Background
In this blog, I share the approach of the Erdos research labs. We now have a logo and have people who have already signed up as foundation members.
If you are interested in being a foundation member(it has a one off fee) please message me.?
I am explaining the methodology of Erdos below in terms of my rationale and inspirations, especially the Semantic tree method of learning - most recently popularised by Elon Musk as a method of learning.?
Two interesting contexts which provide a backdrop:?
1) Much has been said about Nvidia founder’s views challenging the need for coding skills. I think he is right. We are entering a new area
2) Less well known? but equally intriguing is a report which says that for the first time, Google reportedly spends more on compute than on people.?
Relating that to the Nvidia post, we could say that development is going from a creative skill to a productionised / manufacturing-like process. That has profound consequences but also opportunities for domain experts.
With this background, I have used these ideas in a number of areas - both for learning and teaching.?
Below we discuss how these ideas apply to the learning of AI.
The overall idea is to learn AI in a hands-on?
Learning AI is not easy and learners face some unique challenges
The question is: What happens next after the AI course? How exactly do you gain experience after learning in a course?
Essentially, to learn AI, we need three things:
1) A real problem to work with
2) The problem should be relevant to your experience
3) Working with people who know more than you with respect to AI.?
1 . Background
- The Erdos Research labs initiative is targeted to non developers / domain experts who are familiar with the tech industry. Typically, we work with someone who is a domain expert interested in AI / has coded at some point in their life / has learnt some? coding from other online sources
- We use generative AI as a co-pilot i.e. to generate code.??
- The Erdos approach is based on personal tutoring
- While we are a community - we are not a learning cohort in the conventional sense of other courses. We are more of a mentoring program.?
- Overall, we encourage you to share and build your own brand ? ? ? ?????????????????????????????????????????????????????????????????????????
2. Objective
- Design a learning project?
- Build the project using the methodology?
- Reflective learning - put the project into context of wider AI.
3. Procedure - overall steps
- Create a mentoring plan and milestones
- Create the design based on the design methodology below
- Implement / design the build using gen AI
- Reflect the learning experience using the semantic tree approach
- Iterate?
Note that the reflective process is the longest and most interactive - but -it follows the build i.e. you create something first and then reflect it in a wider context using the semantic tree concept. In this sense, this approach is ‘build first’ - i.e. reflection follows build not the other way round.?
4. Deliverables
- A system i.e. working code for learning purposes
- Structured analysis (the paper format)
- A reflective approach (the discussion format in the paper based on the semantic tree)
5. Methodology?
5.1 Overall Approach
- There are two methodology components: a design methodology and a reflective methodology
- The design methodology is based on using generative AI with requirements analysis and test driven development (ATDD BDD and TDD)? as we discuss below
- The reflective methodology is based on learning with a semantic tree. This is the same idea Elon Musk uses. See this reference?
- These tie to the three deliverables: A system ie working code, Structured analysis (the paper format) and a reflective approach (the discussion format in the paper based on the semantic tree)
- The research paper format helps structured thinking. You may or may not publish it - but the structured format helps you to learn.?
- The approach deviates from the classic research paper format - for example,? - the discussion section is much larger based on the reflective learning and the semantic tree
5.2 Design methodology
The design methodology comprises of two parts which tie back to the build using generative AI i.e. test driven development and product discovery.?
5.2.1 Test driven development
- ATDD BDD and TDD are test first methodologies. The overall philosophy is: you write the test first i.e. TDD focuses on writing unit tests before writing the actual code; ATDD involves creating acceptance tests before development starts, focusing on the overall functionality and behaviour of the system from the user's perspective.?
- By defining the tests - we define the end in mind and work backwards to the code using Gen AI.
- Test-Driven Development (TDD)? is more developer-centric and focuses on the technical correctness of the code. TDD results in unit tests.
- Acceptance Test-Driven Development (ATDD) is a collaborative practice that ensures the software meets the business and user requirements. ATDD often occurs at the start of a development cycle, when requirements are being discussed and agreed upon. ATDD results in a suite of acceptance tests that can be used as regression tests to ensure that future changes do not break existing functionality. It also ensures that the product meets the business and user needs.
- Behaviour driven development (BDD) is a development technique that focuses on the system’s behaviour.?
5.2.2 Product discovery
- Product Discovery is a crucial phase in the product development process where the goal is to identify and understand the needs and problems of the target users and to validate ideas for products or features that address these needs. It's about figuring out what to build, for whom, and why, before committing significant resources to the development of the product.
- Key aspects of Product Discovery include: Understanding User Needs;? Defining the Problem; Generating Ideas;? Prototyping and Testing; Validating the Solution; Iterating: Based on feedback; Aligning with Business Goals etc
- There are several methodologies and frameworks for product discovery. We draw on : Design Thinking, Lean Startup, User-Centered Design (UCD), Jobs to be Done (JTBD),? Design Sprints, Story Mapping, Ethnographic Research, Empathy Mapping, Kano Model etc.? Each of these methodologies has its strengths and is suitable for different stages of product discovery. Often, teams will use a combination of these approaches to ensure a well-rounded and thorough discovery process. The key is to maintain a user-centric focus, continuously validate assumptions, and be ready to iterate based on feedback and learning.
5.2.3 Relationship of ATDD to product discovery
- Acceptance Test-Driven Development (ATDD) and Product Discovery, though distinct in their primary focus and stage of implementation in the product development lifecycle, can be closely related and complementary.
- Both focus on User Needs and Requirements. The primary goal of product discovery is to to understand user needs, problems, and the market context. It involves identifying what to build, why, and for whom. ATDD also centers around user needs in the form of user stories or requirements that emerge from the discovery process.
- In the product discovery stage, we define what? success looks like for a product or feature, often articulated in the form of user stories, use cases, or scenarios. These user stories or scenarios from the discovery phase directly feed into ATDD, where they are translated into acceptance tests. These tests define the criteria for a feature to be considered complete and functioning as intended.
- There are of course two different sets of people looking at these two phases
So, we start with product discovery and get a set of user stories and scenarios which are assimilated into acceptance tests - which then are translated into code through Gen AI.?
5.3 Reflective methodology - The semantic tree
The reflective methodology then is used to extend what you have created into a wider picture of AI itself. (known unknowns). This works because complex subjects are not easy to understand linearly. This idea of semantic tree most recently popularised by Elon Musk See this reference ?does not involve memorising facts or learning isolated pieces of information. It focusses a lot on understanding the fundamentals first (the trunk of a tree) and then seeks to add layers of complexity in the form of branches.?
I also used the example of Russian dolls / Matryoshka dolls as a learning mechanism.? You spend a lot of time on the fundamentals and then also a lot of time in how the other pieces relate to each other and to the fundamentals.?
The process actually involves a mentoring / reflective process and the idea of dynamic learning paths for the lack of a better word i.e. . It involves using analogies, critical thinking, applying ideas in domains.
5.4 Implementing the semantic tree using a knowledge graph
- Knowledge graphs can visually represent complex information in an interconnected way, making it an ideal tool for embodying the semantic tree learning approach. Here's how knowledge graphs can be used in this context:
- The first task is to represent the foundational principles. Foundational Principles(trunk) can be represented as concepts (nodes/entities).? These nodes serve as the core from which learning expands.
- Relationships become connections.?
- A dynamic learning path involves proposing the next best connections/questions starting from a core idea.
- You can also model analogies as bridges between concepts.??
- Knowledge graphs encourage personalised learning through: understanding the next best question, understanding gaps in knowledge and critical thinking.
6. Format of paper / artefact
6.1 Title
6.2 Abstract
6.3 Introduction
- What is the problem or research question you are addressing??
- Significance of the problem
- What is already known?
- What is the gap you are trying to address
6.4 Literature Review?
6.5 Design Methodology?
6.6. Results
6.7. Reflective methodology?
This section is as discussed above
The reflective methodology is based on learning with a semantic tree. This is the same idea Elon Musk uses. See this reference?
6.8 Conclusion
6.9 References
Notes
- The views presented here are my own.?
Previous references
If you are interested in being a foundation member(it has a one off fee) please message me.?
