Thursday 17 July 2014

Intelligence

Introduction


One of the "holy grail" problems in Computer Science is referred to as "Artificial Intelligence".  The main issue that I have with this label is that I disagree with the notion of intelligence as being artificial.  Intelligence is either there or it is not.  And even then the spectrum of behavior that can be considered "intelligent" varies based on the person making the observation and his state of mind.

Consider for a second that we travel to another planet and discover a civilization of beings that are fundamentally different than ourselves: in their appearance, physical makeup, method of communication, culture, etc...  How would we judge them to be intelligent?  I would imagine that our limited perception would force us to compare their behavior, and their impact on the environment, to ours.  So, if they do not appear to create cohesive communities that in some way modify their environment (Think of a village with "houses", tools, etc...), would we look at them as basic animals with no "true" intelligence?

Another way of looking at this is to consider a monk that has taken a vow of silence and is in the "constant" state of meditation (excepts that he eats, shits, and sleeps).  We would immediately consider this person to be intelligent because he looks like us and we place him in a mental "bucket" of beings that we consider intelligent.  Yet he does not demonstrate any traits that would be identifiable as intelligent.  He does not communicate, he does not modify or create things, all he does is sit there with his eyes closed.  One issue here is that we only have a limited perception of the monk because we have no way of knowing what is happening within his mind.

As I struggled with the concept of "Artificial Intelligence", and over time gotten to experience and observe life (think raising children),  I realized that I have been "programmed", throughout my life, with certain beliefs that limit the range of what I would consider "possible".

One of these beliefs is that we, as Human Beings, are incredibly special and that our brain/mind is incredibly complex and unique and cannot be re-created "artificially".  We consider the fact that we have a wide range of emotions to be something exclusively "human".  I invite you to dismiss this belief, if only for the duration of this article, so that we can explore a possibility of duplicating, and possibly even improving on our own design.

Modern science has made key discoveries about the nature of human beings and, arguably, the most important one is that of DNA.  To a computer scientist, like myself, DNA looks, tastes, and smells like software code.  Code that works on biological level, instead of the digital level that my software programs work on.  So, if all we are is a product of DNA programming, that creates a biological machine capable of learning and intelligence (aka Human), then one should be able to do the same, or better, using software.

What is Intelligence?


To create "something" it helps to understand how it works and what you expect it to do.  So, to create an entity capable of intelligence I need to define what I consider to be intelligence. 

To me, the simple definition of intelligence is the ability to combine two ideas into a third idea:  A + B = C.  Where 'C' is neither 'A' or 'B' and we are able to infer that 'C' "makes sense" based on our understanding of 'A' and 'B'.  For example, you have a concept of a 'Chair'  and a concept of a 'Wheel' and you can imagine a 'Chair' with 'Wheels' and it "makes sense" to you, and you just invented a Wheelchair or a Rickshaw.

I would hazard a guess that most inventions that we enjoy today did not come around based on random combining of things in our head.  A more probable path likely started with some deficiency (or a problem) that someone had to overcome (find a solution).  Such "deficiency" can be very subtle and does not necessarily have to be related to one's survival, it could be as simple as feeling chilly or hot or bored or lonely.

Let's break down the above definition into its pieces and define what is required to achieve each one:

"A and B" - the building blocks.  How did they get into our head?
"Combining of A and B" - the thought process.  How is that accomplished?
"Making  sense of 'C'" - what makes something "normal" and something else "ridiculous"?

The Building Blocks


What is 'A'?  Ultimately, 'A' is a concept that we have stored in out brain based on our observations of the real world or recursive application of intelligence (a + b = A).  To be able to conceptualize our real world observations we would require a system for capturing an observation of the real world: processing and analyzing this observation, and storing it in our brain for future reference.

Humans observe the real world utilizing five primary senses:  Vision, Touch, Hearing, Smell, and Taste.  We can also be influenced in some way by forces that we cannot perceive directly (like radiation and other forms of energy as well as chemical agents that can interact with our bodies) but I am not going to address that at this time.  Additionally, we have a system for feeling of pain and discomfort that monitors our internal organs and can be considered a "sense" in an indirect way, or a secondary sense.

The way that our senses operate is most certainly pre-programmed by our DNA.  The shape and function of our organs (eyes, ears, nose, skin, tongue) are there to make sure that our mind is fed with a steady flow of sensory data.  So, if we are to emulate human intelligence, at least at first, we would have to design basic senses using the science and materials that we understand.  Currently we are not capable of growing a biological eye and are not capable of processing the data it produces.  But, arguably, we can achieve a very similar result by utilizing cameras and other computer based technology to capture the visible light, detect spacial contrast (shape boundaries), and detect temporal contrast (movement), and we can focus the lens to recursively apply the above process and more efficiently utilize the computing resources.

Think about how we "see" the world.  We do not analyze every little thing we see in front of us.  There are things that draw our attention.  Those are objects that are moving and things that stand out from the "background".  If you enter a room that has white walls and there is a small red hart sticker on the wall, you will immediately "see" the red hart sticker.  If the walls were painted red or if they had multi-color wall paper with random shapes , you would find it very difficult to notice the sticker or find it, even if you were told that there is one there.  Same is true for the objects that are moving.  If you have a wall with painted on butterflies, and there is a live one sitting on that wall,  you will not "see" it until it starts to flap its wings.  Immediately you will focus your eyes on the butterfly and will be able to follow it as it flies around the room and will see the details of its shape and colors painted on its body.  At the same time, because your attention is focused on the butterfly, you will have a limited perception of other things happening in the room or other objects moving. This design ensures that we efficiently utilize our brain power when analyzing the sensory input. But, at the same time it demonstrates one of our weaknesses that stage magicians take advantage of.

Once an image is captured and broken down into basic elements our brain can store them in a way that is generic.  This means that a "concept" of a chair is stored as a collection of properties that have a certain relationship between each other.  And if you see a completely new object that matches those properties, to a certain threshold, you will recognize it as a "chair", even though it might be something completely different if looked at from a different perspective.

I imagine that a similar system can be applied for storing information from other senses and all the properties can be tied together to create a "complete" picture.  Thus smells and sounds and sights can reinforce each other and create ripple effects, when encountered.  Consider that when you encounter a smell of a chocolate chip cookie you will also "see" the cookie in your head and, depending on your state of mind, the ripple effect can be quite deep and follow the relationships between the stored properties by invoking those images and sounds and smells and replaying them in your head.

As we experience the world around us we are constantly adjusting the information stored and the relationships between the individual properties.

Combining of A and B


Once all the possible observations that come through our senses can be stored and connected to each other one can start building a net of relationships between objects and concepts and time.  The main aspect of the information that comes through our senses, compared to the information we conceive in our head, is that we consider it to be THE TRUTH.  It is difficult for us to disregard what we have experienced directly.  This concept can be seen in our culture quite a bit.  Consider for example an expression "I'll believe it when I see it" or the fact that courts consider eye witness accounts to be almost indisputable, when judging someone.  As such, the information that we experience directly is the bedrock for everything else that we have in our head.

With time, and enough "programming", indirect experiences can also achieve similar indisputable status in our head.  An example for this are mathematical concepts, historical knowledge, religious believes, etc...  I view this as an internal optimization for situations where we need to use concept 'A' as a building block, when combining with concept 'B'.  If the status of 'A' is "indisputable" we do not need to use computational resources to verify it's validity.  Thus when trying to decide if 'C' is "valid" the decision is easier if either 'A' and 'B' are already considered "valid".

When we combine two concepts together, in our head, what we most likely do is run a mental simulation and observe the outcome.  This is likely very similar to the type of processes that happen while we sleep.  This is an important idea that I would like to explore further.

Imagine if we had a system in place that would feed simulated sensory input to our brain.  This would in fact allow us to "see"/"feel"/"hear"/"taste"/"smell" even though the experience would not be coming from the external world.  But, more importantly it would allow the brain to create situations that you have not yet experienced in the real world.  By combining various building blocks in random or pseudo-random way, we can have a vast array of possible things happen, even if we would never be able to experience them in real life.  This would explain the experience of dreaming.  Dreaming is an extremely important aspect of our intelligence because it allows us to explore an idea, no matter how ridiculous, in an extremely immersive realistic environment.  Interestingly enough, if the experience in our dream simulation is intense enough it can breaks through the mental protective "wall" and send the nerve signals to our limbs, causing them to move while we sleep.  And, if there is a failure of this simulation system, a person might sleep walk or even act out some of their dream sequence in real time.

One interesting aspect of human wisdom, acquired over the millennia through pattern recognition, is the idea that is expressed in various cultures : "sleep on it".  In essence, it acknowledges the power of dreaming in helping us to find the most optimal decision.  This is, most certainly, because while we sleep we are actively evaluating the experience of the previous day and running multitude of simulations to imagine possible outcomes and their desirability.  If you take anything from this, before making an important decision, especially one that involves various conflicting factors, have a good night sleep - you will be more likely to make an optimal decision in the morning.

During the day, we engage in similar activity consciously.  This is usually referred to as "imagination".  This is an important ability that allows us to make decisions in real time.  But consider the fact that while we are imagining things we are still flooded by, and have to process, all of the sensory data from eyes, ears, nose, etc...  All of that takes computing resources from our activity of imagining.  This can be referred to as "distractions".  Therefore to help you focus you can reduce the amount of sensory data flowing to your brain.  If you are listening to an audio book or a lecture you can close your eyes. If you are in a quite room or in the car, you can turn off the music - conversely if you are in a loud environment with multitude of voices and sounds you can put on headphones with some rhythmic music that you are well familiar with - that will require less computing then songs that you are not familiar with.

To fight distractions our brain has an ability to "focus".  In essence this is auto processing and disregarding of sensory data.  An example of this activity can be an experience where you loose track of how you got from one place to another while driving or walking.  Obviously you traveled that distance, and your eyes were opened, but you cannot recall the experience because the brain auto-processed most of the sensory data and performed the necessary actions to allow you to safely operate the car or walk.  

It is my belief  that large parts of this imagination framework are pre-programmed by our DNA.  But as we get further away from the raw sensory input more and more is controlled by learned experiences.

Making  sense of 'C'


The third piece of the puzzle is the method of recognizing that what we imagined, when we combined 'A' and 'B', actually makes sense.  This is the pinnacle of our intelligence and allows us to create new ideas and solve complex problems.

The concept that will help us understand this process is humor.  In the very basic form we find anything that is abnormal as funny.  In a land where we all wear hats, a guy that puts a tea kettle on his head, instead of hat, will appear funny.  This is because we immediately recognize it as abnormal.  That is also the reason why we laugh when someone stumbles and falls.  Of course, if it was customary to wear tea kettles on our heads , our reaction would not be the same.  This is why our experiences in the childhood, and throughout our life, determine what we consider normal.  It is also important to note that since the idea of "normal" is not pre-programmed by our DNA it constantly changes.

Since our life experiences constitute the universe of "normal", and idea 'C' must be close enough to that universe to be considered "normal".  That can explain why many inventions are accidental.  If an idea strays too far from our universe of "normal" concepts, it can be difficult for us to "believe" that this new idea is "normal"; experiencing it through our senses (by observing the accident) makes it automatically "real" and ads it to our universe of "normal" concepts.

We can use this concept of "normal" testing and apply it to our dream simulations and imagination.  This allows the dream simulation to combine things in random or pseudo random fashion and we will discard the majority of the combinations because they do not pass the "normal" test.

If my understanding of this process is correct, one conclusion that is apparent is that people that have a fine tuned sense of humor are more creative and have an ability to sift through the mental concept combinations faster, and are able to pick the ideas that are actually "feasible" faster and more accurately.

Why?  (or Conclusion)


One question that I did not touch yet is "Why would someone want to create a computer based intelligence?"  The answer depends on your ability to go beyond your existing "belief system".

One of the aspects of our intelligence is the physical limits of our skull and subsequent limit of the processing power of our brain.  The size of our scull and other aspects of our biology, as pre-programmed by our DNA, are there to ensure our survival in the "jungle", or the world where there are many external dangers.  These dangers require us to run fast, to climb high, and to be able to go hungry for long time.  If our head was much larger, it might require us to have much larger bodies which in turn would require us to consume much more food - to feed the biological processes of our body.  So, in the end, the DNA achieved a balance and subsequently created a built-in limit to our intelligence - that was good enough to survive the "jungle".

It would be interesting to see if over the last several centuries the average body size and the head size has increased or not.  If so, it might support an idea that our ability to produce food in abundance allowed humans to slowly evolve to be smarter.

A computer based intelligence does not have to have the same physical limits as human bodies.  As long as we can solve the problems of utilizing thousands of processing cores efficiently and have ability to store humongous amounts of information efficiently and be able to retrieve it quickly, we can have an entity that is vastly more "intelligent" than a human being.

So why would you want this?  To solve all of the extremely difficult problems of our age. Remember the idea that DNA is a programming language?  What if we can decode that language so that we can write our own programs or re-write existing ones?  Can we make humans live much longer? What about hibernation and space travel?  That can probably be solved as well.  There are plenty more of difficult problems to solve that can be seen as both "benevolent" and "nefarious".

Building of the computer based intelligence is the nuclear arms race of the 21st century.  Whoever controls this intelligence "wins".  I imagine many countries and corporations are already in process of thinking through the same ideas, that I presented here, and are putting together solutions.

Companies like Google are especially well positioned to leverage the brain power of its employees and the infrastructure of its data centers to create this type of computer based intelligence.  In fact, I believe they are in the middle of doing exactly this.

What is our society going to be like if we have the technology to live twice as long or "forever"?  Currently the utility of the majority of the population is to service the population itself.  Things are going to change!