Whether through search engines, automatic translators, voice assistants such as Google Assistant or Alexa, robotic lawnmowers or assisted driving systems for cars, artificial intelligence has now permeated our daily lives.
While we are still a long way from the advanced artificial intelligences of popular science-fiction works, the world of artificial intelligence is full of possibilities for simplifying our everyday lives, whether in private life or at work. But what is behind the term artificial intelligence? What are the most common existing forms of artificial intelligence? And how is artificial intelligence used by MCA Concept to better serve companies?
What is artificial intelligence?
Artificial intelligence, or AI, is defined as a set of theories and techniques aimed at giving a machine the ability to simulate human intelligence. A machine capable of solving problems by following human reasoning and logic is generally referred to as artificial intelligence.
Research into artificial intelligence began around the 1950s, with researchers beginning to question the boundary between human and machine and the possibility of a machine able to think for itself.
The study of AI encompasses a number of disciplines such as robotics, statistics, logic and neurobiology.
To function, artificial intelligence relies on more or less advanced algorithms. An algorithm consists of a succession of instructions given in a certain sequence in order to perform a task automatically. Thus an algorithm tells a machine how to solve a problem that has been submitted to it, which is what allows the machine to process data.
You have probably already heard of Google’s algorithm series, which allows the search engine to define for each search query the relevant results to offer from the existing database.
To better illustrate the concept, an algorithm is often compared to a recipe: for example, a recipe for spaghetti carbonara (algorithm) provides the steps to follow (instructions) to obtain a dish of spaghetti carbonara (result) from certain ingredients (data).
Different approaches to AI
AI research attempting to replicate human intelligence in a machine has resulted in two different approaches: symbolic AI and connectionist AI. These approaches are each based on different methods of human learning :
1) Learning through the transmission of knowledge, i.e. knowledge, rules, procedures, what is known as ‘school’ knowledge.
Take a maths problem for example: One can explain the path to solve a maths problem step by step to another person, who will then know how to solve the problem without having to touch the problem themselves. This approach aims to imitate the logical reasoning of humans (symbolic AI).
2) Learning by experimentation, knowledge gained by observing the world and by trial and error, until success is achieved (connectionist AI).
Take learning to swim for example: it is not enough to explain to someone how to swim for them to have acquired this skill, they will have to swim to truly know how to swim.
As mentioned above, symbolic AI attempts to mimic the logical reasoning of humans. A symbolic AI system will therefore attempt to perform a task submitted to it by following a list of explicit rules implemented by the programmer.
This method is at the origin of the so-called “expert systems”, tools aiming at simulating the reasoning and know-how of experts. An expert system will deduce information on the basis of facts and logical rules that have been fed to it.
Such tools can be used in the medical field to make a diagnosis. For example, it can deduce the probability that a patient has a certain disease based on the symptoms observed and the abnormalities present in the body by programming the following set of rules: ” if a patient has symptom X and their body has abnormality Y, then the probability of them having disease Z is N%. ».
Programming such systems requires a fair amount of effort, as more than a hundred rules have to be entered into the system. Moreover, while they work well in very simple situations, expert systems can quickly become ineffective in a real situation where there is a profusion of facts and rules to be taken into account, as it will take a long time to analyse all possible answers.
A symbolic AI system is limited by rules explicitly defined by the programmer. Indeed, it can only act according to predefined scenarios, therefore it does not improvise.
Symbolic AI is useful in the sciences, but is limited when it comes to language processing or image recognition.
The connectionist approach seeks instead to simulate human intelligence in the machine by “teaching it to learn”. In this case, algorithms learn from examples instead of simply receiving specific instructions to solve a given problem. They develop skills in a sort of autonomous way. This is what is known as machine learning.
There are different ways to train machine learning algorithms :
1) Supervised learning : A large amount of “training” data is fed to the algorithm, this data is associated with labels so that the algorithm can know what they correspond to. This allows the algorithm to understand the problem at hand: recognising a certain element in the data it receives. As it proceeds, the algorithm will be able to deduce which features identify the labelled item and will then be able to locate it in unlabelled data.
The algorithm trains itself on this data until it is able to indicate whether or not a specific element is present in the data with a certain level of accuracy.
Let’s take the example of an image recognition algorithm :
Objective : the algorithm must be able to distinguish between a blackbird and a hen so that it can automatically classify photographs of these birds without human intervention.
Training procedure : A set of pre-labelled images of blackbirds and hens are given to the algorithm as a training base. The machine then analyses these images and learns to recognise the distinctive features of blackbirds and chickens.
Later, when the algorithm analyses unlabelled images, the programmer gives feedback to the algorithm to tell it if any of its observations are wrong, for example if an image was categorised as a robin when it was a hen.
Results : the algorithm is able to automatically identify whether an image contains a blackbird or a hen with a certain degree of accuracy.
2) Unsupervised learning : A large amount of training data without associated labels is given to the algorithm. The latter must then identify the underlying trends in these data on its own, and will group the data according to common characteristics.
3) Semi-supervised learning : The algorithm receives training data mixing labelled and unlabelled data.
Deep learning is the result of machine learning. It is based on artificial neural networks, similar to the neural networks of the human brain, combining different algorithms. Several layers of neurons are superimposed and work more or less like a filter, starting with a raw analysis that is refined with each new layer until a precise result is reached. Each layer analyses the information transmitted by the previous layer.
In the case of image recognition, each neural layer will be responsible for identifying a feature of the element it is supposed to recognize. If the system needs to identify a bird, for example, one layer will search the image for a beak, another for wings, etc. As another example, an automatic translator will first try to recognise the letters in a presented sentence, before moving on to analyse the words and so on.
These neural networks can thus learn to identify structures in the data and classify information. They give deep learning systems more advanced capabilities than machine learning systems. Deep learning systems are thus able to handle more complex tasks and analyse a larger amount of data, whereas machine learning systems will tend not to improve after a certain amount of data.
As mentioned above, deep learning systems are particularly good for image recognition or machine translation tasks. But also for assisted driving systems, personalised recommendations on websites or virtual assistant systems. There are even algorithms that can recreate the style of paintings by artists such as Van Gogh on other images, a technique known as style transfer or neural style.
The increase in computing power of computers in recent years as well as access to massive amounts of data has greatly contributed to the development of AI, especially in image recognition. Algorithms are now more quickly assimilated by machines.
And what about companies ?
But, you may ask, what do artificial intelligence, algorithms and business management have to do with each other ? Well, you see, artificial intelligence, specifically symbolic AI, can be put to work for companies. Indeed, they have their role to play in the optimisation of processes within a company. These are algorithms that make it possible to standardise and automate operational processes, such as the creation of an invoice or the sending of a payment reminder. The software executes the tasks according to the instructions that have been programmed by the developers.
Thus the programmers of the MCA Concept team have developed a variety of algorithms for the MCA Kale management software (and its business variations) to simplify processes and automate redundant tasks in any business.
These algorithms based on symbolic artificial intelligence assist you in your daily tasks and alleviate the administrative burden caused by repetitive, often uncomplicated but time-consuming tasks.
The algorithms allow the MCA Kale software to perform these tasks efficiently, reducing input or calculation errors as they follow the instructions given to them to the letter.
This artificial intelligence technology also allows the software to exploit a large amount of data for activity reports and statistical analysis at a stroke. What would take days and days for a human being is accomplished in the blink of an eye by algorithms.
With the support of algorithms in tedious tasks, you are free to focus on higher value activities.