Dear All,

Welcome to intro2ml!

This is going to be your course nickname this semester (instead of using one of the 4 confusing section numbers). And guess what? You’ll have access to the material here: intro2ml.com

I’ll be writing to you at the beginning of every week, sharing the plan, and maybe a small recap of what we’ve done the week before. I’ll be sharing some thoughts that are relevant to my work, and some news that might be relevant to us that week.

Why These Long Emails? Who Has Time to Read Them?!

The reason I’m going through the exercise of writing you an extensive e-mail, is simple: writing helps me think, and I believe it would help you as well. Although this is not a writing course, I’d like to convince you this semester that it is probably the skill you’ll need to hold onto the most, to preserve and grow your intelligence in a world that is going to be flooded with it. What you’ll need most is the ability to direct it, and that requires practice. Expressing yourself well will ensure you’re in the driver seat, rather than just a passenger.

My claim is that writing is an effective tool to practice thinking, which I’d define as: the manipulation of representations to align a simple model with a complex reality, that we have to adapt to. And this is, after all, exactly what both machine learning and scientific modeling attempt to do: compress reality into simple representations, a.k.a. models.

The Questions

For this week, I’d like you to think about two questions:

The first: “what is a model?” A mathematical model. A computational model. A fashion model. A model student. A car model. An architectural model. A 3D model. A descriptive model. A probabilistic model. A simple model.

Machine Learning is concerned with getting machines to build models of the world, which it perceives in the form of ‘data’.

The second question becomes: but “what is data?” More accurately “what are data?”. Data is the plural of datum, which means “given” or “reference”. That is, the world is ‘given’ to us through our senses, and we try to make sense of it with models.

You’ll find that the way we learn about the world and adapt to it is deeply embedded in (or dependent on) our experience (obviously); starting from our perception, down to every small conception we build of it. Human learning is data-driven in a very similar way machine learning is. So by learning machine learning, you also learn about yourself.

Machine Learning is About Action and Perception

For us, humans, action is as important as perception for building models. Perception has no absolute grounding in reality; that is, we can’t see reality as is because we collect an extremely limited amount of data through tiny sensors, in a tiny location in space and time. The way we choose what to pay attention to and what to ignore is rooted in our actions.

We choose what to sense, and build simplified models of those senses, to plan our actions, in order to survive. Not only are actions necessary, but the consequences they have on the world become the next thing we perceive, in a dynamically evolving feedback. Learning lives in this very complex and nonlinear action-perception loop.

Fast forward to the present, this is where we are in the evolution of the field of Artificial Intelligence. AI agents are computational systems that build models about the world by using a combination of perceptions (data) and actions (tools). They still don’t do it exactly the way we do, but there is no reason they can’t be.

This week I’ll talk about the various types of models, of data, and of ways models are distilled from data. You’ll find that the important principles are quite simple once you think deeply about models and data.

Your Tasks

For this week, before coming to Thursday’s class, I want you to:

• Discuss the two questions on data and models with your class-mates who are taking the course. Push beyond what is obvious. Question assumptions.
• Come up with an example of data, and think of an example model that is extracted from it. The example doesn’t have to be a machine learning model: e.g. Newton’s second law is a model of how forces on objects are related to their mass and acceleration. A simpler model we have of the world is: tomorrow the sun will rise. Not guaranteed, and technically inaccurate because the Earth is revolving, but it’s a useful representation given the data.
• Think about: how that data is collected, and how big it would have to be to give rise to a good model.

Be creative, and try not to use a search engine or an LLM.

Looking forward to seeing you in class!

Best, j.b