The nature of intelligence

Written by tautvilas | Published 2016/07/23
Tech Story Tags: philosophy | artificial-intelligence | technology | machine-learning | education

TLDRvia the TL;DR App

Movie still from Ex Machina

I have always been interested in the subject of Artificial Intelligence. It is because by building AI we are learning valuable lessons about ourselves. After all, we consider us to be intelligent, but are not really sure what that means. AI is an attempt to reverse engineer our mind and to define intelligence by creating an abstracted version of it. Can AI become smarter than us? What is the true nature of intelligence? These are the questions that make me wonder.

Very recently there have been some astonishing AI advancements with models based on Deep Neural Networks (DNNs). Apparently today AI can perform image identification better than humans and win against the world champion of board game GO. We have lost tic-tac-toe, checkers and chess to a digital mind long time ago. These board games are closed environment discreet systems that have winning conditions and rules strictly defined. Older AIs used mostly brute-force and sheer computing power to win against humans. DNNs however are solving problems using evolved pattern recognition. This way they can tackle more fuzzy and human-like problems that have vast solution search space.

The holy grail of the field of AI is to develop the so-called seed intelligence or general purpose intelligence. A program that would exhibit such properties would be able to modify itself and perform general purpose tasks that have been given to it. In its versatility and adaptation it would be similar to human intelligence. Since such intelligence would be digital it might be many times more efficient than our meaty brains and could quickly modify itself to become much more advanced than homo sapiens.

So we have been able to develop AI that outmatches humans in specific computational tasks, but still have no good clue how to build an AI that is generic, can adapt itself and creatively solve a wide range of tasks. Artificial neural networks might be a part of this puzzle, but something important is still missing. It seems that we have solved separate parts of intelligence, but we still don’t know how to glue them together. I feel that in order to unravel this problem we need a better general understanding of the nature of intelligence.

Many people know about IQ tests. We imagine that a very smart and intelligent person must have a high IQ. IQ is our definition of intelligence today. However these tests have also been criticised for not taking into account such aspects as creativity and social intelligence. General IQ tests cover just a narrow space of human intelligence related to pattern recognition. Also these tests are very human specific. Most of us consider our pets to be somewhat intelligent, yet they would not score anything on any human IQ test because these tests are based on language. If we want to find the roots of intelligent behaviour we have to go deeper. We have to start exploring non-verbal intelligence that can be found in animals and babies.

If you observe a newborn baby you will see that it is always in constant motion while awake. From the first day the baby tries to understand its environment. The movements of the limbs is really a baby questioning his world: how does this feel? It tries to move itself towards the feel good state. It does not have to be taught to do it. In a way the learning process is driven by the baby from the first day. We do not recognise this because the question asking is non-verbal, but the baby is constantly exploring and gathering information from the environment surrounding it. When the baby becomes a toddler and still can not speak he continues exploring the world by pointing fingers at things, waiting for them to be named and explained to him.

Yet in many countries this questioning pattern is broken at school by the education system. Education is usually based on a different definition of intelligence. It considers pupils as dumb and empty when they enter school. In school the pupils are taught information as if it has to be ‘put into’ their heads so that they can become smart and intelligent. Such education system is neglecting the fact that intelligence comes from the bottom up to the top, not the other way around.

That is why education only works if the student is curious. If he is asking a question then he is ready to understand the answer. If there is only one way flow of information from teacher to student then nothing can be taught. Maybe that is the reason why we hardly ever learn anything new from TV. When we are watching it we are flooded with answers about the questions that we never really asked in the first place.

Our AI models are being constructed using the same outdated education pattern. The neural networks are passive in their nature. They are ‘dumb’ and rigorous training has to be performed on them with huge data sets in order for them to learn how to perform a certain task. After being flooded with information they are able to answer specific questions. But they still lack the ability the newborn has — to question their current environment.

So if we want to construct general purpose AI the question asking has to start from the beginning non-verbally and grow up to abstractions and language from that bottom point. This implies the necessity of body. Body is the ground in which the mind grows. It gives a way to start learning. If the abstractions that are learnt by AI are not rooted in our non-verbal physical reality then they can not be truly universal.

Currently the progress of robotics is lagging behind artificial neural networks. We can only construct robust robots with a few sensors and several motors. To the contrary human fingertips have approximately 2500 receptors per square centimetre. There are large dedicated areas in our brains for processing all this sensory input. If we could construct an exceptionally sensitive and agile robot with a DNN and enable it to explore its environment this creation would have a chance of becoming the first universal learning AI.

After all, it looks to me that the nature of true intelligence is not the ability to give an answer, but to ask the right question. Non-verbal questioning using body is natural to every sentient being. It might be the fabric and essence of all intelligent behaviour as we know it.


Written by tautvilas | Developer, CTO, dad
Published by HackerNoon on 2016/07/23