Can AI be defined as advanced intelligence

What is Artificial Intelligence (AI)? Definition and examples

While buzzwords like AI and big data are on everyone's lips, it is often not clear what exactly artificial intelligence is and what examples there are for its application. Put simply, AI is the simulation of human competence through automated systems. This includes a range of methods from simple algorithms to signal processing such as image processing and machine learning. The opportunities created by artificial intelligence cover a range of automation, optimization and support for people in their activities and allow us to concentrate on creative and highly complex, novel tasks.

Table of Contents

What is Artificial Intelligence (AI)? A definition.

Artificial intelligence is the ability of a system to simulate human behavior. This imitation can take many forms: From a robot to software to a physical object, many things can give the impression that it has cognitive abilities. The term is usually used to refer to self-learning algorithms in computer science, more specifically in the area of ​​data science and machine learning (ML).

In addition to this general definition of Artificial Intelligence (AI), there is also the general distinction between strong and weak AI. These categories are important to discuss because the state of research and business cannot compete with images from media and literature - especially science fiction.

What is strong / general AI?

The Terminator, HAL 9000, the Matrix, or the humanoids in “I, Robot” are all examples of strong artificial intelligence. Strong AI can identify, evaluate and solve problems independently - regardless of the area. She constantly acts under new conditions and can easily adapt to new situations - just like humans.

But in reality we are still a bit removed from these images. This generalized, general artificial intelligence requires much bigger leaps in research and development than is currently possible. The strong AI has to “learn to learn”, which is a great challenge due to the flexibility required.

The principle in the area of ​​ML that comes closest to a strong AI is the area of ​​reinforcement learning. Here an agent (computer program or, for example, a robot) is placed in an environment and learns to fulfill a goal as optimally as possible by means of fixed actions (e.g. moving forward and grasping). But here, too, the framework conditions are predetermined based on the environment, actions, goal and reward, and thus the complete variability of human action is far from given.

What is weak / specialized AI?

In contrast to the strong AI is the weak AI. Weak AI focuses on the (almost) optimal solution to a specific application problem. There are many examples of the use of this type of artificial intelligence. Predicting weather, categorizing image material, face recognition, language to text translation or even autonomous driving are all areas of application for weak artificial intelligence.

As a result, all currently existing applications in the field of artificial intelligence are from the weak AI category. The aim here is to create a statistical model that solves a task in the best possible way. All types of sensors (e.g. cameras, audio or vibration sensors) and systems (e.g. ERP, web analytics, service) serve as the data basis. Based on this, one lets the AI ​​“learn” in order to make predictions or categorizations - classic examples from the field of machine learning.

The history of artificial intelligence

Even if artificial intelligence has experienced a real hype in recent years, the principle - especially in a theoretical nature - has been around for more than half a century. In the first half of the 20th century, it was above all the area of ​​science fiction that presented the first concepts of autonomous machines and vehicles.

This was followed in 1956 by the definition of the term “Artificial Intelligence” at the so-called Dartmouth Conference and thus the field was also formally defined and opened for further research.

One of the first and most impressive examples was achieved in the field of natural language processing over a decade. The so-called ELIZA system by Joseph Weizenbaum used syntactic restructuring (from statement to question) of written input in order to work like an intelligent system.

A lot of research followed, as well as two so-called “AI winter”, in which it was recognized that the then exaggerated promises cannot be kept. In these times, research funds for research in the field of AI were frozen, but gradually thawed again and again under critical scrutiny.

One of the most famous events in modern times is likely to be Deep Blue's win against Kasparov in chess. In 1997 it was shown that a machine could be better than a human in a very specialized, albeit highly cognitive, area.

Since then, events have rolled over: From IBM Watson’s win in Jeopardy to the definition of Big Data and the development of deep learning, there have been many methodological and technological developments that keep AI strengthening.

Definition of terms - AI vs. ML, NN, DL and Big Data

Artificial Intelligence vs. Machine Learning (ML): What's the Difference?

Artificial intelligence is often equated with machine learning. But machine learning is only one part of artificial intelligence. The difference between AI and ML is that AI also includes simple systems that mimic human behavior.

A simple example of AI, but not ML, is a rule-based tic-tac-toe bot: This manages to follow an optimal routine in the game, which comes very close to human behavior. However, it is not based on variable learning of states such as machine learning, but rather firmly defined rules.

Therefore, AI covers a much broader field, while ML defines the special application of learned concepts using statistics.

Artificial intelligence vs. neural networks (NN) / deep learning (DL): what's the difference?

Neural networks and deep learning are often mentioned in the same breath as artificial intelligence - but what is the difference? Neural networks - i.e. the artificial implementation of a human neuron - are often used as a method in machine learning to solve classification tasks. Therefore, neural networks are a method of machine learning, which, as discussed, is in turn a sub-area of ​​artificial intelligence.

Deep learning, on the other hand, is a specific, advanced type of neural network. Since this approach also allows sub-concepts of an input to be generated, predictions are much more multi-layered than with classic ML algorithms. This contributes to improved results. Nevertheless, deep learning is also a method in the field of AI.

Artificial Intelligence vs. Big Data: What's the Difference?

Big data is data that goes beyond ordinary, locally and easily processed data sets due to its volume, its variability, its speed, its quality or its value. Consequently, artificial intelligence denotes a range of methods for using data, while big data denotes the associated data itself.

The role of AI in business and society

Why is AI so remarkable?

When broken down, artificial intelligence is nothing more than a lot of data, statistics and programming. So the question arises: why has it been revolutionizing the technology industry for some time?

The added value of artificial intelligence is that the algorithms are able to do things that cannot be done manually. Both in terms of precision and volume, AI enables data to be processed which humans or conventional algorithms cannot.

The easiest way to illustrate this principle is by using various examples. Face recognition is an algorithm based on neural networks and extracted from an image using image processing in order to subsequently identify or extract other information about the face (e.g. age, gender) depending on the application.

This procedure can be scaled as required using AI: No matter how many videos or images are processed, the artificial intelligence is able to handle it. This would not be possible with human support: neither the exact recognition of faces nor an almost arbitrary mass can be achieved manually.

But other use cases, such as autonomous driving, show that many technological innovations would not be possible without AI. Automatic communication with other road users, making an optimal decision in milliseconds and successfully navigating even in unfamiliar areas - all of these are just a few starting points of how AI is superior to conventional methods.

Artificial intelligence in business

One of the most common questions related to artificial intelligence is what role AI will play in the company. Is it a fringe technology, human supporter, or even the future that makes decisions autonomously?

While the range of opinions on this is as wide as the questions themselves, some things are relatively reliable. On the one hand, every managing director should make it clear that artificial intelligence and machine learning will no longer leave the stage of the global economy.

Similar to the Internet and digital technologies, we see AI as a central capability in every company in the future. Whether as a separate data science unit or as AI software will be determined individually, but that AI is used is inevitable. Especially with the establishment of the Internet of Things, it is useful and necessary to understand what artificial intelligence is capable of and which limits have to be observed.

Corresponding to other core competencies, it will also become more and more central to anchor and promote a “data driven mindset” in the company. By means of training courses and the proactive promotion of data-driven ideas, a company manages to be able to act in the long term.

Artificial intelligence in medium-sized companies

When we talk to companies about AI, many see large companies as the main players. But following our previous idea that AI will be ubiquitous, we need to extend this assumption. Not only DAX companies have to invest in AI competence, but also medium-sized companies.

While selling artificial intelligence as a product will seldom be central in medium-sized businesses, the potential must not remain unexploited when it comes to optimization and innovation. Only those who are themselves able to successively position the company digitally using simple methods will be able to keep up with large companies in the future.

Will artificial intelligence soon make jobs superfluous?

This question should be formulated more precisely: "Which jobs does AI make superfluous?" As is often the case with technological progress, AI will also bring such great improvements in certain professions that human influence will be superfluous.

Which professions are affected by the advancement of AI? Especially those who carry out a very repetitive task or do a special task that can be solved very well by the algorithms of machine learning or automated. Examples include market researchers (NLP and Big Data), parcel deliverers (with the replacement drone), truck drivers (autonomous driving) or recruiting agencies (data mining).

Does AI belong to the future?

In addition to the many positive aspects, there are also a few critical voices who argue that everything will soon be dominated by machines. In positive terms, these voices pose the question of whether artificial intelligence will be omnipotent in the future.

As already mentioned in the previous questions, we see this in an aspect-dependent manner. On the one hand, AI will have a very strong impact on certain parts of society and the world of work. Be it the automation of certain professions or various personalization algorithms that keep people more and more in their own comfort zone.

In contrast, there is the great argument that artificial intelligence has so far been very far removed from the human ability to think. The strong AI is still a product of science fiction and technically, there is still a long way to go.

Nonetheless, all companies and scientists who continue research in Artificial Intelligence should ask themselves what happens when we are ready. These considerations are manifested in the topic “Ethics in AI”, often also in reference to the 3 laws of robotics. They say that - no matter how advanced the AI ​​is - the AI ​​must never hurt a person, must obey people and protect itself.

Taken together, we see three aspects: AI will change large parts of our daily being, but we are far from a general AI that trumps people in all areas. And even if we should get there one day, there have been considerations for decades as to how AI can also be made ethical.

Receive further articles on the topic of Data Driven Company directly by e-mail:

Artificial intelligence methods

It is now known what AI is and why it is relevant. But how does it work in detail? Here we would like to introduce how artificial intelligence is implemented in everyday life. Machine learning methods are dominant here, but simple data processing approaches such as image processing also fall into the field of AI. Here we would like to give a brief overview of the main points of the methods used.

Supervised learning

In the area of ​​supervised learning, two main categories of algorithms are used: prediction and classification. The goal of prediction is to predict numerical values, i.e. numbers. Simple examples are the prediction of sales figures, temperature or the duration of a process.

The classification on the other hand fulfills the task of classifying something into one of two categories (“labels”). Examples are object recognition ("dog / cat"), error recognition in the manufacture of products or the classification of customers in certain segments.

What all algorithms in the field of supervised learning have in common is that they first train a statistical model on the basis of training data and then apply this model to new data. A model in this context only represents a statistical equation that has been optimized until it predicts as much training data as possible as precisely as possible.

Supervised learning is used where you want to make the best possible prediction based on existing data. The most complex step is often the labeling of the data. Depending on how much additional information (e.g. age, gender, origin, length of membership, etc. for a customer) is taken into account in the algorithm, the more data is also required to train a reliable model. And this data must all be provided with training information in order to provide the model with a parameter with which it can train the prediction.

Unsupervised learning

In contrast to supervised learning, there is unsupervised learning, in German unsupervised learning. In the area of ​​unsupervised learning, there are a number of methods, all of which aim to identify patterns in the available data.

However, the procedure is not based on training attributes as in supervised learning, but uses information within the data to form meaningful groups.

One example is the area of ​​clustering. In clustering, distances between individual data entries are calculated in order to combine entries that are as identical as possible in a group (a cluster). This happens, for example, in the area of ​​customer segmentation. Every customer has many attributes (age, gender, place of residence, number of purchases, preferred delivery method, favorite category, etc.) and can thus be set in relation to every other customer. In this way you can find customers who behave in a similar way, who can then be viewed collectively.

Another example in unsupervised learning are recommendation algorithms. Events that occur together (e.g. purchases of goods) are analyzed for their commonality in order to make recommendations based on them. The large amounts of data from all customers are used together, for example, to evaluate which products are often bought together - and then to suggest the other to those who have only bought one.

You can find these and other categories for machine learning in our article “What is machine learning? Machine learning definition, algorithms and examples ".

Reinforcement learning

Reinforcement learning is seen as the third main category of machine learning and thus also artificial intelligence.In reinforcement learning, a so-called agent (for example a virtual avatar or a real robot) is used in an environment (for example a room) in order to achieve a goal. For this purpose, the agent can use previously defined actions (e.g. driving straight ahead, steering, braking).

The agent's job is to find an optimal way to achieve the goal. For this purpose there are resources (for example energy) that are consumed according to the actions used.

As already indicated, reinforcement learning is used in the field of autonomous driving, but such algorithms can also be used in offer or price optimization.

Signal processing

Signal Processing deals with the translation of signals from the environment into interpretable data. In general, this AI method is about processing signals (for example from sensors such as a camera or microphone, but also vibration, energy or temperature).

The aim here is to extract as much information as possible from a one-dimensional signal in order to carry out appropriate subsequent actions. An example would be that a production machine is switched off before it overheats.

From a formal point of view, signal processing is no longer part of machine learning, but it can definitely be processed using machine learning methods. Subsequent processing steps in particular are usually carried out using ML.

Image processing

What can be seen as a sub-form of signal processing is image processing. For this purpose, images or videos recorded by cameras are processed into information to be extracted.

A very prominent example that has now found its way into our everyday lives is facial recognition. The face is extracted from the image via the front-facing camera of our mobile phones and then classified into “owner” and “non-owner”.

Image processing is also used in the area of ​​object recognition. Regardless of whether it is people, animals or products: it is now possible to extract very reliably from each image which objects are on it.

Natural Language Processing

Another sub-category of signal processing is Natural Language Processing (NLP), called natural language processing in German. Speech is one of the most efficient ways of conveying information after seeing. Therefore, language processing is also very interesting in the field of artificial intelligence.

NLP covers both speech and text. The aim is not only to extract the text itself, but above all to extract the context within the information used.

An example of the processing of speech in the field of artificial intelligence, in particular text-to-speech and speech-to-text, are virtual assistants such as Siri or Alexa. Both are based on the fact that natural language can be translated into digital information and back in order to interact with other information sources such as the Internet.

Another well-known use case for AI in NLP is sentiment analysis. For this purpose, words are assigned certain emotional connotations (for example, the word “fail” has a negative connotation) and then new texts are analyzed using this model. This makes it possible, for example, to classify emails in the service area according to category and / or importance.

Data types: structured and unstructured

One of the advantages of artificial intelligence is that not only structured, cleanly recorded data can be processed in databases, but also unstructured data such as images, text, video or audio can be used.

This variance in data types is also reflected in the processing methods shown: Classification and clustering as well as signal processing often work directly with unstructured data.

Big data and artificial intelligence

In combination with the high compatibility of AI with data types, other big data factors such as volume or speed also play a role in artificial intelligence.

Many of the methods used in AI are specially designed to deal with large amounts of data. In general, the following principle applies: the more data, the higher the variance of the processed data, the better the analysis.

Thus, all in all, artificial intelligence is not only able to deal with big data, but large amounts of data even promote the results of AI.

Examples of the use of artificial intelligence

Spam detection

One of the most common use cases, which hardly anyone sees superficially, is the detection of spam. Ten years ago the Internet was flooded with unwanted advertising of all kinds, but now you hardly ever come across such news. Regardless of whether it is on social media, YouTube or e-mail providers: Spam has been successively reduced in recent years using artificial intelligence and has now been almost completely removed. The methods used are primarily machine learning models, possibly supported by simple statistics.

Virtual assistants

Virtual assistants like Siri or Alexa are a very nice example of the use of artificial intelligence. They combine text-to-speech with speech-to-text so that the devices can understand people's verbal utterances and reproduce structured content. In addition, even deeper neural networks work to understand verbal expressions - such as the difference between “playing” in different contexts such as music or games.

face recognition

Presented as dystopia in older books, face recognition is now more and more about us. Not only used by the state and the police, but anchored in more and more products, face recognition is based on image processing and deep learning. The example of China shows that it can also take on negative proportions: Face recognition is used in full to implement a “social score”. This score is defined by people's behavior - i.e. where who is and why - and allows or prevents certain actions in society. And this example is really close to a dystopia.

Predicting sales

Predicting numbers is one of the core use cases for AI. Using time series analysis or other ML algorithms, the attempt is made to know in advance how the company is developing.

Particular attention is often paid to sales. Whether it's total sales for an organization or on a product category basis: If you can estimate how the business is developing, you can react accordingly in Operations or Marketing.

Prediction of product lifespan

A very common use case in Industry 4.0 is the prediction of products. Based on the production and usage data, machine learning models are trained that try to define when a product breaks.

Whether it's a car, a production machine or an electronic device - all products provide data and can have a different lifespan based on their use.

The advantage of knowing when a product should be repaired or serviced allows it to be used for a variety of purposes. Whether it is the so-called predictive maintenance, the maintenance terminated before a machine reaches its limits or the delivery of spare parts to consumers before something breaks - there are many aspects that can be used to reduce costs and increase the user experience.

Receive further articles on the topic of Data Driven Company directly by e-mail:

Summary

In summary, we would like to explain artificial intelligence again simply:

  • Artificial intelligence simulates human competence
  • AI can include simple algorithms such as rules or image processing, but is mostly associated with machine learning
  • AI is so attractive because of the ability to make decisions fully automatically, with high performance and on an almost infinite number of inputs
  • Another advantage is the ability to process not only structured, but also unstructured data such as images and sound
  • Specific examples are spam detection and classification, virtual assistants, video processing or forecasting of sales or other numerical metrics

As you can see and we have explained in detail, artificial intelligence will be integrated more and more into our everyday and economic life in the coming years. The risk of people being completely displaced as a result is relatively low, both technologically and cognitively. The opportunities, however - fully automated processes, superhuman performance on special problems and relief for people in many repetitive areas - are extremely attractive. All in all, we look forward to the successes that artificial intelligence, machine learning, data science and big data will bring in the coming years.

Categories Analytics & Machine Learning, Big Data