With each study of the human brain, neurosciences absorb the main headlines in the press. All progress generates a question in which both hopes and concerns are linked: will it ever be possible to technically imitate the human brain? Although computers already surpass imagination with their processing power, in their complexity the human brain remains superior in many areas. Will this change at any time?
These questions fall within the field of artificial intelligence (AI) research. The main objective of this discipline is to try to technically recreate the human brain and its functions through computing, neurology, psychology and linguistics. In addition, AI research approaches have also revealed information about self-perception as human beings and the understanding of “intelligence”.
Artificial intelligence, which has its own free will and acts autonomously, remains a fantasy. However, in many areas of everyday life this visionary technique already plays a central role in many areas and, in most cases, goes unnoticed. Both the definition of artificial intelligence and its applications are unknown to many. In the field of medicine, it is applied to develop diagnoses and treatment plans, and thanks to AI, market forecasts are much more powerful and, with its help, Google’s search algorithms are much more dynamic. This technology is behind personal assistants such as Cortana or Siri, each car that drives itself and each computer that is responsible for the selection of new employees. In the United States, some legal documents are already being created with the help of artificial intelligence. All this makes it evident that AI research has found varied and useful areas of application in the last decades.
The Internet in general, search engines in particular, and therefore online marketing are affected by these rapid innovations. Thus, a basic understanding of AI technology is also beneficial for SEO. Some of the questions that arise then are: What is artificial intelligence and how does it work? What are the objectives of this research and what current applications are known? What opportunities and risks does it involve? And finally, what is the impact of its evolution for online marketing and SEO?
- Definitions of artificial intelligence: vision and reality
- How does artificial intelligence work ?: AI methodology and history
- Risks and opportunities of artificial intelligence
- Artificial intelligence in the digital world
- Definitions of artificial intelligence: vision and reality
- What is artificial intelligence?
One of the definitions of artificial intelligence identifies it as a branch of computing that aims to create a technical equivalent of human intelligence, and not only do computer scientists work, but also experts from other fields of knowledge. For their part, there are many theories and methodological approaches to determine the characteristics of “intelligence” and the ways to simulate it.
Achieving a more precise definition of artificial intelligence is impossible, mainly due to the complexity per se of the concept “intelligence”. If for humans those abilities that are considered part of intelligence are already controversial, determining them for machines is even more complex. Some of the questions involved include whether the machine should be primarily optimized in its rationality or whether other human traits such as intentionality, intuition, or the ability to learn should be included rather. Possibly social skills, empathy, or a sense of responsibility may also be expected to play a relevant role. Thus, the question revolves around what technology must create: rational capacities or artificial humanity.
However, differences also arise regarding the relationship of similarity with humans. Should the machine be designed like a human brain? This simulation approach aims to create an exact replica of brain functions. Or is the machine rather expected to act like a human, that is, the end result to be as identical as possible? This phenomenological approach tries, above all, to determine exactly what people would receive from artificial intelligence, regardless of the technical process that has to be carried out.
Defining artificial intelligence is a difficult task. In 1950 mathematician Alan Turing developed a test to measure artificial intelligence. With a series of questions, the Turing test determines if a machine is recognizable as such. If the machine’s responses cannot be distinguished from a person’s, then its artificial nature is confirmed. However, for current AI technology, this definition is not very helpful, since artificial intelligence is developed especially for technical areas today. In this case it is less about the machine being able to communicate, but rather that it performs highly specialized tasks efficiently. For these types of technologies, a restricted version of the Turing test is used, in which, if in a same area a technical system has the same capabilities as a human being (such as a medical diagnosis or a game of chess), this It is considered as an artificial intelligent system. At this point, then, two definitions of artificial intelligence emerge: one “strong” and the other “weak”.
The vision: strong artificial intelligence
The strong definition of artificial intelligence refers to an intelligence that is capable of replacing people in their entirety with their various capacities – this universal approach to man as a machine has existed since the Enlightenment, but remains fiction.
There are several dimensions of intelligence that are part of strong artificial intelligence: cognitive, sensory-motor, emotional, and social. Most of the current applications of artificial intelligence are related to the area of cognitive intelligence, that is, logic, planning, problem solving, self-sufficiency and the individual construction of perspective.
The vision is that artificial intelligence could develop an autonomous consciousness and a will of its own. With this long-term goal, AI research enters the traditional territory of philosophy and raises many ethical and legal questions. In fact, the question of the legal capacity of smart machines remains totally confusing.
The reality: weak artificial intelligence
For its part, the weak definition of artificial intelligence focuses on the development and implementation of artificial intelligence in clearly defined application areas. This is precisely the point at which current AI research is located, whose application areas, almost entirely, are today in the field of weak but widely specialized AI – such as, for example, the development of cars. freelancers, the development of medical diagnoses or the creation of intelligent search algorithms.
The field of weak artificial intelligence research has been able to record progress in recent years. The development of intelligent systems in individual areas turned out to be much more practical and, at the same time, ethically of greater applicability than the investigation for the development of a “superintelligence”. The application areas of weak artificial intelligence are very varied, being particularly successful in sectors such as medicine, finance, the transport and marketing industries and, of course, the Internet. It is already foreseeable that, at some point, technologies of this type will penetrate almost all areas of life.
How does artificial intelligence work ?: AI methodology and history
How to describe the operation of artificial intelligence? An AI is only as good as the nature of its technical representation of knowledge. Here two basic methodological approaches emerge: the symbolic and the neural.
In symbolic AI knowledge is represented by symbols, working with so-called symbol manipulation. Symbolic artificial intelligence refers to information processing “from above” and operates with symbols, abstract contexts, and logical conclusions.
Neural AI represents knowledge by its artificial neurons and their connections. Neural artificial intelligence refers to information processing “from below” and simulates individual artificial neurons that organize into larger groups, forming an artificial neural network.
Symbolic artificial intelligence
Symbolic AI is considered the classic approach to artificial intelligence. This is based on the idea that human thought can be reconstructed from a higher conceptual-logical level, regardless of specific experimental values (top-down processing strategies). Thus, knowledge is represented in abstract symbols, including spoken and written language. Based on algorithms, machines learn to understand and use symbol manipulation. The intelligent system gets its information from so-called expert systems. In these, symbols and information are classified into a specific type of classes, especially in conditional logical relationships of the form “yes … then”. The intelligent system can access this knowledge database and compare this information with yours.
Traditional applications of symbolic AI are word processing and speech recognition, as well as other logical disciplines for, for example, mastering a game of chess. Symbolic artificial intelligence works according to fixed rules and can solve more complex problems as computing power increases. The most recognized example was when in 1996, with the help of symbolic AI, IBM’s Deep Blue won against world chess champion Garri Kasparow.
- Building an expert system is based on data equipped with certain processing rules. An example:
- All trees are made of wood.
- Wood is flammable.
- X is a tree.
- So X is flammable.
Based on such logical connections, the expert system is intended to mimic people’s knowledge. Expert systems are almost always limited to one specialty, such as medicine.
Symbolic AI performance depends on the quality of expert systems. In the early days, developers had high hopes that, with the advancement of technology, expert systems appeared to be more powerful, and with that, the dream of strong artificial intelligence seemed tangible. However, the limits of symbolic AI became increasingly apparent because, beyond the complexity of the expert system, symbolic AI remains relatively inflexible. With exceptions, variables, or uncertain knowledge, a strong system is barely manageable. Furthermore, symbolic AI has many limitations when it comes to autonomous knowledge acquisition.
Being too rigid and not very dynamic, this technology fails to meet expected expectations. As a consequence, in 1970 the so-called “AI winter” takes place, which lasted until the early 1980s and in which financial support for AI development plummeted. In this context, there is a revolutionary reorientation of this technology: the development of machine learning systems. Working on artificial neural networks revives AI research.
It was Geoffrey Hinton and two of his colleagues who in 1986 revived the investigation of neural AI and, with it, that of artificial intelligence in general. With the development of its backpropagation algorithm, the foundations of deep learning were created, with which almost every AI system works today. Thanks to this learning algorithm, deep neural networks can constantly learn and grow at their own pace, thus overcoming the challenges of symbolic AI.
Neural artificial intelligence (also known as subsymbolic or connectionist AI) says goodbye to the principle of representing symbolic knowledge. As in the human brain, knowledge is segmented into small functional units (artificial neurons) that connect with larger groups (bottom-up or bottom-up approach). The result is a branched and diverse network of artificial neurons.
Neural artificial intelligence tries to mimic the brain’s functioning as precisely as possible, as well as artificially simulating its neural networks. Unlike symbolic AI, the neural network is “trained” – in robotics, for example, with sensory-motor training data. Thus, based on these experiences, AI generates increasing knowledge. The great motivation is that, while your training is relatively time consuming, the system is capable of automatic learning. Therefore, there is also talk of machine learning systems or Machine Learning. This turns neural artificial intelligences into highly dynamic and adaptive systems that, in many cases, are not fully understandable to humans.
However, the construction of an artificial neural network almost always follows the same principles:
Numerous artificial neurons are layered on top of each other. They are connected by simulated connections.
Currently, deep neural networks are already in place. “Deep” means that they work with more than two layers. The intermediate layers are stacked hierarchically, in some systems the information is sent upwards through millions of connections. As an example: AlphaGo (Google DeepMind) has 13 intermediate layers, Inception (Google) has more than 22 layers.
The top layer or input layer works like a sensor, this includes any input in the system, be it text, image or sound. From this point on, the inputs follow specific patterns in the network and are compared to the previous input. That is, the network is trained and fed through the input layer.
In general, the deepest layer or exit layer has only a few neurons, one for each category to classify (image of a dog, image of a cat, etc.). The output layer shows the user the result of the neural network and can, for example, recognize the image of a cat that was previously unknown to him.
There are three basic learning methods for training neural networks: supervised, unsupervised, and reinforcement learning. These methods regulate, in different ways, how an input leads to the desired output in a system.
The vast majority of AI’s most recent achievements are due, in part, to neural networks. In the context of research into innovations and under the concept of deep learning, we are committed to the extraordinary achievements of machine learning systems – be it in speech and handwriting recognition or in autonomous cars. Thanks to deeper neural networks, AlphaGo, Google’s DeepMind, defeated Go’s South Korean professional gamer Lee Sedol in 2016 (Go is one of the most complex strategic board games in the world).
Google’s Inception, actually an image recognition system, creates amazing visual daydreams that in 2015 went viral with the hashtag #DeepDreams. This “side effect” of the system was discovered by its developers randomly as they searched to find out exactly how artificial intelligence works.
Risks and opportunities of artificial intelligence
From the blind optimism of progress to the simple negation of technology, smart technology elicits a wide range of reactions. This mainly has to do with future positive and negative predictions about how these technologies could change people’s lives. What opportunities and risks are associated with artificial intelligence? Here are some of the top positions for AI enthusiasts and skeptics.
The range of advantages offered by artificial intelligence is very wide. The main benefits of this technology are related to the world of work, its high performance and its economic prospects.
“We will be surrounded by machines that will produce faster, at a lower cost, with better use of resources. In the medium term they will exceed human capacities. In the first age of computing, when calculations were made and Moore’s Law [the power of computers doubles every two years] was accepted, the price of components dropped and their computing power increased. That is going to keep happening. For example, in the case of cars, humans are not going to manufacture them or drive them.
Proponents of this new technology point out the main advantages provided by artificial intelligence:
Employment and simplification of work: New technology could provide valuable jobs and ensure a global economic boom. Experts agree that the technology could have drastic effects on the labor market. A committee at Stanford University studied the future prospects for artificial intelligence and concluded that it is currently impossible to estimate whether the impact on the market will be positive or negative. However, it is highly likely that people will have to earn a living through means other than work. This is why supporters of universal basic income see artificial intelligence technology as a great opportunity: the traditional wage labor model could become obsolete. Even for Tesla CEO Elon Musk, one of the greatest benefits of artificial intelligence is the possibility that humans have more free time.
Comfort: Advocates of artificial intelligence recognize the potential of these changes in improving everyday comforts. From the autonomous car to the machine translation software: both events can be a great relief for consumers.
Extraordinary performance: Artificial intelligence can also have a positive impact in areas of public utility, after all machines have a lower error rate than people and their performance is enormous. Thanks to the great versatility of smart machines, AI is especially promising in sectors such as health and justice. Although experts do not expect that in the future, judges will be replaced by artificial technology, AI can help to identify patterns in a faster process and, therefore, to approach more objective judgments.
Thus, artificial intelligence could represent a major economic boost for the IT industry and adjacent economic sectors and, as a consequence, increase overall prosperity.
Futuristic projects: Last but not least, artificial intelligence inspires humans’ natural desire to discover, an approach that is already in place for exploring for oil wells or for controlling robots on Mars. It could be assumed that the areas of application of artificial intelligence will continue to expand with the progress of technology itself.
Some eminent experts like physicist Stephen Hawking or Silicon Valley icon Elon Musk also warn against the risks of artificial intelligence. These critical voices find support in important initiatives such as the research and lobbying organization Future of Life Institute (FLI), which is in charge of mobilizing renowned critics to raise awareness about the responsible use of technology.
Some of the most discussed artificial intelligence risks are:
Inferiority of humans: A possible risk that many fear and that is usually a central theme of science fiction is the development of superintelligence. Under this term we understand a technology that optimizes itself and, therefore, that is independent of the human being. The relationship between people and this superior technology could be problematic and the fear of skeptics lies in the fact that at some point the machine surpasses man. However, researchers find it virtually impossible for deliberately malicious artificial intelligence to exist. On the other hand, a risk that does seem real is one where AI is so competent that its activities become independent – activities that could later become dangerous for human beings. Additionally, there is great disagreement on whether or not control of AI technology could be lost. Future of Life reports on its page about myths and misconceptions about superintelligence.
Technological dependence: Other skeptics do not see such inferiority as a risk, but rather recognize the danger of man’s increasing dependence on technological systems. Critics claim that the fact that robots are already being tested in nursing in healthcare, makes man become an object supervised by technical systems. In this process, users risk losing a part of their private lives and self-determination. These concerns are not only related to the medical sector, but also to AI-based video surveillance systems and intelligent network algorithms.
Data protection and power division: Smart algorithms can process growing data sets more efficiently. Especially for internet commerce this is a positive point. For critics, data processing using AI technologies is going to be increasingly difficult for consumers to understand and control. However, given the necessary resources and knowledge, it would be the companies and the experts who would have exclusive control of said information. Logically these are not exclusive risks of artificial intelligence, but rather problems of the digital age. Now, given the amazing capabilities of AI technologies, the warning voices are getting louder.
Bubble filters and selective perception: Cyber activist Eli Pariser points to bubble filters as an additional risk of artificial intelligence. The danger is that, increasingly, search algorithms present the user with information derived from their previous behaviors (personalized content), thus offering an increasingly skewed view of the world. Skeptics say that AI technologies promote selective perception and intensify an increasing “ideological distance between individuals.” In 2016, Microsoft published a report on this model of divergence in access to information through filters.
The results of such a study relativize this artificial intelligence risk, pointing out that this is a problem that already exists in traditional journalism and that the impact of new technologies is still only partially demonstrable.
Influencing opinion: Furthermore, according to critics, AI technologies could also consciously influence public opinion. The reason for this concern is those technologies that know their users down to the smallest detail or the use of social robots that influence public attitudes. Detractors of AI argue that with the increased intelligence of these techniques, the risk of determining collective opinion is increasing.
Weapons technology: Another important risk of artificial intelligence is its application in the military field. In 2015, hundreds of researchers and scientists, supported by FLI, warned about AI-based autonomous weapon systems. Those who participated included Stephen Hawking and Elon Musk, but also Apple co-founder Steve Wozniak and DeepMind co-founder Demis Hassabis. In an open letter they called for a ban on all AI-based weapons technology that could be used without “serious human control.” The fatal combination of, on the one hand, artificial intelligence and war and nuclear threat on the other, is repeatedly pointed out by other parties as well.
Labor market: The disputed risks of artificial intelligence in the labor market mainly refer to the loss of jobs. Skeptics fear that AI technology could turn people into unnecessary beings, whether by cleaning robots, nursing robots or by autonomous transport systems. In medical ethics, the use of robots is highly controversial. The fear is that the supply of robots that take care of health care tasks may lead to social coldness, especially in the last phase of human life.
Discriminatory algorithms: One of the many advantages of artificial intelligence is that, compared to people, artificial technology provides neutral results much more often. However, the AI technique has also shown time and time again people’s prejudices against gender or ethnicity. Tay, the Microsoft chatbot, quickly mimicked the racist language of users, surveillance technologies classify “black neighborhoods” as problematic districts, and job search platforms show male users those best-paying deals. This problem is not a secret, which is why the British Standards Institute published a revised version of the ethical guidelines for robots.
Artificial intelligence in the digital world
What then is the role of artificial intelligence in the digital world? First, it should be noted that for the inexperienced user, artificial intelligence on the Internet is barely recognizable. Many companies also avoid using the term even if their products work hand in hand with AI. This is because the rise in fascination with artificial intelligence is directly proportional to its bad reputation. Consumers are often skeptical of the implementation of AI technologies in everyday life. It is also sometimes difficult to define when a technical service can be defined as “intelligent”, as the smooth transitions of the implementation forms and the different definitions of artificial intelligence often make it even more confusing.
With the widespread use of voice assistants, people are more likely to get used to using artificial intelligence. However, on the Internet there are also a wide variety of applications in which AI technology plays a crucial role. The list of active AI algorithms and programs that use artificial intelligence is extensive. Google dominates this market with its innovations, presumably with a two to three year development advantage compared to other companies. Now, how exactly is artificial intelligence integrated into the well-known search algorithms? And what impact does it have for online marketing and especially for SEO? Here are some examples of typical techniques and innovative programs in the sector.
Techniques and applications
Machine learning: Machine Learning means that an artificial system gains knowledge from experience. This learning data enables the system to recognize patterns and regularities. Machine learning uses both symbolic and neural artificial intelligence.
Deep Learning: Deep Learning is a secondary machine learning class that works exclusively with neural AI, more specifically with artificial neural networks. Deep learning is the foundation of most of today’s AI applications.
Visual classification: used to develop recognition of text, objects, faces and symbols.
Auditory classification: it is implemented in voice and sound recognition.
Social Computing: it is in charge of analyzing different types of digital content (blogs, social networks, online games or wikis). Patterns and rules of social behavior are deduced from the results. Thanks to social computing artificial social agents are developed.
Opinion Analysis: “Opinion mining” (also called “Sentiment Analysis”) refers to the methods by which the Web is searched for the opinions and feelings of users. The data obtained is then used to determine the users’ points of view about specific topics, events and people. Opinion analysis enables automatic and personalized processing of customer inquiries.
Customer support (phone, web) and digital assistants: In the support area, advances in AI play an important role. In particular, voice recognition software works with artificial intelligence.
Search algorithms: Artificial intelligence is one of the many components with which search algorithms are optimized and its importance for ranking is constantly increasing.
Crawlers: Crawlers are used by search engines, among other things, to search for information on the Internet with which an index is created. A tracker learns from the examples and becomes able to draw relevant conclusions.
Computer vision systems: Machine vision, especially facial recognition, is often used in the field of security technology, such as in traffic or in public spaces. Services such as Facebook have implemented it to identify their users more easily. Facebook is currently capable of finding a face in millions of photos in seconds, even if the person is not looking directly at the camera.
Virtual Actors and Bots: In the development of computer games, AI allows virtual players to act more humanely. So-called bots have been developed to simulate human activity on the Internet. In particular, social bots especially act as artificially intelligent robots.
Group simulation: thanks to artificial intelligence it is possible to predict the behavior of groups of people. This is used both in video game development, in security technology or in the analysis of viral dynamics.
Effects on SEO
Innovations in machine learning systems represent a wave of great change for the industry. With the acquisition of DeepMind, Google, the pioneer of search engines, demonstrated that its search algorithm aims to increasingly specialize in the field of artificial intelligence. Increasingly, Google is taking on new AI research companies, such as the British companies Vision Factory or Dark Blue Labs, and integrating them into its DeepMind team.
The biggest impact of Google’s AI initiatives is its intelligent RankBrain algorithm. The extremely efficient algorithm for detecting new searches was implemented worldwide in 2015 and slowly, along with links and content, has become one of the top three factors in the ranking. RankBrain’s specialty is turning text searches into mathematical entities. In this way, the intention behind the search can be more easily recognized. However, how exactly this form of artificial intelligence works remains a mystery.
The influence of AI on Google search is undeniable. SEO expert Mark Traphagen quoted CEO Sundar Pichai as saying, “We are moving from a mobile first world to an AI first world. We want to create a personalized Google for each of the users ”(freely translated from English, see image). Creating a Google for each person, conceiving a complete individualization of searches through artificial intelligence is a huge challenge for SEO
RankBrain artificial intelligence classifies searches to the extent that it converts those data that are already known to it into hypotheses and generalizations and applies them to each respective entry. Its behavior changes because it is constantly fed with new data. Thus, Google no longer works with weekly updates made from a human measurement, but instead relies on real-time calculations from machine learning systems. The fact that algorithms were previously partly identifiable and that artificial intelligence is now characterized by its dynamism and personalization makes SEO very difficult.
Below we explain with some basic concepts the role of artificial intelligence in SEO. Something that every SEO specialist should keep in mind is that, on a daily basis, artificial intelligence acquires knowledge about the quality of web pages that it will surely apply to future rankings based on the user’s experiences and indications. Google knows where users click, what links they use, how long they stay on a page, and how likely they are to react to ads. The following tips can be of great help to SEO:
1. User signals are very relevant: It is no longer just about clicks, but also the length of time or, as some studies indicate, social signals. Four factors are decisive in this regard:
Time on site: average time spent on a site.
Bounce rate: The bounce rate includes both short visits to a site and those in which the user only opens one of the pages.
Click Through Rate: The click rate is the frequency with which users click on advertising banners or sponsored links.
Social Signals: Social signs are Likes, Shares and comments on a web page or its contents.
2. Semantics before keywords: The RankBrain was originally developed to better understand the longest or most unknown searches up to then. As a result, Google increasingly interprets everyday language and the intent behind it. Therefore, SEO has shifted its focus to semantics rather than classic keywording. In other words, for a good search engine ranking, the quality and relevance of the content of a web page are increasingly relevant.
3. Google recognizes user satisfaction: Google evaluates user signals and therefore ranks the quality of the site more accurately compared to how algorithms did before RankBrain. Therefore, it is more important than ever to ensure a high level of usability for the user. Page speed is also crucial to user satisfaction and retention. Something that is especially important is the navigation menu, since both the user and the AI must be able to locate themselves easily. In short, perfection in content and technique is equally demanded.
4. Include all the departments of your company in online marketing efforts: The bigger a company and the more efforts it invests in its online presence, the greater its team of professionals in marketing, SEO, social networks and UX managers. Whoever wants to innovate or respond appropriately to artificial intelligence changes has to join forces.
Although the rankings are becoming more flexible, the good news is that search engine optimization doesn’t change much. For example, excessive use of keywords is no longer considered the only measure of SEO success. Slowly, and increasingly, the industry has been focusing on user satisfaction and focusing on your target group enjoying and visiting their website as often as possible.
The impact of artificial intelligence on rankings does not differ much from classical algorithms. AI-based algorithms do not necessarily work differently, but are more effective and accurate, recording more than is really relevant to Internet users. As a consequence, existing SEO strategies should not be discarded, but rather adopted with greater expertise.
It only remains to say that in 2016 five Silicon Valley digital giants (Google, Amazon, Facebook, IBM and Microsoft) joined forces to boost their research in artificial intelligence. The sensitive nature of this news regarding consumer privacy did not take long to alert the community because, after all, these companies own most of the total data set. Furthermore, this initiative is obliged to develop common ethical guidelines for the treatment of artificial intelligence and the need for common ethical principles in this area is undeniable. Driving the development of AI technology towards more profitable paths will be the central task of society in the coming years and decades.