Everyone is talking about artificial intelligence, the advantages and the dangers. Adventurous definitions are sometimes expressed in this regard. We try to make a definition and explain using a concrete example what you need for an AI application and how everything works - and why the author and philosopher Richard David Precht is not entirely right with his definition of AI.
What is AI – Interview with Richard David Precht
INOSOFT AI Expert Patrick describes Artificial Intelligence as an algorithm or method, that makes independent decisions. He criticizes the fact that everything is lumped together in the current discussion.
Richard David Precht, for example, defines AI in an interview as statistical systems that evaluate and process a great deal of data, recognize patterns and, once they have recognized the patterns, are able to vary them independently.
As an example, he refers to a chess computer that is better than all the world chess champions, and a Go computer that has beaten the best Go player in the world. From this, he deduces that algorithms, once they have understood how something works, can develop something new and are thus superior to humans.
Why Precht is not right
We can agree with Precht's statement about data analysis and pattern recognition. But the comparison of AI with humans is misleading. What people always disregard is the fact that a Go or chess algorithm can certainly play chess or Go but we humans can do significantly more. For example, when we drive to work in the morning, we have to operate the car and, at the same time, make sure that we don't run over any pedestrians. This algorithm can't do all that. It can only play chess or Go. And these games are also an ideal field of activity for algorithms, because the set of rules is strictly defined. The algorithms have been taught moves by experts and have developed them further as a result. However, algorithms cannot invent new things.
We can use the methods of AI to extract knowledge from data and thereby perhaps optimize things. But we always set limits for the algorithm within which it operates and we provide the training data on which it builds-up. This has nothing to do with new creation and creativity.
What do you need for an AI application?
In advance, you have to think about a precise task. Then you need data, i.e. thousands, tens of thousands or hundreds of thousands of data sets, depending on the task, to show the model what to do with them. The data must be prepared before training. You need a framework for processing and appropriate computing power. The algorithm then analyzes the data with the aim of finding a pattern and detecting deviations. Examples would be the detection of erroneous transactions in credit card transmissions or the analysis of medical images to find out cancerous tumors. However, because of AI’s lack of traceability, a human must always check the results.
Face recognition to try out
Image recognition, in this case face recognition, is also the basis of our small application. This is also referred to as machine learning. The application was used to automatically recognize employees as such when they log on to the company network.
If you want to try it out: Simply upload a search image and a reference image and experience how the system detects the person you are looking for in the photo: KI.INOSOFT.de
We define Artificial Intelligence as a toolset for dealing with data. However, AI is only suitable for certain areas. For a task, you should therefore try out whether you choose AI as a method or a conventional analysis technique.
Where it can be used, artificial intelligence is very helpful and therefore always worth a try.