How Do Humans Learn?

Machine vs. human

We have just seen how a simple rule-based computer program involving if-else comparison statements cannot make sense of complex visual data such as images. While computers can process any amount of data and perform some basic operations on it, these programs are unable to infer any meaning out of visual data.

The human brain, where all learning takes place, effortlessly understands data received through the eyes as images. It is also able to meaningfully identify areas in the image and assign them appropriate names and labels. For example, try to identify a snail, a monkey, a butterfly, a cat, and a ladybug in the image below. Click these objects to reveal their labels:

Find the objects

We have seen that while identifying the objects of interest in an image is difficult for a computer, this very same task is easy for humans.

Brainy humans

Our brains can see the data received from our eyes and “interpret” this into meaningful boundaries of objects. This ability of humans is called vision, and it is a dominant part of human learning. It is also the hardest of all tasks for a computer. It also lets us appreciate what human brains are capable of.

Humans are “inquisitive” beings. We have a deep urge to know and to name each new concept that we can see or experience. Initially, we often need supervision in recognizing a new object and naming it. However, with experience or after having seen some instances and variants of the same concept many times, the human brain is able to retain the representation of the concept. Based on this general representation of the object, we are able to recognize variants of the same object. We make mistakes all our lives. We just hope we make less and less of them as we grow in our experience and conception of the object. We learn.

What is important to note here is that humans store a representation of the object and its features in their brains, not the exact details of the object. We know the general conception of a cat; we retain a representation of catness in our brains, not precise instances of all the cats we have seen.

Let’s train how to spiral

Let’s understand how humans learn from observing and looking at multiple examples of the same concept. We’ll continue with the example of spiral galaxies. You are looking at the galaxy images and labeling each of these with the type of galaxy. We are interested in learning only the concept of spiral galaxies. So the images of spiral galaxies are labeled as “Spiral.” All other types of galaxies are marked as “Not Spiral.” When we are learning, we have a trusted elder or a supervisor with us who can point out the object as well as its name to us. It is up to us to learn the association. Click the images in the grid below and let the supervisor teach you the concept:

Training: Spiral vs. Not Spiral galaxies

Our learning gets tested often

Closely look at the images above. Each image is labeled by its type. Try to figure out why some galaxies are labeled as “Spiral” and others as “Not Spiral.” Once you feel confident that you have mastered this, you can try your hands on the given task. You have been assigned the task of labeling a fresh batch of galaxy images from the James Webb telescope. Let’s see how many of these you can correctly identify:

If you make a mistake, the universe will give you its feedback in red.

Testing Spiral vs Not Spiral galaxies

Humans generalize their learning

These were a different set of images that were shown to you in the training phase, yet you managed to classify most of them correctly. You have done a good job! Did you notice how you were able to correctly label the new images even though you saw them for the first time? Can you figure out how that happened? Well, you looked at some galaxy images and also their names/labels.

Let’s look at some features or characteristics of those spiral galaxy images you might have retained in your mind. These images helped you learn the general concept of spirals rather than rote-learning the specific images shown to you.

Feature

Description

Spiral Arms

The most distinctive feature of spiral galaxies is their spiral arms, which extend from the central bulge outward.

Central Bulge

At the center of a spiral galaxy is a dense, spherical structure known as the central bulge.

Disk Structure

The galaxy's spiral arms and the central bulge are embedded in a flat, rotating disk.

Halo

Surrounding the disk and central bulge is a spherical halo

Rotation

Spiral galaxies rotate, with the stars in the disk moving in orbits around the galactic center.

Most likely, this is the representation of the spiral galaxies that your brain learned from the galaxy data.

Even though you don’t remember the exact details of the individual images down to the individual pixels, you were able to retain the key features of the spiral galaxies.

Every mistake is a learning moment

So far, so good. We have seen that our brain is able to retain important features of the images and use this to identify objects around us and classify them with already-learned names.

To err is human.

However, just as we have seen in the previous example, sometimes we make mistakes and incorrectly identify specific objects. When this happens, it means our concept isn’t complete and has some gaps in it. We have to then update our learned representation of the general concept. The human mind has the ability to update its understanding by taking in feedback from the errors that it makes in perceiving and understanding objects. Click the “Next" button to walk through the learning process pipeline:

Can machines learn from the way humans learn?

Let’s review what we have learned so far about how humans learn, using a flowchart as an illustration.

Press + to interact
Human learning pipeline
Human learning pipeline

We can call this a human learning pipeline. The whole point of this detour was to go back to our simple computers and figure out what essential changes need to be made so that they can emulate human learning. That is what machine learning scholars do. They study natural phenomena that display intelligence and learning and design algorithms and computational models inspired by nature.