive Environment:

   • We prioritise the mental health and well-being of our team members, providing support and resources as needed.
   • We offer flexible working arrangements to accommodate different schedules and personal needs.

By embracing diversity, minimising barriers, and fostering a welcoming environment, we believe that Marni will thrive as an inclusive and supportive community where all team members can reach their full potential.

I have not yet decided on this. At the moment, the project is self funded and there is the option for it to be entirely not for profit or a hybrid for profit.

I have not yet decided on this. At the moment, the project is self funded and there is the option for it to be entirely not for profit or a hybrid for profit.

As a long time elite tutor for UUHNW families, I have a number of experienced large scale education philanthropist connections. However, as stated above, I am not certain which option to choose and even whether it should start out in one mode and switch to another model after. Ultimately, I would like Marni to be a potential local source of income for teachers and tutors in developing countries, so at some stage, there would have to be at least some element of profit involved.