Human in the Loop: the human in the machine


Human in the Loop ML

Man in the machine – a buzzword familiar from science fiction novels of the early 20th century. What this term is about in the 21st century is clear: it is about Artificial Intelligence. The development and training of AI requires the intervention of natural intelligence at many points: human in the loop. In this loop, the human acts in a similar way to a teacher.

AI needs instructions

The grand vision behind Artificial Intelligence is the independent learning of programs. Software evolves on its own, learns new things and applies these insights in practice. But before it gets that far, every system needs training. It’s like in school: if the student makes mistakes, the teacher intervenes. And the student learns from it. He will avoid making similar mistakes in the future. The same applies to Artificial Intelligence.

What is the advantage of Human in the Loop – abbreviated HITL? The principle is quite simple:

  • Human already has a large reservoir of experience. He recognizes mistakes and points them out. Man alone has the ability to decide correctly even intuitively.
  • The machine scores with its incredible speed and fast access to large amounts of data.
  • The combination of these two features optimizes the system.

So Human in the Loop uses the combination of human and machine intelligence to create machine learning models. Human and machine hand in hand: Humans are unbeatable at making sensible decisions on a small database. Machines, on the other hand, access a gigantic amount of data. Their advantage lies in precision.

Human in the Loop in practice

What does the interaction between humans and machines look like in practice? Human in the Loop works in different ways. For example, the human factor already plays a role in the creation of training data. In supervised learning, this has been common practice for a long time:

  • Humans mark input data to stimulate the machine to learn.
  • Humans evaluate the output data to constantly adjust and improve the algorithm.

Human in the Loop is a continuous feedback loop that is also used after development. The concept is based on a cycle. The program never stops learning. It constantly evolves based on the reactions of the users.

Areas of application for Human in the Loop

Human in the Loop can be used effectively in two areas of Artificial Intelligence work:

  • For training: Before market maturity or actual use, the human in the machine optimizes the learning of Artificial Intelligence.
  • For tuning and fine-tuning: already existing AI systems get more accuracy from the human factor.

The Human in the Loop concept can be used for a variety of AI projects. These include NLP (Natural Language Processing), computer vision, sentiment analysis, and a variety of other use cases. Any deep learning AI can benefit from some human intelligence inserted into the loop at some point.


Human input for AI wanted? Via clickworker you can get AI training data created by humans, data labeling as well as data validation.

AI Training Data Sets

An example is the automatic marking of certain objects on images. The program should mark the detected object (for example a traffic light) with a box. During training, the teacher can focus directly on difficult cases and help the program with his reactions: He marks the object himself in case of faulty output.

The larger the amount of data, the faster a concentration on the doubtful cases has an effect. This enables the system to solve similarly complicated cases increasingly on its own. Gradually, the loop of output and reaction leads to better and better results.

Man or machine? The level of trust decides

Human in the Loop is a circuit (loop). The system is constantly improving by allowing checks at critical points and gaining new knowledge from the reactions to decisions.

But at what points should humans intervene? The answer to this question depends on the level of trust one chooses. How high should the bar be at which erroneous decisions are no longer accepted?

  • Below a certain level of trust, the human is allowed to provide feedback to encourage the algorithm to continue learning.
  • What falls within the realm of trust is still left to the system.
  • As development progresses, the program improves more and more. This also raises the bar. Ideally, human correction is no longer necessary at the end.

The idea behind HITL is that humans and machines work together. Constant feedback promotes the continuous optimization of algorithms.


Human in the Loop is a concept that contributes decisively to the improvement of Artificial Intelligence. Because the main problem in the development of AI systems is often the lack of training data. This is where HITL comes in: The human in the flow of the program feeds the algorithm with correct results based on natural intelligence. Within certain limits, this principle can compensate for the lack of data. Gradually, the system adapts the structure of these decisions and constantly becomes more adaptive. This contributes significantly to the quality of Artificial Intelligence.


Dieser Artikel wurde am 17.December 2021 von Jan Knupper geschrieben.


Jan Knupper

Leave a Reply