Artificial intelligence is a fundamental component of the digitization process that is changing our society. What even a few years ago we thought could only be possible in a science fiction film is now a reality: we talk to computers, phones show us the fastest route to the nearest gas station, and our watches know if we have done enough physical activity.
Technology is getting more innovative, and researchers, engineers, and programmers take on the role of teachers by teaching computers how to learn on their own.
Machine learning is interesting for science and IT companies like Google and Microsoft: even the world of online marketing can change through artificial intelligence developments. This article explains the evolution of so-called artificial intelligence (AI, in Italian AI) in recent years: what exactly does machine learning mean? What machine learning methods are there? Finally, why should marketers focus on self-learning systems?
The History Of Self-Learning Systems
Robots and automatic machines have been part of human life for several hundred centuries. Several authors of romantic literature had already addressed artificial intelligence, and even today, robots are the fascinating protagonists of books, films, and video games. The relationship of the human being with the thinking machine has always oscillated between fear and fascination.
The fundamental studies on machine learning only began in the 1950s: an era in which computers were in their infancy and artificial intelligence was still a distant dream. If in the previous two centuries scientists such as Thomas Bayes, Adrien-Marie Legendre, and Pierre-Simon Laplace had laid the foundations for future research, only with the work of Alan Turing did the concept of a machine capable of learning finally materialize.
“In such a case, one would have to admit that the progress of the machine had not been foreseen when its original instructions were put in. It would be like a pupil who had learned much from his master but had added much more by his work. When this happens, I feel that one is obliged to regard the machine as showing intelligence. “
In Italian: “In this case, it should be admitted that the advancement of the machine was not foreseen when the original instructions were inserted. It would be like saying that a pupil has learned a lot from the teacher but has added even more with his work. When this happens, one is obliged to recognize that the machine shows signs of intelligence. “
Alan Turing in a lecture in 1947. (Quoted by BE Carpenter and RW Doran (eds.), AM Turing’s Ace Report of 1946 and Other Papers )
In 1950 Turing developed the Turing test, still known today: it is a game in which the computer pretends to be a human with another human being. If the machine can convince the human being that it is talking to a natural person, it has passed the test. Two years later, Arthur Samuel developed a computer that knew how to play checkers and improved his skills with every game: the program could learn.
Finally, in 1957 Frank Rosenblatt developed Perzeptron, the first algorithm that could be taught, thus essentially an artificial neural network. From then on, researchers began giving their computers increasingly complex logic exercises to solve. The machines solved them sometimes better, sometimes worse.
Meanwhile, large companies have become the main financiers of the development of machine learning. With Watson, for example, IBM has developed a computer that has an immense archive of information and can answer questions that are asked in natural language.
To give the world a demonstration of its abilities, in 2011, the computer was invited to participate in the famous television program “Jeopardy,” from which it came out as a winner. This event is reminiscent of the 1997 chess match between world champion Garri Kasparov and another IBM computer, Deep Blue. Again it was the machine that beat its human opponent.
Google and Facebook use machine learning to understand their users better and offer them more functions. DeepFace, Facebook’s computer, can now identify faces on an image with 97% accuracy. Thanks to its Google Brain Project, the search engine giant has dramatically improved the voice recognition of the Android operating system, photo search on Google+, and video recommendations on YouTube.
What Is Machine Learning?
Essentially machines, computers, and programs only function in the way previously established for them: “In case of A, do B”. Our expectations of modern computer systems are getting higher and higher; nevertheless, programmers cannot predict all conceivable cases and consequently teach computers all their respective possible solutions.
For this reason, the software must be able to make decisions independently and to react appropriately to unknown situations.
For this purpose, however, there must be algorithms that allow the program to learn. This means that the first step is to fill the machine with data; subsequently, the computer must understand its structure and create a model to make associations.
When we talk about self-learning systems, we also think of terms related to them: it is essential to know them to understand the whole world of machine learning better. Below we provide a list with definitions of the most important ones.
Artificial intelligence (AI) research attempts to create machines capable of acting like humans: computers and robots must analyze their environment and then make the best decision. They must therefore behave intelligently, according to the criteria of human beings. And here arises the problem of defining these criteria, even if it is not yet clear according to which parameters we should evaluate our intelligence.
Currently, artificial intelligence, or what is considered as such, cannot simulate a human being in its entirety (including emotional intelligence). Instead, partial aspects are isolated to be able to solve specific tasks. This is commonly referred to as weak artificial intelligence.
Neuroinformatics is working at computer design based on the model of the human brain. This branch of artificial intelligence research considers nervous systems from an abstract point of view, i.e., free of their biological properties and limited to their mode of functioning. Artificial neural networks are primarily abstract mathematical processes rather than actual manifestations. Therefore, a network of neurons is woven, made up of mathematical functions or algorithms, and which, like a human brain, can cope with complex tasks. The connections between neurons are differently powerful and can adapt to problems.
The term “big data” refers to a massive amount of data. However, there is no definite point from which we stop talking about data and start talking about big data.
The fact that this phenomenon in recent years has enjoyed significant media coverage is due to the origin of this type of data: in many cases, the flow of information consists of user data (interests, travel profiles, vital data) collected by companies such as Google, Amazon, and Facebook to adopt more precisely to their users.
Such amounts of data can no longer be evaluated satisfactorily by traditional computer systems: after all, such software can only find what the user is looking for. For this reason, self-learning systems are needed, which can discover connections unknown until then.
With data mining, we define the analysis of big data. The collection of mere data has in itself a relative value. However, the accumulation of information becomes interesting when the relevant characteristics are extracted and evaluated similarly to how gold is extracted. Data mining differs from machine learning because it relies on applying recognized models rather than finding new ones.
Different Methods Of Machine Learning
Essentially researchers make a distinction between supervised learning and unsupervised learning through step-by-step intermediate levels. The algorithms used for this purpose are very different from each other. In supervised learning, the system is fed with examples. The developers indicate which value to associate the corresponding information with, for instance, whether it should belong to category A rather than B.
From this, the self-learning system deduces certain information and draws a model that it will be able to recognize in the future, so it will be able to handle unknown data better. The main aim is to minimize the error rate more and more.
A well-known example of supervised learning is spam filters: based on specific characteristics, the system decides whether the email should end up in the inbox or the spam mailbox. Should the system make a mistake, you can later manually set the parameters again, which the filter will refer to in the future.
This way, the software will get better and better results. This filtering program is based on the Bayes theorem, for which it bears the name of “Bayesian filter”.
Concerning unsupervised learning, unsupervised learning, the teacher’s figure is no longer present, who in supervised learning has the task of instructing the system and giving feedback to the decisions that the machine has made. Autonomy. Instead, the program attempts to recognize recurring patterns on its own. To do this, for example, it has the possibility of using clustering.
The collection of all the data then selects an element whose characteristics are analyzed and compared with those already examined. If the program has already analyzed similar items, then the current object will be added to them; otherwise, it will be isolated.
Systems that rely on unsupervised learning are also implemented in neural networks. Examples of applications in network security are found: A self-learning system can recognize abnormal behavior here. Although, for example, it is not possible to associate a cyberattack with a known group, the program can still acknowledge the presence of a threat and therefore proceeds to raise the alarm.
In addition to these two main directions, there are also partially supervised learning ( semi-supervised learning ), reinforcement learning, and active learning. These three different methods belong more to supervised learning, and nature and forms of user involvement distinguish them.
A distinction should also be made between so-called shallow learning and deep learning. While the former is a relatively simple method where the results are rather superficial, the latter is more difficult to understand. The complexity of these data is because they are natural information, i.e., those that appear in speech recognition, written language, or the face. For humans, biological data is easy to process, while it is not for a machine, as it is difficult to assimilate mathematically.
Deep learning and artificial neural networks are closely related to each other. The way neural networks are trained can be defined as deep learning: deep because the network of neurons is organized on different hierarchical levels. The first level consists of incoming neurons that detect the data, start with their analysis, and send the results to the next neural node. Eventually, the information, each time more and more precise, reaches the output level, and the network emits a value. The groups, on the whole, very numerous, which are located between the entrance and exit, are called hidden layers.
To better explain deep learning, Google’s image search can be used as an example. The network behind the search algorithm provides the query “cats” only with images that depict domestic cats. This mechanism works because Google’s self-learning system can recognize objects within the image. When Google registers a new idea in its catalog, the system’s entry neurons process the data; after all, according to computers, images are composed only of numbers.
On its way through the various levels, the network filters only the necessary information until it finally decides which objects are in the image (for example, a cat). In the training phase, the developers also provide a category for each image so that the system can learn. If the machine were to provide false results, then believe that images with dogs are the right ones to search for “cats”, the developers can adapt individual neurons. Like the human brain, the latter have different weightings and thresholds that can be fine-tuned in a self-learning system.
How Does Machine Learning Work In Marketing?
Machine learning is already of great use to marketing today; mainly large companies use this Technology, first of all, Google. Self-learning systems are still so new that they cannot simply be purchased as out-of-the-box products yet. In their place, the giant Internet companies develop their strategies and play a pioneering role in this area. Since some, despite the commercial interest, still follow an open-source approach and collaborate with independent research, advances in machine learning are increasingly rapid.
Alongside the creative one, marketing also always has an analytical aspect: statistics on user behavior (purchasing behavior, number of website visitors, use of apps, etc.) play an essential role in the decision of specific advertising measures. .
Usually the rule is that the more significant the amount of data, the more information can be extracted. To be able to come up with such a mountain of features, you need intelligent programs. This is where self-learning systems step in: educated programs recognize recurring patterns and can make reliable predictions, unlike humans, who are often biased towards data.
Usually, those who analyze the data start with expectations that end up being natural prejudices, practically inevitable for human beings, which are often a reason for distorting the results. The more data analysts review, the greater the divergences in the interpretation of the data. Even though intelligent machines themselves may have biases because humans themselves have unknowingly instructed them to have preferences, they act with greater objectivity than experts in the flesh when it comes to hard facts. It can therefore be said that machines provide more reliable analyzes.
Self-learning systems improve and simplify the representation of analysis results, as happens in automated data visualization, a technology in which the computer independently chooses the correct presentation of data and information. This is particularly important because it allows humans to understand what the machine has discovered and predicted. With substantial data streams, it becomes difficult to represent the results of the assessments on your own. For this reason, the visualization must also take place via computer calculations.
However, machine learning can also influence content creation – as is the case in generative design. Instead of designing the same journey for all users, consisting of the customer’s steps to purchase a product or service, dynamic systems can create personalized experiences thanks to machine learning. Although copywriters and designers still provide the content viewed by the user on a website, it is up to the system to integrate the specific components for the user.
In the meantime, self-learning systems are also used in the design: for example, with the Project Dreamcatcher, an innovative CAD system created by the research group of the Autodesk company, it is possible to have components designed by a machine.
Machine learning can also be used to organize chatbots better. Many companies already employ programs today that carry out a portion of customer support with the help of a chatbot. However, in many cases, users quickly get annoyed with robots: the capabilities of current chatbots are usually minimal, and the possible answers refer to a manually managed database. A chatbot that is based on a self-learning system and has good voice recognition can convey to customers the feeling of really communicating with a person and, therefore, passing the Turing test.
Amazon and Netflix demonstrate another benefit of machine learning for marketers: recommendations. Among the essential success factors of these large companies is also predicting what the user will want to have. Self-learning systems can recommend other products to the user regardless of the data collected. What was previously only possible on a large scale (“Our customers like product A, so most of them will also be interested in product B.”) is now also possible on a smaller scale thanks to modern programs (“Customer X liked products A, B, and C, so she will probably also be interested in product D.”).
In summary, it can be seen that self-learning systems will influence four different aspects of online marketing:
Programs that work with machine learning and that have been well educated can process vast amounts of data and make predictions for the future. Marketers can draw more accurate conclusions about the success or failure of advertising campaigns and measures.
Analyzes take time if done manually. Thanks to the self-learning systems, the speed of work increases so that you can react more quickly to changes.
Through automatic learning, it is easier to automate processes. As modern systems can autonomously adapt to new circumstances with the help of machine learning, even complex automation processes become possible.
The software can assist countless customers. Since self-learning systems detect and process data from the individual user, they can comprehensively deal with them. Personalized advice and specially developed customer journeys help you apply marketing measures more effectively.
Other Fields Of Application Of Self-Learning Systems
Marketing isn’t alone in finding different uses for machine learning: self-learning systems are also effective in many other areas of human life. In part, these contribute further to the advancement of science and Technology. In some cases, they are used in more or less large gadgets to simplify everyday life. The fields of application presented are just some of the possible examples. After all, it cannot be excluded that machine learning will be present in all aspects of our life in the not too distant future.
What is true of marketing has even more significance in science. Intelligent big data processing dramatically eases the work of empirical research. For example, particle physicists can detect and process much more data and ascertain any anomalies thanks to self-learning systems. But machine learning can also help in medicine: some doctors are already using artificial intelligence to make diagnoses and treatments. In addition, machine learning is also helpful for the prognosis of diabetes or heart attacks.
Nowadays, robots are omnipresent, especially in factories: for example, they are used in mass production to automate work steps that are always the same. However, these are not truly intelligent systems, as they are only programmed to perform a single specific action. When self-learning systems are used in robotics, they need to be able to solve new tasks. Of course, these advances are also of great interest to other sectors: from space travel to housework, robots equipped with artificial intelligence can take action in many areas.
Autonomous cars are an excellent showcase for machine learning. Only machine learning makes cars move autonomously and safely in traffic, rather than just on test drives. As it is not possible to program all possible situations, autonomous vehicles must refer to intelligent machines. But self-learning systems don’t just revolutionize means of transport in traffic: intelligent algorithms, for example, in the form of artificial neural networks, analyze traffic and develop more efficient ways of managing it, such as by smartly turning on traffic lights.
Machine learning already plays a significant role for the Internet. We mentioned spam filters: through constant learning, the filters are refined and can recognize unwanted e-mails better and better to banish them from the inbox in an increasingly reliable way. The same is also valid for combating viruses and malware: thanks to new intelligent technologies, computers are better protected from malicious software.
Even the ranking algorithms of search engines, most notably Google RankBrain, are self-learning systems. Even if the algorithm does not understand user input (because it has never been searched for by anyone so far), it can infer and propose inherent to the request.
Even within your own four walls, intelligent computers are becoming present: hence, ordinary homes become smart homes and intelligent homes. The Moley Robotics company, for example, develops a clever kitchen equipped with mechanical arms that prepare meals. Even personal assistants such as Google Home and Amazon Echo, thanks to which it is possible to monitor systems and devices in your home, use machine learning technologies to understand their users’ needs better.
But many people don’t give up on their assistants even when they’re out and about: Siri, Cortana, and Google Assistant allow users to give commands and ask questions to their smartphones via voice command.
Since the beginning of artificial intelligence research, the ability of machines to play has always been an excellent stimulus for researchers. Self-learning systems have been tested in chess, checkers, and even go, the well-known Chinese board game among the most complex globally, challenged by human competitors. Game developers also use machine learning to make more engaging projects. Game designers can use machine learning to create the most balanced gaming experience possible and make virtual opponents better adapt to the behavior of human players.