Artificial Intelligence : Let’s clarify some misconceptions !

“Artificial Intelligence” : trendy term, exciting for many, mysterious for others, suspicious for some but definitely a hot topic for the majority.

These last years, everyone is talking about AI and everyone is curious to explore more and exploit the potential of this field.  

However, when discussing about this subject with different persons, I noticed confusions and misconceptions related to the definition of AI, its sub-fields, its applications, …

So, I decided to write an article to clarify 6 main misconceptions related to Artificial Intelligence.

Misconception N°1 : AI is a new field

This year, AI is celebrating 70 years old. The history of artificial intelligence dates to the 1950’s when Alan Turing proposed a test for machine intelligence known later as The Turing Test. This test is a method of inquiry done by a jury (humans) to determine whether a computer can think like a human being. The term Artificial Intelligence was coined later in 1955 by computer scientist John McCarty (the father of AI) at a conference in Dartmouth, United States.

The period between 1980’s and 2010’s was the era of machine learning breakthroughs like IBM’s Deep Blue, KISMET, AIBO, ….

The boom of AI was drived then by deep learning breakthroughs like SIRI, IBM’s Watson, Eugene Gootsman which is a 13-year-old virtual Ukrainian boy that passed the Turing Test in 2014.

In 2018, Google said that his project Google Duplex has passed The Turing Test and you can judge by yourself by watching this video : https://www.youtube.com/watch?v=D5VN56jQMWM

Misconception N°2 : AI is a subfield of Data Science

At first glance it seems to many that AI is a subfield of Data Science, but AI is actually a subfield of Computer Science.

In fact, Data Science is a recent term that study and extract insights from massive quantities of data (unstructured, semi-structured and structured) to make business decisions. AI is more about perception, reasoning and action. With AI, machines try to imitate humans’ intelligent acts.

It’s clear that every field or discipline is distinct but when it comes to deal with enterprise digital and innovative transformation, different techniques and processes interconnect and can be used interchangeably to conceive solutions and create value depending on the context.

Another term appears here “Data Intelligence” that includes both big data, analytics, machine learning, and statistics techniques.

Misconception N°3 : AI is one branch, one technique

AI is in fact a field, a discipline that includes different branches like Computer vision, speech recognition, speech generation, Natural Language Processing, Machine Learning, Deep Learning, Chatbots and dialog systems, …

Each branch deals with analyzing different types of data (image, audio, text, …) and involve the use of different techniques of processing depending on the context and need.

Let’s zoom, for example, on 2 fields : Computer Vision and Machine Learning.

Computer Vision is the field of analyzing images and videos to extract information (e.g. silhouettes extraction, …), track motions, build 3D models, detect events, …This field requires qualifications like camera calibration, image processing, mathematical morphology, pattern recognition, …

Machine Learning is a set of techniques based on mathematical and statistical approaches and used to develop computer programs that can access data and use it to learn from themselves and improve their performance in solving tasks without being explicitly programmed.

And talking about techniques, it is good to know that there are 4 subareas of Machine Learning (Supervised, Unsupervised, Semi-supervised and Reinforcement) on which many AI solutions are built. You have probably heard about these areas. If not, I will explain to you briefly each one :

In Supervised Learning, the dataset is labeled, and we will classify a new observation among predefined classes.

In Unsupervised Learning, the dataset is not labeled, and we will discover the structure of the data after applying an algorithm (clustering method).

In Reinforcement Learning, machines learn to take optimal decisions by interacting with the environment and, by observing the result of certain actions.

There is also another type of learning called semi-supervised learning which is a combination of supervised and unsupervised learning. 

Misconception N°4 : There is only one and common definition of AI

Although AI is ancient, there is no common definition of it. University of Helsinki and Reaktor define 2 criteria to categorize a system or a solution as AI. These criteria are Autonomy and Adaptivity :

1.    Autonomy is defined as the ability to perform tasks in complex environments without constant guidance by a user.

2.    Adaptivity is defined as the ability to improve performance by learning from experience.

Based on this two criteria I tried to make my own definition of AI that I share with you : "Autonomous and adaptive systems imitating intelligent human behavior and solving real problems".

With this definition, I keep key concepts of AI (autonomy and adaptivity), I insist on the real aspect of AI and I explain that the systems are imitating intelligent human behavior and not intelligent like humans who have mind and conscious. You can find other definitions of AI among different sources in the Internet but in my opinion, any definition should be based on these 2 criteria.

Misconception N°5 : AI is all about prediction

It is such fascinating to predict a price, a churn, a choice, a behavior, …and many AI systems allow this. But let’s agree on the fact that when we talk about prediction in AI, we mean an algorithm that has been trained on historical dataset and applied to new observations to predict the probability of a particular outcome.

But prediction is not the only aspect of AI strength. With AI we can cluster observations, detect anomalies (e.g. : detect fraud in insurance or in financial transactions, …), detect suspicious behavior or unusual motion using intelligent video surveillance, interpret facial expressions, …and many other applications.

Misconception N°6 : AI will replace humans in the future

If you type the word “Artificial Intelligence” on Google image search, you will find that AI is represented, in most images, as a humanoid robot thinking or controlling the world. I don't think this is an accurate representation since it gives the impression that AI is made for unrealistic scenarios and science fiction where robots will replace humans. AI offers, however, solutions to many realistic and present problems related to diverse fields such as health, environment, security, ...In certain applications, AI helps humans become faster and more efficient in their tasks and decisions, so it will rather complement than replace humans.

The term “Artificial Intelligence” involves philosophical questions

Can Intelligence be artificial ? Intelligence isn’t related to the presence of a conscience ? and conscience isn’t specific to human being ? Personally I prefer saying machines imitating human Intelligence but for those who are interested to find the answers to these questions and want to understand the philosophical side of AI, I suggest you to read : What has AI in Common with Philosophy ? of Professor John McCarthy http://jmc.stanford.edu/articles/aiphil.html

And to get more elements about AI, I suggest you to take AI elements course made by University of Helsinki and Reaktor https://course.elementsofai.com/, a great course with illustrative examples, trying to demystify and simplify AI.