AIFP Part 0: Why learn about AI from first principles?

Hi, I'm Eric. A very warm welcome to this series. It is written for those who envision a better world with the aid of AI and want to build that future themselves regardless of their technical background.

A little bit about myself: I'm a PhD student in Natural Language Processing at the University of Cambridge. I've supervised MPhil and MEng students in their NLP projects, where they went from limited research experience to submitting a 50-page thesis. Some went on the publish their work. I'm also a R&D leader at an AI startup in Beijing and a technical consultant at an AI startup in London. I've supervised 50+ Cambridge Engineering undergrads in their informatic courses and feedback has been great. That's my qualifications for talking about this topic.

Why am I writing this series? I know a couple of friends personally who would benefit from this series. And I believe we can do so much better if we can engage more brilliant minds, you, in thinking about AI in both technical and non-technical aspects. That's my WHY for this project. Now, let me explain why you change-makers want to learn about AI from first principles.

Bridegway AI is about connecting people, like what a neural network does internally. Illustration generated by DALL-E on prompt: "A minimalist abstract design representing artificial intelligence".

---

From Inventing the Wheel to Real Change

Imagine living in an ancient tribe that survives by hunting, gathering, and basic farming. One day, after a long day’s work, you’re heading back to your camp when you see the village’s eccentric tinkerer running down the village with a round piece of wood in his hands. He’s shouting, panting, and brimming with excitement:

I've made a wheel!

Let's call this ingeneious, nameless, pre-historic inventor "the wheel guy". The wheel is undeniably one of the most important inventions in human history. It revolutionized productivity and paved the cornerstone for cultures. But did our “wheel guy” single-handedly advance human society?

No, as you can tell from common sense. Technology and inventors alone aren't enough to change the world. People need to apply that technology to solve real problems. The wheel is later adapted to make carts for moving things around, plows for farming more lands, and chariots to conquer more territory. Would the “wheel guy” have known about the challenges in gathering, farming, or warfare and devised every single innovation himself? Certainly not. In the same way, the world today is not just exciting for AI engineers (who are essentially “inventing the wheel” of our era), but also for the change-makers in every field who see possibilities for making things better.

But one thing is for sure. To make carts, plows, and chariots, you have to understand what a wheel is. I would imagine the prototypes were first conceived when their makers were learning about the wheel. AI may not be as fundamentally revolutionary as the wheel (or, is it?), but if you are trying to introduce AI-driven innovation in your field, you will need to have a working understanding of the technology.

While you can't leave everything to the Engineer, rejoice in the fact that it is perhaps in learning about AI when you will come up with the world-changing idea. That's why this series is written without any assumptions in technical background - you don't need to know all the maths to conceive how to change the world. What would you build in your field with the wheel of AI? Let's find out together.

---

Starting from First Principle

You must have heard of AI before. Perhaps there are already specialized AI tools for your field and maybe you are already using them to your advantage. Why spend time learning about AI if you already know what they are?

This series aims to help you develop your own system for understanding AI technology from common sense. Think of it as growing your own knowledge tree of AI. So far, you have seen various fruits on many branches. But perhaps you are not sure how they are connected to the main trunk. And worse for change-makers, because you can't see the whole picture, you are not sure which branches are more likely to grow, and which are withering and doom to fall. To grow that tree, you need to start somewhere solid. Something you already know and have been applying throughout your school years and careers. That foundation is what I call "First Principles" (or "axioms"). The first "wheel guy" story is a typical example of starting from first principle. We reason about the opportunity for change-makers in the AI era with a common-sense analogy. You don't need maths to understand that.

More excitingly, you will be able to decide what can you bring to the tree - where to cultivate and harvest your own fruits. Starting from first principle gives you the edge of having minimal assumptions, so you may be an even better innovators than the "wheel makers". Engineers that have been in the field for long may be so used to some assumptions that they are refrained by them by default. You can avoid that. Breakthroughs and change often happen in questioning existing assumptions. What this series aims to bring to you is the foundation to raise educated questions to your wheel-making friends. Questions that they will appreciate, not discarded because your questions show that you "simply know nothing about AI".

---

A Roadmap: How you will get there

I will take a top-down approach in explaining, starting from the most general ideas. For every topic, we will try to re-invent them ourselves. Here are the topics we are going to cover in sequential order. A brief overview for each chapter is provided after the syllabus.

Artificial Intelligence from First Principles (AIFP)

Syllabus.

Part I. The Big Picture.

1. What exactly is an AI system?

2. A framework to (start) understanding any AI system.

3. From ideas to AI models: a complete example with AI-assisted rapid prototyping

4. A short history of AI technology.

5. A brief overview of AI capabilities as of 2025.

   
   

Part II. Technology

1. The standard AI "weaponary": Classifier, Predictor, Retriever, Generator.

2. Training AI: ways to learn from data.

3. Running AI: what you need to provide AI services.

4. Evaluating AI: setting up useful "exams".

Part III. Capability

1. Language Generation: how do AI write like human?

2. Image Generation: how do AI paint pictures?

3. Visual Detection: making AI "watchdogs".

4. Information Retrieval: finding the right thing faster.

5. Predicting the Winner: a better chance than lottery.

Part IV. Use Cases

1. AI-Assisted Writing: a dream that has come true.

2. AI-Assisted Visual Design: the promised land?

3. AI-Assisted Security and Surveilence System: a commericialized technology.

4. AI-Assisted Search: the "long-tail" is getting longer.

5. AI-Assisted Scientific Research: finding the most-likely successful path.

Part V. Future

1. Where is the AI technology headed? The Big Problems and Opportunities.

2. What's left for human in an AI world?

3. AI Safety and Societal Impact: Can AI become destructive? How?

Chapter Overview.

In the first part, we are going to cover the first questions to ask when we see something new: imagine you just see an aeroplane for the first time as a child. The likely questions are as follow, and they correspond to each of the five posts:

1. What is an aeroplane? What is not an aeroplae?

2. How can such a giant thing take off from the ground?

3. Who built this areoplane? How did they build it?

4. Who invented the plane? How did the first plane evolve into what we see today?

5. How fast can this plane go? Is it going to become even faster?

   
   

In the second part, we are going to look at the underlying technology that makes AI powerful. These are the equivalent of the jet engine for AI. You are going to learn the language of the AI "wheel-makers". We will keep it high level without diving into maths and codes (most of us won't need them anyway thanks to AI). The goal is: you will be able to understand the challenge of your wheel-making friend, and help him/her in overcoming them.

In the third part, we move on to understanding what AI can achieve if you put the technology in Part II together. Now we are talking at the level of fighter jets, cargo planes, and unmanned drones. If building an innovative solution is like building an army, then knowning the capabiilties is like knowing your soldiers. I aim to cover the most widely-known capabilities that have been the building blocks of real products and solutions. You will be able to understand how these capabilities arise from piecing together the AI Technology in the previous part. You will also be able to think about what capabilities are most relevent when you apply AI to your field. If you need examples, move on to Part IV.

In the fourth part, we dive into examples of real AI use cases to show you how multiple capabilities can be/have been combined to form a solution. Some of these cases are already highly established (e.g., AI-Assisted Writing); Some are more preliminiary in terms of acceptance at the time of writing (e.g., AI-Assisted Scientific Research), but nonetheless have exciting outlooks. The goal is to provide chances to explore AI's capabilities and limitations in real contexts. By this time, you will be able to evaluate for yourself if the change you have in mind can truly be powered by AI.

In this fifth part, we deal with the big question: the future. In 2025, it is apparent that predicting specific future technology stack (e.g., what algorithm will people use to train future AI) is doomed to fail. In the past 5 years, the field has seen major updates to its learning paradigms for more than 3 times. But every change-maker cares about the future. So these questions remain valid: What can we say about the future of AI? What's the role of human in a post-AI era? How is AI going to impact our society? What can we do to ensure AI "do good"?

---

So here's your roadmap. I hope you are excited about the journey as I do. Let's make a start together! First article in part one: What exactly is an AI system?