05/18/2023

How To Think

notes on learning fast, inductive bias, proactive inference, first principles, and curiosity

Background

In 2014, I was in 11th grade and obsessed with Veritasium – the YouTube channel created by Derek Muller, an Australian-Canadian science communicator. Veritasium's videos cover a wide range of scientific topics, including physics, chemistry, biology, engineering, and mathematics. The channel aims to explain complex scientific concepts in an engaging and accessible way, often using experiments, demonstrations, and interviews with experts.

Derek Muller, the host of Veritasium, has a background in physics and is known for his enthusiasm in sharing scientific knowledge. 

At one point in the summer of 2014, I read Muller’s PhD thesis Designing Effective Multimedia for Physics Education, which he completed in 2008. I’ve always been interested in the theory of knowledge and learning how to learn and how to think, and why some people are curious and others not. 

His thesis was actually my first ever introduction to Cognitive Load Theory and Constructivism (I love social constructivism). 

—

Last weekend, I learned how to animate some blobs in a few hours (for an ad I produced that will be shown on a big screen in Times Square next week – eeek!) using some youtube tutorials and written articles (multimedia). As I was finishing up, someone asked me how I learned to animate so fast. I didn’t know what to tell them at the moment. I kind of just… did it? Followed the steps, I guess?

—

This morning, I was in a journal club meeting where we discussed a paper surrounding emergent abilities of LLMs when the topic of inductive bias came up. Inductive bias guides machine learning algorithms by incorporating prior knowledge or assumptions, influencing their decision-making and generalizations from data. Inductive bias is not inherently bad. In fact, it is essential for effective generalization from limited training data. This piece does a great job of why we should care about inductive bias.

—

As I was swimming this evening, about 9 hours after the journal club meeting, the answer to the question “How do you learn things fast” emerged for me. 

I realized how I think to learn fast can be attributed to a principle I had internalized from when I read Muller’s thesis on how to learn Physics concepts when I was 16. 

The idea went something along the lines of approaching the pursuit of knowledge with a completely open mind, shedding your cognitive biases at the threshold. I remembered that he emphasized the importance of essentially embracing a blank canvas, untainted by preconceptions, allowing novel ideas to flow unimpeded. 

—

My internalized principle

I reread relevant parts of Muller’s thesis tonight to find exactly where the idea I internalized came from. It turns out that it’s actually rooted in misconceptions more than preconceptions. 

Misconceptions can 

1. give learners a false sense of knowing so they reduce the amount of mental effort they put into learning which you could relate to the Dunning-Kruger effect 

2. interfere with recently learned science – this is called proactive inference or interference – favouring prior knowledge or misconception over new information 

What I had internalized was avoiding proactive inference if and when it ever came up in my learning. Bear with me here because, yes, misconceptions cause harm as noted above, but they can also be useful.

Avoidance

In Muller’s experiment, learners with a high attention capacity at the time of learning, allowed them to avoid proactive inference. But as soon as other tasks were added or distractions were added, the avoidance of proactive inference disappeared, reinforcing the focus of mental effort on learning the concept. 

The notion I internalized of avoiding misconceptions and proactive inference was not the primary aim or finding of his thesis. It was one of the manipulated variables in the experiment conducted to test his hypothesis. The primary aim was to find the optimal multimedia design for Physics education. 

Experiment

To explain what actually happens when we try to avoid misconceptions and/or proactive inference, and how they can actually be useful, I need to explain the experiment a bit. 

Essentially, Muller tested the learning of Physics students from different years – meaning they had different levels of understanding: from fundamental students to advanced. The different prior knowledge the students had was used to predict how useful the multimedia lesson would be. Different prior knowledge meant that they were coming with different levels of preconceived notions. It was hypothesized that novice learners potentially prefer more instructional direct lessons with reduced sound effects and dialogue, and advanced learners wanting a bit of a narrative and more dialogue. 

Conceptual change

The relevant discussion of the experiment for us is focused on conceptual change as the thing required to learn new things fast. Conceptual change refers to the process of revising or altering one's existing understanding or mental representation of a concept or topic. It involves a shift in an individual's beliefs, misconceptions, or prior knowledge towards a more accurate or advanced understanding.

Conceptual change → Cognitive conflict

Conceptual change requires creating cognitive conflict. Cognitive conflict is a state of mental discomfort that arises when individuals encounter information conflicting with their existing beliefs. It plays a crucial role in facilitating conceptual change by prompting individuals to recognize inconsistencies or inadequacies in their current understanding, and seek a more accurate understanding. By experiencing cognitive conflict, individuals engage in deeper thinking, reflection, and analysis to resolve the discrepancy. This transformative process allows them to revise or replace their previous beliefs, leading to conceptual change and the development of more accurate and sophisticated understandings. Creating opportunities for cognitive conflict in educational settings can promote meaningful learning outcomes by challenging learners' existing beliefs and encouraging the pursuit of more accurate knowledge.

Students, fundamental or advanced, must maintain a balance of relative focused cognitive conflict for conceptual change. Remember the hypothesis? Novice learners potentially prefer more instructional direct lessons with reduced sound effects and dialogue (lesser cognitive conflict), and advanced learners wanting a bit of a narrative and more dialogue (more cognitive conflict). 

Misconceptions are required for cognitive conflict. Awareness and attention is needed to not fall down the Dunning-Kruger or proactive inference paths. 

Creating cognitive conflict

According to The Theory of Multimedia Learning, creating cognitive conflict can be done by how you design your learning, which impacts how a person thinks 

How do you design learning to create cognitive conflict?

1. misconceptions 

2. having new conceptions available – is the “lesson” (new concept) in the tutorial or after the tutorial? What creates cognitive conflict to the point of healthy engagement but not losing focus? (reminds me of Flow by Mihaly Csikszentmihalyi)

3. the new conceptions being plausible is an ongoing battle between facts and emotions because we are biased towards previous ideas, previous beliefs and changing someone's worldview is not easy – first principles thinking

4. making sure the new conceptions solve the new problem but also give answers to old phenomena that couldn't be answered

To get a comprehensive solution to “how to learn new things fast” that is applicable universally, we probably need to go beyond the four steps outlined in Derek Muller’s thesis and gain a deeper understanding of the mindset surrounding behaviour change, as well as the broader realms of socialization and ontology.

But these four steps deeply resonate with me as this is exactly what I believe I’ve been unconsciously doing to learn new things fast. 

Connecting the dots between inductive bias and proactive inference

Thinking about inductive bias this morning reminded me of proactive inference since both involve the influence of prior knowledge or information on the learning and processing of new information.

And proactive inference is what I internalized as what not to do when I was 16 years old – which I have now been practicing for almost 10 years in my thinking and learning. 

It is important to remember that there are differences between human learning and algorithmic learning. Algorithms typically have explicit inductive biases that are built into their design, while human preconceptions can be more diverse and variable. Algorithms can process large amounts of data more efficiently but lack the nuanced understanding and context that human cognition can provide.

Fortunately, we transcend algorithms. Our consciousness, capacity to embrace the present moment, and inherent freedom of choice enable us to acquire knowledge in a manner distinct, superior, and intricately nuanced compared to machines.

—

Just like inductive bias is not inherently bad and is almost needed for effective generalization from limited training data in ML; misconceptions and proactive inference are also not all bad. They sure must be avoided, but their avoidance must not lead to losing focus. Having them in the first place is key to cognitive conflict. 

Essentially, challenging preexisting beliefs at every stage of reasoning and trying to have the new answer solve both the new problems and old unsolved problems. 

Distilling it down

Basically, what you want is to create conceptual change through cognitive conflict and first principles thinking. And this is how I think I do it: 

1. Increase + maintain attention – use intrusive thoughts specifically surrounding misconceptions or preconceived notions related to the topic of interest to be curious rather than reject new ideas
   (drop ego) 

2. Be present
   (practice Noticing) 

3. Maintain a flow zone
   (practice mindfulness) 

I have written about 

• the idea of Noticing in How To Know What To Do 

• presence and mindfulness in It’s All For The Plot

• ego in Quarter Life Crisis

It takes time to achieve conceptual change. I have been practicing it unknowingly for almost 10 years. I think intentional practice can probably make internalization quicker.

Learning design is not a revolution in education. The medium of learning design, communication, learning environment doesn't matter as much as the actual thinking that goes on in the learner’s head. 

To learn is to think. That’s why the question isn’t “how to learn”, it’s “how to think”

Now that we know: acknowledging misconceptions → avoiding misconceptions → cognitive conflict → conceptual change → learning new things fast

A new question arises: can intentionality in this practice lower the activation energy and thus, time required to become faster, better thinkers? I’d love to be part of that experiment. 

This piece is 42/50 of my 50 days of learning and writing. Consider subscribing to support and hear about new posts. 

Other notes:

1. The Science of Thinking is one of my favourite videos by Veritasium, which goes a bit deeper into how the brain works, how we learn, and why we sometimes make stupid mistakes.

2. Researchers Perry Zurn and Dani Bassett would describe the type of curiosity and connecting-the-dots I did in this piece as the busybody type of curiosity. The busybody collects stories over time creating loose knowledge networks. They talk about it in their book Curious Minds.