How useful are AI systems really?

Whether AI systems do good or bad is determined by humans, provided they take a few simple rules into account. Opinions differ even on the question of what artificial intelligence actually is.

For it has not even been conclusively clarified yet what natural intelligence is. Let us agree that it is literally about understanding, and understanding is preceded by a learning process.

AI is supposed to automate digital processes "intelligently". IT systems have of course been used for a very long time to automate processes. There is the good old cron job, for example.

An action is performed at certain intervals, for example, the recycle bin is emptied every five minutes. Computers do this job extremely reliably, they do not get tired while working and do not find endless repetitions boring either.

However, they would never come up with the idea themselves to put the garbage by the road to make room for new data. Man had the idea. He recognized the need and wrote an algorithm.

In fact, people are still behind most automatic IT actions today, not intelligent systems. There are good reasons for this. Because what is often forgotten about AI is that humans are in charge. He should consider what a system should learn and, above all, from whom, i.e., what the goal of the whole thing is.

Who influences whom?

In 2016, Microsoft launched an experiment with a Twitter chat bot equipped with artificial intelligence. However, after some time, the fictitious teenager named Tay began to tweet in an increasingly racist and misogynistic manner. He had apparently been specifically influenced by a group of Twitter users.

Microsoft decided after 24 hours to make the profile inaccessible to the general public and to stop the experiment. Microsoft had certainly not foreseen that the experiment would be deliberately influenced, otherwise the company would have defined rules in advance.

In this case, it was about social norms, a rather squishy matter compared to the management of an IT landscape. But even the functionality of an IT landscape is not as easy to maintain as it seems, because it requires setting priorities:

• Which system will be restarted first in the event of a failure?
• And which update will be implemented first?

Such decisions are usually situational and the more components that need to be weighted, the more complicated it gets.

Impact Assessment

Everyday life shows in many examples, from employee bonuses to government support programs, that too simplistic target systems can lead to undesirable developments.

At the beginning, they fail to assess the consequences in a complex system, and later they lack readjustment. This is precisely where AI can help. Neuronal networks can play through the interactions of individual components at lightning speed and in this way help to create and maintain an optimal system state.

What exactly is "optimal" is a matter for man to decide. Perhaps you still remember HAL 9000 in the feature film "2001: A Space Odyssey". It is possible that this increasingly neurotic, astronaut-murdering mainframe computer has led to AI still being eyed very suspiciously by many.

Instead of supporting the human, he wanted to protect himself from his shutdown and complete the mission. For this, he could even lie purposefully. Clearly a consequence of poorly prioritized goals, lack of impact assessment and learning from the wrong example - analogous to Tay.

My conclusion is accordingly: AI can help us make much better decisions in very many areas, especially when things get complicated. After all, these are the situations in which people often do exactly the wrong thing.

However, a prerequisite for this is that we first think about what we want to achieve with the help of AI and from whom the system should learn.