Mimicking??

Can you tell how you learn? As an adult?

It may be worth thinking about.

We tend to learn a lot by reading, ingesting information, and also mimicking — doing what others do.

What do we do then that is worth learning?

Social media use is about creativity, applying filters and themes, adding sounds. Using hashtags. Mimicking the behavior of successful social media persons is said to be the quickest way of boosting your results.

Learning for me has been always superbly fascinating. I am your normal homo sapiens, having gone through the educative stack. I was very fascinated and thriving well in STEM subjects.

Found a career in IT, which was somehow a place that I knew I would land in.

Yet there was constantly a curiosity towards a few things tingling in my mind:

What is this “learning” everyone is talking about?

When and why would we ever stop learning?

What and how does an AI system learn?

Is learning quicker without emotions?

AI does not have emotions. We do have emotions, and indeed the motivation and actions of learning are quite heavily dependent on controlling your emotions toward the subject of learning, as well as having a control of the frustrations and the pain of using energy towards learning.

The latter fascinates even more now in 2024, as there is an explosion of AI tools and of course — our friend ChatGPT. The Large language models (LLMs).

Real-world learning for me — just like for everyone else, has contained painful moments. There are some exams that didn´t go as well as you would like. Some were a breeze.

I cannot deny the existence of pain. However it is also a reasonable question that would we feel the emotional fulfillment, and even grateful self-compassion about having learned something, if we did not feel any of these supposedly negative emotions?

We sort of achieve a milestone.

Tacit learning is something that happens informally.

Learning is not just about thriving and meeting the goals - one after another. I think in the human experience, this is core to successful learning:

Are you persistent? Can you overcome the obstacles along the path? Can you stay motivated, even though there wouldn´t be necessarily anyone actually pushing you forward?

AI and computers are said to learn group-wise, very fast. Computers basically share and copy the experiences with speed of light. They can spread the entirety of a learned corpus, just by copying the deep-learning models (accumulated data and code) that has gone into the training period. Then every agent (computer “persona”) is equally fit as anyone. They do not need to necessarily do individualized, custom learning. In fact this is the very promise of a AI world: there are a lot of network effects, and even perhaps this winner-takes-all thing that might be unhealthy for the markets in large.

The automated pilots of Teslas learn from various terrains and experiences all the time — and they can almost immediately share the learnings: they become better all the time! Imagine if human drivers would basically learn in a similar manner. I think it would be very beneficial.

Why do I say that.. computers learn — as if I had suddenly had a Eureka moment? I say it because this article is about the nuances and philosophy of learning in general.

I love learning! Inner drive to understand the world

I am totally addicted to exploring what is around me, and trying to understand it all. Yes. That would pretty accurately describe the life-long journey that I see is actually worth while.

As a side-effect I am immersing also to learn about what is the AI, learning in AI, the mathematics used in it currently, and perhaps even make little bit of contributions to this area? Who knows.

Supervised learning?

Supervision means in AI context, that first a AI model is trained from the data. Then, after training, we use and grade (confirm, whether AI results are ok or false). So the answers given by AI (called “predictions”) are graded against ground truth.

“Ground truth” concept

Imagine that you had to learn mathematics all by yourself. Or a new language. It is doable, yet there is considerable help if there is someone to give you feedback.

AI (machines) learn also by getting feedback, through a error term.

Error term

The error is easy to calculate in certain contexts, and it may be harder in other applications.

Automated supervision in AI?

What about if our AI could actually automatically teach and supervise itself? It is possible. We will talk that in the next episode.