Was ist AI?

What is AI?

What does artificial intelligence mean?

Artificial Intelligence - An introduction, explained in an easy-to-understand manner.

We come across the term artificial intelligence (AI) more and more often. Unfortunately, even in this area of ​​technology, many people cannot imagine what exactly is meant by it. Or you ask yourself the question: What can AI actually do for me?

With AI, one generally tries to transfer human thinking and learning to a machine or computer.

If you want to get closer to this topic and enter the term in English spelling into a search engine, over five billion search results will come up.

Anyone who has dealt with one or two articles will find that they are almost overwhelmed with technical terms and word battles.

“Artificial intelligence, machine learning, deep learning, neural networks, weak AI, strong AI, BIG-DATA, data driven business, master data management, marketing automation, user experience design, TensorFlow, etc..”

To name just a few examples.

In many cases, free “white papers” with application examples, the so-called “use cases”, are offered for download, in which practical examples are intended to help you understand the ingenuity behind AI and how companies now have “no problems” thanks to its introduction " have more.

Many offers and articles suggest that AI is “the solution to your problems” and that “you absolutely need AI now” to solve them. But which problems are actually meant and do you really have problems for which AI can be solved? can be a solution?

In order to be able to form an opinion on this, we will try to explain the term and how it works to you very simply and without mathematics, so that you can begin to understand the basic functionality of an AI.

The aforementioned terms and many others from AI are in fact “somehow” related.

In connection with all the different terms, it should be noted that artificial intelligence (AI) is an umbrella term and terms such as “deep learning” or “neural networks”, as well as many others, fit into the overall picture of AI.

It is impossible to explain these terms and their complexity in detail at this point. However, the terms “weak AI” and “strong AI” should be carefully discussed below.


As of today, strong AI is wishful thinking and is far from reality. This refers to “robots” or “artificial life forms” that are modeled more or less 1:1 on people in their thinking and behavior. However, such strong AIs currently only exist within the science fiction scene. One of the most famous strong AIs is probably the popular android Lieutenant Commander Data, played by Brent Spiner, from the Star Trek series “The Next Generation”.

If you don't know him, you might be more familiar with the "Terminator", played by Arnold Schwarzenegger, although he is more of a combination of man and machine.


Weak AI is what we encounter in everyday life today, even if we may not be aware of it in many areas. Examples of this include activation via facial recognition on mobile phones. Or if targeted advertising is suggested on a website, you receive product suggestions from Amazon or other online shops, or even film suggestions within platforms such as NETFLIX or Apple TV. Lastly, examples such as Amazon Alexa or Apple's Siri, i.e. voice assistants, or speech-to-text converters should be mentioned. The latter enable us, for example, B. to speak a message with our voice in WhatsApp so that the spoken words are then available as text.

In order to develop strong AI, computers would have to have their own consciousness and intelligence. Or emotions, like us humans. But that's exactly what they don't have, and we're still miles away from that despite ongoing technological progress.

Weak AI, in turn, as we use and know it today, is based on the computer systems we know today, combined with complex mathematical algorithms. Also referred to in technical language as “applied math”.

If you like, AI is nothing more than a complex algorithm or combinations of many algorithms.

In simple words, what is an algorithm?

An algorithm consists of mathematical formulas, clearly defined instructions, many individual steps and it is finite. An algorithm must be finite in order to reach a result. If it were infinite, you would probably wait forever for a result.

An algorithm always requires an input before it can deliver a result, i.e. an output. So, for example, if you want a voice assistant like SIRI or ALEXA to do something for you, in this case you first have to input something into the algorithm in the form of voice commands before the algorithm responds with an output. Example input: “Hey Siri, what time is it?”. Example output: “It’s 9:09 a.m.”

This may sound very simple and logical at first, but it is not. For example, how does SIRI know how the words and sentences you speak are “decomposed” and “machine translated”, where this query is then forwarded to be answered, and how does SIRI know at the end what will be “answered” to your question as output should? So what exactly happens once you have formulated your entry (in our example, the question about the current time)?

The answer to this seems simple at first: your input “disappears” as if by magic in a complex algorithm that does exactly the things that you cannot see. Complex mathematical operations, formulas, calculations, comparisons, mappings, routing to other systems, retrieving information from billions of records, sorting, all back again so SIRI can respond to you with one voice, etc.

Would we now look in detail at how this all happens within one or more algorithms, e.g. E.g. how your language is broken down into individual phonemes etc., this will probably take a while and is not the point at this point. However, this example is intended to make it clear that a seemingly simple request can lead to very complex processes.

It is important to know that an input into an algorithm initially represents a “problem” for it, and an algorithm consists of a finite number of individual mathematical steps with clearly defined instructions for action in order to solve the problem through the subsequent output. That's what it was developed for.

The task of an algorithm is to react to the output and to use functions to calculate a solution that was formulated by an input.

In mathematics, a function is computable if it can be formulated for calculation using an algorithm. Simple example:

1+1 is predictable.

1 + TOAST BREAD is initially not calculable as long as TOAST BREAD has not been defined more precisely.

Algorithms are developed in a very complex manner and, as already indicated above, are used in many areas today.

We'll take a look below at how this can happen within an AI:

Artificial intelligence in the form of weak AI has the goal of “learning independently” and thus solving certain processes, tasks or even predictions and getting better and better over time. And ideally so that the results are better than those that a human “calculates”.

The question is: will this succeed?
The answer to that is: within tasks where AI makes sense, definitely.

Why? Here's an example: One of the oldest and best-known AI algorithms is probably the chess computer. The algorithm was given basic features or the rules of the game, as well as the moves of already known and important games, and the more it played against other opponents, the better it became. In the end he defeated the chess grandmasters of our world.

A weak AI has no consciousness, no emotion or real intelligence, but it has the advantage that the algorithm or algorithms “only” have to deal with what they were ultimately developed for. The chess computer has some advantages in these areas.


- He is never tired or dependent on any other form of day or time.

- He doesn't think about anything else besides chess, which we humans inevitably do subconsciously, for example thinking about the family, problems or other things.

- He can be his own opponent. So he also plays against himself. Computer against computer. In doing so, he tries out new moves that he is not yet familiar with within the stored rules of the game, if necessary a million or billion times. He “learns” when I win and when I lose and “remembers” everything down to the smallest detail. He can also play against you and other real people at the same time, while continuing to train in the background and also learn from your new moves that he doesn't yet know.

- He can carry out these arithmetic operations in fractions of a time that are unthinkable for us humans.

- He “remembers” the games and moves in a database, which he can retrieve more quickly and precisely like us humans. It doesn't matter whether a game is one day old or 20 years old.

Algorithms are also used for this. However, they are really very complex, so-called artificial neural networks.

If we want to understand this in a nutshell, we have to take a short excursion into ourselves as “humans”:

When we humans learn and remember something new, neurons are recruited within our human biological nervous system during the learning process. This creates neuronal connections (synapses). The more we learn or the more we train, these synapses become stronger. They are like “circuits” that form. We can recall these learned “circuits”. If we unlearn these again over time, these “circuits” can dissolve again.

In 1943, Warren McCulloch, an American neurophysiologist and cyberneticist, together with Walter Pitts, an American logician, managed to create an artificial neuron model and thus electronically and mathematically recreate parts of neurons in the nervous system of living beings. They are therefore considered the founding fathers of today's neuroinformatics. Even today, many artificial neural networks are still based in many parts on the networking of the so-called Mc-Culloch-Pitts neurons.

To understand this in more detail, you have to know that computers today are so-called control systems. IF, THEN… Also known to many as 0 or 1, or ON or OFF. Example: If a user clicks “Print”, then “Print”. If “dark”, then “light on”, etc.

Today we have the aforementioned founding fathers to thank for the fact that, through complex mathematical algorithms within artificial neural networks (and thus AI as a whole), we can at least better "break down" the "IF, THEN..." structure and use an algorithm " “Teach” to “further develop” themselves without a programmer having to specify every single step.

This means that artificial neural networks are able to “learn”, store and also link their own “IF, THEN…” structures within the “IF, THEN…” structures. By linking “the individual learned steps” with each other, patterns emerge that allow AI to become increasingly better at avoiding errors. In addition, an AI usually does not unlearn anything like us humans, unless the AI ​​is explicitly instructed to do so.

To clarify this a little better, let's go back to our example with SIRI and the question about the time. In a pure “IF, THEN…” system without the use of AI, a programmer would have to take every single case of “input” into account and program every single step manually.

The question or input you ask SIRI could be different:

- Hey Siri, what time is it? or
- Hey Siri, what time? or
- Hey Siri, what time is it? or
- Hey Siri, what's the time doing?

So there are different ways to just ask for the time. Let's assume there are 1,000 to 2,000 different ways to ask for the time. Each of us can certainly think of 5-10 examples. Outside of AI based systems, these 5-10 possibilities would be manually programmed by a programmer according to classic “IF, THEN…” rules. Examples:

IF “Hey Siri, what time is it”, THEN “Say time”

IF “Hey Siri, what’s the time?” THEN “Say the time”


Such manual programming can be made a little more intelligent by having a programmer link different queries together in order to be able to evaluate certain “assumptions”. In our first example of the query there is e.g. E.g. the word "LATE", in the second request the word "TIME". Such links then increase the probability that the time will be output even if a request does not correspond 100% to a sentence as it was stored in the programming.

Such links recognize certain patterns within queries, which can also lead to correct output.

Very simply presented:

IF “Hey Siri” AND “late” OR “time”, THEN “say time”

In practice this would still not work, but at this point it should show you roughly how it works.

Based on these examples, it should be clear that it is impossible to do manual programming for every request in advance. Not even if, for example, you had started a worldwide telephone survey before programming or developing software with the aim of noting how others would ask about the time in order to then take these questions into account in programming. You or your programmers would probably suffer a collapse at some point in trying to prepare all of these different options for programming in any meaningful way.

If, in addition to the actual, clearly formulated questions, manual “patterns” are to be programmed, which “assumptions” are to be taken into account from the combination of different questions and terms, one thing becomes relatively quickly foreseeable: you lose track. You could also say: It ends in chaos.

Another example are electronic vacuum robots, which can now be found in many households. How should a programmer program a robot vacuum from the start so that it doesn't constantly bump into objects in your house, or if something is changed, take this into account from the start? Apart from millions of different households, where each robot vacuum cleaner needs its own program due to the different circumstances and facilities. This is also an impossibility, programming something like this manually and taking every eventuality into account in advance.

The solution is not to program everything manually, but rather to accept all inputs or requests that come from all over the world in a variety of forms and then dump them into an artificial neural network consisting of complex mathematical algorithms.

However, each neural network has its own individual tasks. One calculates e.g. B. Requests from SIRI, another trains a vacuum robot. Different neural networks can then be linked together.

Within these neural networks, new "IF, THEN..." instructions are automatically created, as well as complex connections between each other, from which certain patterns can be recognized. Without a programmer having to program it manually as described above.

Because the AI ​​is constantly training itself and increasingly through the input it has learned, it can increasingly recognize things on its own and come to solutions, even if, for example, B. has never heard a word that was requested to SIRI before. And in doing so, she continues to expand her “knowledge” in the background.

In the beginning, you may even feel a little sorry for the AI ​​if it doesn't have enough information to "learn independently" from the input and recognize patterns.

This clearly means that the more data an AI receives through input, the better it gets at completing its task. The less data it has, the more erroneous the output is.

Every AI therefore has a so-called learning or training mode. If it receives input that cannot initially be assigned to a pattern due to too little data, the AI ​​forwards such requests to human teams.

The human teams look at the requests to the AI ​​that could not initially be assigned and assign the AI ​​appropriate output options.

Over time, the AI ​​becomes better and better due to the mass accumulation of data (BIG-DATA) and the requests to the human teams become fewer. She also begins to train herself using the data collected and to continue to expand and improve pattern recognition.

According to current data, SIRI alone processes around 3.2 billion voice requests per week. Mass data like this is the ideal “fodder” for AI-based systems.

Companies like Google, Amazon, Facebook, Instagram, to name just a few, also store massive amounts of data. In particular about user behavior.

Music or movie suggestions like those from NETFLIX don't just come from what you've listened to or watched in the past. For this purpose, for example, data from all over the world is evaluated in order to make appropriate suggestions to you.

If you are thinking about whether AI can be useful for you or your company, there are a few suggestions that we would recommend you think about:

1. What question would you want to ask an AI and what would you like the corresponding output to be? And at the same time in this context: Would AI possibly be better at providing answers than existing human project teams?

2. I would like to solve a problem that e.g. B. has to do with prediction models that are difficult to answer by humans? For example, within production processes, logistics, transport and much more.

3. Are there areas in which errors repeatedly occur that an AI could possibly optimize?

4. I am thinking about replacing certain areas with AI in order to save costs in the future. Above all, ask yourself whether you already have large amounts of data from the past few years that could be used to feed an AI. There is also the option of purchasing external data that can potentially help solve problems within an AI.

Repeatedly, AI can be used very sensibly in the area of ​​certain prediction models if, for example, inquiries depend on many factors that overlap, such as weather, procurement of goods, delivery, availability, personnel, advertising.

AI can also make sense in places where you yourself have noticed that, for whatever reason, the “limit” of what is possible seems to have been reached and you no longer have any idea what to do next.

You should also keep in mind that AI can solve problems in many areas, but in many areas it cannot. Furthermore, AI is not a problem solver for everything and AI was developed by us humans. This means that errors can occur within AI systems and the algorithms used. When AI is used in a company, it is usually for a specific area for which it was developed and trained. AI cannot be easily applied to other areas or simply transferred.

We would be happy to advise you in the area of ​​artificial intelligence and help you analyze whether the use of AI can make sense or not. The implementation and introduction of AI in companies is also part of our range of services.


AI in our everyday lives

As strange as it may sound, the EURO-Fighter fighter jet is built at the highest limit of instability. Normally you would build an airplane so that the wings are long enough to give the airplane the highest possible and safest lift and, for the sake of safety, you would rather have to fly a large loop to turn the airplane in the air and perhaps design the aircraft a little lower at maximum speed so as not to push the materials to their maximum load limit. The EURO Fighter has short wings to make it more maneuverable, is designed for maximum acceleration and much more. Without the use of AI, the EURO Fighter would prove to be unfit to fly in some areas. In order to keep it in the air, the AI ​​has to calculate appropriate prediction models of the pilot's future reaction, as well as taking into account many other things besides altitude, air pressure, weather in general, surroundings such as desert, temperatures, speed, weight, G-forces, etc . And this happens in fractions of a second, constantly and constantly, while the plane is in the air.

Back to blog

Leave a comment

Please note, comments need to be approved before they are published.

Courses & Training