Why "Artificial Intelligence"(AI) is merely an "Artificial Smartness"(AS) and a dumb inefficient one too!

The tech world needs its buzz word to tickle the interest in the enthusiasts as well as in the investment community. These buzz words are a great way of creating artificial bubbles of hope for change and prosperity. Dot Com, CleanTech, Web 2.0, Bots, Bitcoin, Virtual Reality, and now artificial intelligence and machine learning.

The overwhelming buzz and the fuss that is been created around AI now gives me a sense of another bubble and as intellectuals and responsible individuals, it is our duty to challenge any such home runs and invite an open debate around the subject, a debate which is based on facts and figures and not paranoia and emotions.

I am writing this article to vote and celebrate human intelligence and to prove that the dumbest person living today is smarter than the smartest artificial smartness!

In this article I want to argue why the " Artificial Intelligence " is only " Artificial Smartness " and has nothing to do with "Intelligence "

Neurons

Let us first understand what is AI and why the word "intelligence" itself is a gross overstatement in the word "Artificial Intelligence".

Neurons are the functional units of our brain and there are billions of them( I haven't counted mine though!). Each Neuron holds a piece of small information as an electrical signal. Many Neurons fire together in a systematic way for pulling out a memory or creating a vision in your brain or giving senses(Yes, Neurons are so powerful that they can even make sense out of some of the dumbest political speeches!). There are various ways of Neuron collaboration which has fascinated a whole generation of intelligent humans who are studying this behavior under Neuroscience. Neurons can learn and unlearn and relearn things.

In 1943, Neurophysiologist Warren McCulloch and mathematician Walter Pitts wrote a paper on how neurons might work. In order to describe how neurons in the brain might work, they modeled a simple neural network using electrical circuits. (Read the history of Neural Networks).

And then from there on there has been a quest to build and perfect this computer model of human Neurons.

Essentially the theory assumes that Neurons stores and conducts through an electrical signal.

As semiconductor technology has progressed, it has allowed exponential semiconductors to be put in a small density area and making their collaboration more and more computationally powerful. Also, it has accurately followed Moore's law who had theorized that the number of transistors in an integrated circuit will double in every two years. And so the computer has evolved. Along with this, the Neural Network or the quest of replicating a human brain in the computer has seen rapid effort plus pace. But this model still had one fundamental problem. Humans learn from their environment and other humans and other beings. Humans are connected. Neurons of many people are connected. Think of a team in a company and each personnel as a computer, their brain being a Neuron model computer. Then if the collaborate on a project, a project is implemented faster as the knowledge from different brains gets connected. The same thing happened with the advancement of internet and connectivity. The individual computers and computing devices got connected to form a Network of Neural Networks. And to keep our obsession with fascinating names we call them Deep Learning Networks or DeepNet or whatever! And different networks and technologies together are called Artificial Intelligence.

So what these Neural Networks or Networks of Neural Network Do? If you give them a problem, they supposedly find a better solution than humans after learning sufficiently. For example, Face Recognition. You show the neural network photos of 100 different people, these programming abstracts will extract what is called features and create a path between the features in a weighted way which is called a model. Then when you show a photo of one of the trained faces but with different pose or emotion, the network can still tell which is this person. The more photos they see of a person and different persons, the better model gets. These are finite problems. The solution of finite problems are called Machine Learning. Artificial Intelligent is moving around in a world and learns from it, act to maximize the result. The learning model is Reward-Penalty Model.

Reward-Penalty model-based learning is one where if a computer had taken a decision X and it resulted in a more suitable outcome then Computer rewards that decision by adding weight to the decision, else it subtracts weight as a penalty. Video Games are using this technique since as long as the history of the game goes. Here also, the more finite the "World", the better the computer gets. Chess is a perfect example of how AI works. This world has only 64 places, 32 Characters, finite rules, a reward and a penalty, and an end game.

But problems like driving are much more complex, there are ethical, reflex, emotional, cultural and a whole lot of other issues. So for building even a working model of a self-driving car, a computer has to gather data from driving of about 1 million miles, as against a human who can learn driving with 100 miles of driving and perfect the art in say 1000 miles.

So, Artificial Intelligence or connected Neural Network can solve finite problems in finite world which has finite sets of fixed rules with better accuracy and efficiency with reward-penalty model. The model, and the network becomes more complex as the world becomes more complex and the rules ends to become infinity. By connecting and disconnecting electrical connections between semiconductors which are perceived as Neurons

Intelligence Vs Smartness

The ability to acquire and use knowledge is called Intelligence. Anyone who can learn quickly and use knowledge more efficiently is called more intelligent.

There is another word which unfortunately often interchanged with the word intelligence and that word is Smartness.

Smartness is the ability to use knowledge more efficiently for maximizing the result in whatever that is applied to!

Smarts may not be intelligent, Intelligent may not be smart.

Smart and Intelligent is rarest of the rare trait.

For example, if you are a sportsman, you have to know when to use a stroke or move, how to interpret the same of your opponent and getting better at your game. So, smartness deals with learning the rules of a game and then getting better at them by practice.

Bill Gate is one of the smartest businessmen in our time as he smartly monopolized a market that still remains monopolized. Steve Jobs was another smart man, but a smart visionary who could envision the future and latched onto it in a smart way. Einstein was an extremely smart (perhaps the smartest) Physicist of our time. Warren Buffet is the smartest investor of our time.

Smart people may not be intelligent because they haven't proved their ability to learn other things and subjects and pioneer multiple of them. Perhaps pioneering one domain is so challenging that often it leaves no scope for these individuals to even think of other things.

Intelligence is to acquire and perfect knowledge. Knowledge is not limited to a single field. It is the ability of learning and using what you have learned. If I have to name the most Technological Intelligent human that we know, he has to be Leonardo da Vinci. The man has learned about arts, painting, architecture, maths, philosophy, mechanics, avionics and shown traces of genius by producing great results in most of these fields. The other name that people may argue is perhaps Nicole Tesla. In modern-day, Elon Musk probably is the only name that comes to mind when it comes to Technological Intelligence.

Here also the word intelligence is not merely limited to "applying knowledge" in technology, but there are vast other things which deserve intelligence. They include Social Intelligence, Emotional Intelligence, Psycho-Physiological Intelligence and so on. I have my fair share of doubt about emotional and social intelligence of both Tesla and da Vinci. Elon Musk needs no bias to be left out of the realm of both.

And like many false measurements, Intelligent Quotient or IQ is the most horrible KPI to measure intelligence. For a record, the man with most IQ served most of his life as a bouncer in a bar.

Intelligence also deals with the ability to learn what one has to learn and learn effectively such that smartness can be brought in to use the knowledge efficiently for better outcome.

Fundamental Flaws with Neural Network And Artificial Intelligence

Firstly, the fundamental assumption flaw with Neural network is that Neurons are triggered and fired with electrical signals.

In 1947, four years after actual model of Neural network was conceived, John Carew won Nobel prize for his epic work on Synaptic Inhibitory System which shows that connected neurons can fire without cascading a brain activity and established the presence of GABA substance.

Further studies established a chemical substance called Acetylcholine being a major chemical component responsible for neuron systematic connection.

This is very important knowledge to consider while defining and hypothesizing Artificial Intelligence( you can now start calling it artificial smartness by all means). Because the chemical model of systematic Neuron connection is beyond modern computation.

Cascading electrical activity is expensive. Let's do the Maths. An average human needs 3000 calories a day to survive and thrive where about 60% is used by the brain. 1 Watt is Energy or Calorie Consumed par second. In this way, a human consumes 1800 cal/ 86400 sec=.02 Watt a day. And humans can play both finite as well as infinite problems. On the other hand, a Laptop consumes about 600Kwh of energy for 8 hours on time. Are we even talking about any match? Take into account that this energy is not free, it has to be produced and harvested.

If Computers were to fed on the food we eat, one computer would need food needed by 1000 humans almost daily

Computer food is cheap, but not cheap enough. I will leave it on the reader to explore energy consumption growth as computers and data centers have grown.

As we discussed AI is based on Reward Penalty model, which is built around human learning model. We do not take decisions that result in an undesirable outcome. And we are bound by rules. But there are two fundamental reasons for learning: Survive and Thrive. With average humans Survival Instinct is much more dominant, Smart Humans puts more focus and energy on thriving. Artificial Intelligent on the other has no such issues. It needs to win and win an end game. Winning is solving a problem successfully. For example, when I am teaching my son the science and art of Chess, I prefer to challenge him but lose sufficiently so that he remains interested. We all do this. Winning is not the only objective in life, winning heart, soul, winning the trust and many other things come into the picture in our world. No such obligations exist in the computer world.

A transport problem where you want to go from one place to another is really a finite problem and a smart computer program can solve it more efficiently than a human because most average humans are not as smart. But a smart human-like Elon Musk can beat any smart computer program by bringing a whole new dimension to the solution (like he brought in the concept of Hyperloop). This out of the box, out of the rule thinking and exploring infinite possibilities and finding a solution in infinity is a hallmark of human intelligence and smartness.

So inability to consider non existent rules and objects in learning is the biggest drawback of so called AI.

And any learning mechanism that works around thriving based on finite rules is just smartness.

An Artificial Intelligence is way less efficient than even the dumbest human on planet in Food vs learning ability, The intelligence is merely a smartness of optimizing finite rules in a finite world, the mechanism can not identify the problem and needs a problem definition, doesn't have "Survive" model and only relies on "thrive model", can not balance between "Reward" and "Penalty" based on other forms of world like emotional and social world, can not find a solution in a non-existent world, can not create new rules or alter rules.

In terms of Intelligence itself, Einstein was way more intelligent than all the AI combined, Elon Musk is multi-million times smarter than the Deep Mind, da Vinci's smartness and Intelligence would suppress so-called AI for next 100 years and beyond.

It is you to think should we really call AI an AI or we should call it AS. And if at all we should focus on an Intelligence, it must not be into something which is a magnitude of time inefficient than human intelligence.

Investment in research/development/ventures in AI is over 100s of billions of dollars if not trillion. Have we put even 2% of that into innovating better knowledge acquisition tools for the humans?

Have we ever thought seriously in developing systems that can produce more da Vincis and Einsteins and Steve Jobs and Elon Musks? Because they are the ones who will see the future, understand the present, explore the unknown and solve humanity's great challenges.

And those who still do not want to get into the debate because it is unpleasant, please do a simple experiment.

Please do not cook food for a day, Your brain will find a way to get the food, be it dropping out to one of your friends or moving into a restaurant, or just eat some raw fruits. Now abandon electricity supply to all Google data centers for a day and ask the artificial intelligence to figure out it's food. Can it?

For me personally, both Human Intelligence and Smartness is always going to stay ahead of their artificial versions simply because of the underneath fundamental principles.

Can we now please change the name to Artificial Smartness for a fairer comparison to that of humans?

Disclaimer: I have been writing Neural Networks and so-called ASs since the past 20 years, and this article has nothing to do with my ability/inability to comprehend and implement an AS solution.