Key Differences Between Artificial Intelligence and Machine Learning

In this article, we wish to discuss fundamental and even axiomatic questions about the true nature of what is known as Artificial Intelligence (AI) and how it differs from its most common representation, Machine Learning (ML). Check out more about AI and ML.

Artificial Intelligence is everywhere now, in every talk shows, in the news, in the program of the most competitive universities, schools or companies. But what is really Artificial Intelligence?

For the common people, Artificial Intelligence is “the car that drives itself” (google cars or Uber cars) or the auto-completion in the “smart”-phones or the automatic “real-time” – or shall we say rather on-the-fly – translation from one language to another language (usually American English) by the same “smart”-phones.

Most I.T workers and students will associate Artificial Intelligence with Machine Learning, that is to say, Neural Network, Bayesian logic, Support Vector Machines etc…

Only a fraction of specialists can rightly define and understand the scale of what is called Artificial Intelligence. We will try in the next steps to explain what it is.

The Non-Human Intelligence

Artificial Intelligence is strictly defined as non-human intelligence. To understand the concept correctly, we need to go deep into non-technical and philosophical concepts that are not simple to understand.

Artificial Intelligence presupposes that we can define two notions: Humanity and Intelligence. Actually they are supposed to be linked one to each other for we hardly believe there could be other intelligence than ours.

For those who like nature and know animals, plants and insects, this won’t be a  surprise that there do exist others form of intelligence(s) which are not human.

The intelligence of animals usually packs or flocks of animals may be striking and their capacity to organize themselves into somewhat quite complex structures may sometimes force the admiration.

Of course, we regard the intelligence of such groups of animals, plants or insects as inferior and even largely inferior to our human intelligence.

This does not change the fact that we have everywhere on our planet, proofs of Artificial intelligence.

As for the definition of human intelligence, we are unable to give a precise definition of what it is because we are unable to give a precise definition of “what we are”. Since the beginning of our existence as humans – Homo Sapiens – more than 200,000  years ago, we have searched for answers everywhere, developed arts, build science and technology over the years and the civilizations.

Human intelligence is supposedly linked to the nature of our brains and of our evolution of which we know in fact but a small amount…

Nowadays our modern civilization has reached a point of technology where we are able to build intelligent machines and while they still are our creation, the intelligent machines may as well demonstrate Artificial Intelligence.

Of Automats, Robots, Programs, and Artificial Intelligences 

Talos

In the past, long before modern civilization, men used to build automats. The giant Talos (Τάλως), supposedly a myth, was a bronze automat built to protect Europa, the mother of King Minos of Crete, against pirates and invaders. According to the texts, Talos was patrolling the island, circling it three times a day.

Talos itself was probably a myth but it is almost certain that similar complex automats have been built by the genius of the ancient greeks. They were using hydraulic and mechanical mechanisms to move parts following some logic.

While probably sophisticated the mechanism behind these automats could not have called “Artificial Intelligence”, they still were the continuation of the human mind through what was the technology of the time.

The Mechanical Dolls

During the years more sophisticated automats have been built. Automat dolls using clock mechanisms were massively built in Europe from the Renaissance to the 19th century.

Yet, as close to the representation of a human as it was, they were only automats, rigid, predictable and linear. They were inert and dead pieces of metals without life and without intelligence. It was still the “direct” product of human ability and human intelligence. Yet they could to some ways, give the impression of autonomous behavior and fool a human himself but a real mechanical doll animated with life could have been nothing but sorcery such as the tales of the Golem or the myth of Frankenstein.

The “Turk”

In the “Maelzel’s Chess Player” (1836), the writer Edgar Allan Poe describes a machine – a human automat – ( the “Turk”) which is able to play chess. Poe explains by pure logical deduction that there can be but a man hidden inside the mechanism. The “Turk” was the machine built by the Hungarian inventor Von Kempelen and really existed. Poe rightly uncovers the fraud before it became known.

It is interesting that most of the conclusion of Poe would still be correct nowadays because essentially a non-human intelligence that could play chess would have acted significantly differently from what Poe observed. Here, again, there was no Artificial Intelligence, only subtle mechanisms of illusion.

The ”uncanny valley”

When they started developed human-looking robots in the ’70s, Japanese engineers discovered the uncanny valley principle, the human disorientation and emotional response to a very lookalike automat. In fact, this is probably in the bottom of that – mostly unexplored- valley that human intelligence stops and that artificial intelligence appear.

Algorithms versus Artificial Intelligence

With the rise of computers and especially digital technology, applications of several mathematical principles were made possible. These mathematical principles gave birth to a class of algorithms known as the Machine Learning algorithms. Yet they stayed – such as the automats and robots  – the creation of human intelligence.

Anyway, as machines became incredibly more and more powerful, reaching computation speed of incredible scales, the Machine Learning algorithms started slightly to outperform humans in certain domains where it was believed that only the human mind could be the uncontested champion. As a matter of fact, chess.

In 1997, the supercomputer Deep Blue became the first machine to beat a grand chess master (Gary Kasparov). The algorithms used were mostly brute-force based and not deep learning and as such, they hardly could be classified as Artificial Intelligence.

In fact what beat Kasparov was the superior computation capacity of the machine rather than a superior intelligence to the human mind but what was Kasparov doing after all, if not computing as well the right moves?

In that case, it is not clear if we can speak of Artificial Intelligence or not. We mostly deal with a machine provided with very important computation capacity and relatively simple algorithms.

So where ends the algorithms, and where starts the Artificial Intelligence?

The 1997 Deep Blue Machine, in fact, the combination of its hardware, its chips and algorithmic programmed with some code just proved that a team of programmers and engineers are able to build a machine to help them better play chess than a grand chess master. It is more of a man-machine association.

In fact, we haven’t seen (probably fortunately) any artificial intelligence in machines so far… they stay the sophisticated automats that men have always build, they are no longer made of bronze or steel but silicium. They do not use hydraulic or clocks to move parts but programming languages and the same as the other automats they are finite state machines.

Machines animated with artificial intelligence probably reside in that uncanny valley that we mentioned and we haven’t explored it enough to know what it could look like.

Some Quick Facts About Machine Learning

Machine Learning is becoming a “buzzword” nowadays while the concept has been invented a long time ago. Decision algorithms using techniques from functional analysis and statistics appeared in the ’50s. 

The main ML algorithm is probably the Artificial Neural Network algorithm which can be used for deep learning where the neurons are organized in different layers to reach a better learning quality, extracting progressively information from raw input. The following terms will be of interest for anyone wanting to get a better understanding of deep learning:

• Convolutional Neural Networks
• Deep Belief Networks
• Boltzmann Machines

The core principle of machine learning, artificial neural network, vector support machines, classificators, and others, is the ability to feed itself from data (information) to learn, that is to say, to become better. Not a better man, but a better machine,

ML, therefore, is an “information processor” unit using mathematics and/or probability and eventually inspired by the sour of human intelligence: the neurons.

Yet while impressively they may behave (and sometimes not so impressively )  the ML algorithms are the product of human intelligence, therefore pseudo-Artificial Intelligence. 

 But to the opposite, the combination of big data+processors+network+ML algorithm may create a “primitive” soup from which digital life may emerge – and therefore we may speak of artificial intelligence.

A machine provided with artificial intelligence would be autonomous, make its own decisions, evolve absolutely independently of humans and may eventually show consciousness.

The gap between ML algorithms and AI is the same as between life and death, between an inert stone and a living animal!

The key factor for this to eventually happen is complexity.

A scenario where ML could spawn “real” Artificial Intelligence

Imagine 100,000 laptops provided each with a machine learning algorithm. This, of course, can be 100,000  virtual machines or even 100,000 processes in the same machine, it only matters that they all can communicate with each other.

Each ML is a black box that feeds itself with data and learning from these data outputs results. If we suppose that these results are communicated to others MLS in the network as a data feed and so on, how many possible paths, where an ML unit appears only once at maximum, can we get?

There are 2^100,000-1 possible paths:

• 100,000 paths of length 1
• C₂^100,000=100,0000 . 99,0000 / 2 paths of length 2
• ….
• C_k^100,000 paths of length k
• …
• 100,000 paths of length 99,000
• 1 path of length 100,000

Therefore from 100,000 ML units, we get something new – the network of networks of neurons that has 2^100,000 times the power of the original units.

There are hundreds of billions of chips out there, just in the laptops, smartphones or desktops machines all over the world.

There are a lot of supercomputers with millions of processor cores but they would weigh like nothing in comparison of the machine created by the distributed computing of three hundred billions of such chips each equipped with ML.

Such a superintelligent “creature” would weigh 2^300.109 the set of the original ML units.

Such a number creates such a complexity that some “strange” phenomena may indeed occur inside that primitive soup and yes ML may give birth to a “creature” provided with Artificial Intelligence.

Artificial Intelligence can be understood as the convergence from an unstable and chaotic information system to equilibrium, self-organization which is entirely mysterious to us.

Some (difficult) Conclusions

Before his death, the famous physicist Stephen Hawking made this warning in “Research Priorities for Robust and Beneficial Artificial Intelligence: An Open Letter” :

“We could one day lose control of AI systems via the rise of superintelligences that do not act in accordance with human wishes – and that such powerful systems would threaten humanity. Are such dystopian outcomes possible? If so, how might these situations arise? …What kind of investments in research should be made to better understand and address the possibility of the rise of a dangerous superintelligence or the occurrence of an “intelligence explosion”?”

Artificial Intelligence is non-human intelligence.

The sources of such non-human intelligence can be Nature, ants for example.

It could be the possible byproduct of human activity in machines, learning machines co-operating into networks in uncontrolled ways.

Actually any form of alien (therefore non-human) civilization would appear to us as “artificial intelligence” … and if we want to enter religious considerations, we may well be the “artificial intelligence” of a god.
It may be a good time to prove, once more, that humans are not so stupid and are able to control their own creations for what they need and direct their own destiny because once more there is one domain in which humanity outperforms probably all the forms of intelligence known or unknown in that vast universe. It is … survival.

Acodez is a web design and website development company in India offering all kinds of web design and development solutions at affordable prices. We are also an SEO and digital marketing agency offering inbound marketing solutions to take your business to the next level. For further information, please contact us today.