In Part 1 of Building Intelligent Machines, we discussed that we can define Intelligence as a measure of magnitude of interaction with the world. More the ways or dimensions of interacting with the world, more the intelligence. Decision making is a big part of increasing the dimensions of interaction with the world and understanding how humans take decisions can help us build intelligent machines.
Human decision making is based on information. There is a lot of information we are born with, which is stored in our DNA. For example, making a sound, crawling, eating and may such things which we know as soon as we come to life. They are part of our DNA and we don’t have to learn them.
But that information alone is not sufficient for humans to survive. We keep on gathering information from our world to refine our understanding of it. We learn from it, and that as you know is not a very simple process. Sometimes we fail, sometimes we succeed.
In this post, I’ll discuss the process of learning. The scope of human behavior and characteristics discussed here are only related to rational decision making. Let’s leave the complex human behavior to the experts.
As in the Part 1, there are few questions to ponder over while reading, so I’d request you to keep your notepad handy.
How do we gather information ?
We are born with five senses — vision, touch, taste, sound and smell. Consciously or unconsciously we keep on gathering information from this world, the environment around us.
What happens after we have gathered this information? Perhaps a better question to think is — Why do we gather information? This question gives us the idea of goal. The most important goal of any life is — to survive.
Without a goal, gathering of information is not useful and we rarely learn from something that is not useful. All learning needs to have a goal. Goals can also be unconscious, may be hidden and not very clear to us, but we respond in accordance with our goal.
Learning, at its most primal, emerges from a process of observation and response. Before we deep dive into this, I have a question for you. What is this object?
I know most of us will call it a tree, which is true. But, let’s think for a while on this question — Why did you call this object a tree?
Have you seen it before or somebody taught you that it’s a tree? Maybe you saw it in a book. Maybe you “just know” that it’s a tree. Let’s go deeper…
Do you remember all the tree you ever saw?
The last tree you saw?
The first tree you saw?
I am sure the answer to all the above questions is No!
The way our brain is able to identify a tree, is based on some kind of similarity. We might think Hey, this thing is going up from the soil, has leaves, has a stem and many other characteristics that collectively we call as Tree. Brain doesn’t store everything, it processes information very differently. There is no databases inside.
In some ways, you have understood the fundamental principle of the tree.
We have gathered the information and we’ve made it “our own”. We “learned” what a tree is. We were observing it through our eyes and our goal to gather this information is to know the world better.
After gathering the information, our brain does some processing according to the goal and that process is what “learning” is. When we interact with the world, or go through the process of observe and respond. We discover the underlying principle of it.
We understand this idea scientifically as well. In computing world we can create or discover a space or environment where two things are closer even if there is no immediate similarity between them. The closeness depends on the characteristics or some kind of similarity. These spaces are called latent spaces. For example in latent space these two trees would be closer.
The main problem is to search for that latent space. This search can be of two types — supervised and unsupervised. We will have a detailed discussion about it in the next post.
Observe and respond
Coming back to our decision making framework “observe and respond” we have introduced another important part of the process which we call “learning” which happens to be the outcome.
Through this principle, we can also have computing machines to go through the same process of observation and response, where they can start expanding the interaction with their environment and as a result learn the characteristics which will give them more decision making power.
To have a machine with power of decision making, we should also think about where machines can be better than humans in decision making? Since we still don’t understand human emotions and other aspects of intelligence, we can only create purely rational decision making machines or software.
Humans have both positive and negative relationship with emotions. For example, while driving, anxiety and hurry can become a cause of an accident whereas in a competitive game, the same emotions can make a miracle happen. Positive emotions have pushed humans to expand their boundaries.
Machines don’t get drunk and they don’t sleep. So, we know we can make use of pure rational highly analytical decision making machines. I believe driving should be better left to machines. What are the tasks you think machines can do better than humans?
In the coming post we will discuss how we can apply the principle of observation and response to create intelligent machines. A computer’s world is a vector world made of vectors in a vector spaces. The fun part is creating a way to communicate our world to the computational world.
Co-Founder and CTO of Spext. Wanders in deep thoughts of science, spirituality and human nature. Often goes to trek Himalayan trails.