Can an AI be trained without data?

It is the ability to learn and improve over time that is at the heart of AI, which requires large amounts of data to train the algorithms. The question often arises whether you can train an AI without data.

Before we get into the topic, we should first understand what is meant by “training an AI without data”.
In the traditional approach to training an AI, developers feed the algorithm large amounts of data and the algorithm uses this data to learn patterns and make predictions. However, when training an AI without data, the goal is to create an AI that can learn and improve without pre-existing data.

Unsupervised learning

A first possibility for AI training without data is so-called unsupervised learning.
Unsupervised learning is a type of machine learning where the AI is not given labelled data, meaning there are no predefined answers for the algorithm to learn from. Instead, the AI relies on finding patterns and relationships in the data on its own.
This approach is often used in tasks such as clustering, which involves grouping similar data points together.

Reinforcement learning

Another approach to training an AI without data is reinforcement learning.
In reinforcement learning, the AI is not given pre-existing data, but a goal to achieve. The AI interacts with an environment and receives rewards or punishments based on its actions. Through trial and error, the AI learns which actions lead to rewards and which lead to punishments, and adjusts its behaviour accordingly.
Reinforcement learning has been used in a variety of applications, including AI games and robotics.


Even though both forms are technically possible, in reality, however, some kind of data entry is still required.
Unsupervised learning requires raw data to be fed into the algorithm, and reinforcement learning requires an environment with which the AI can interact.
In short, this means that you always need certain training data and information to train an AI. However, it is possible to develop an AI that requires minimal data input or can generate its own data. The goal should be to minimise the amount of data input required and to ensure that the data fed into the AI is of high quality and unbiased.
AI Training at

Also at, the AI is trained on the basis of a wide variety of documents. Some of our models have already been so well enriched with information that no new training is needed for the application. In some cases, however, new data sets can help to make further position data from the document types useful.
Since our AI was developed in-house and our servers are located in Germany, we can guarantee a very high level of data protection (GDPR, Schrems II).

Read more