Artificial Intelligence

In computer science, an algorithm is an exact computational specification for solving a task. A learning algorithm (or self-learning algorithm) is an algorithm that receives sample data (learning data or training data) and computes a model for the seen data that can be applied to new data samples.

Data Science is an interdisciplinary scientific field that deals with methods, processes, and algorithms for extracting insights from structured and unstructured data. The professional field of data scientists requires knowledge of mathematics, business administration, computer science, and statistics. Data scientists identify and analyze available data resources, determine needs, and develop concepts to use the data profitably.

Fraunhofer IAIS offers data scientist trainings.

Foundation models or large language models (LLM) are large language models with billions of parameters that are trained on huge data sets. They rely on "self-supervised learning", i.e. learning that does not require annotations. Examples of such models are ChatGPT, GPT-4 or BARD, and they produce high-profile results and broad social discussions.

On the one hand, they can be fine-tuned for specific tasks with labelled data. On the other hand, large models can also perform tasks directly by formulating a mission and explaining it with a few examples if necessary. Clear advantages lie in the high flexibility for the application of such models. Disadvantages include, in addition to the enormous computing power that is currently still required for training, the associative mode of operation of the models, which seek plausible formulations and are not oriented towards lexicon-like fact fidelity. On closer examination, inconsistencies and jumbled information sometimes become apparent. Furthermore, it is usually completely unclear how the models arrived at their results. However, transparency and traceability are of elementary importance in cases of critical decisions, such as in the health sector. Despite the already enormous potential of LLMs, there are still important research questions about their trainability, reliability and fairness.

Artificial Intelligence is a discipline of computer science, that deals with how a computer can mimic intelligent human behavior. Neither is defined what "intelligent" means, nor which technology is used. One of the foundations of modern Artificial Intelligence is Machine Learning. Subject matter experts distinguish Weak AI and Strong AI.

Weak AI implements AI methods to solve narrowly defined tasks. While it can already surpass human capabilities in individual areas, such as image analysis, Weak AI falls far short of reaching the same level for broader context tasks or tasks that require world knowledge. All current AI solutions are examples of Weak AI and have a limited horizon: An Artificial Intelligence that can recognize images cannot necessarily play chess.

Strong AI stands for the vision of using AI techniques to emulate human intelligence to its full extent and outside of individual, narrowly defined fields of action. Strong AI has so far only been found in science fiction. Since Artificial Intelligence emerged in the 1950s, there have been predictions that Strong AI would become realizable in a few decades.

Hybrid AI combines data-based Machine Learning, world or expert knowledge, and logical reasoning. Knowledge and the respective conclusions are directly introduced into the learning process, for example, in order to emulate the human ability to correctly understand meanings from the context and to make the AI system more robust overall.

Only trustworthy AI applications guarantee IT security, control, legal certainty, accountability, and transparency. For this reason, guidelines for the ethical design of Artificial Intelligence are being developed within companies and at the societal and political level. These focus, for example, on the dimensions of ethics and law, fairness, autonomy and control, transparency, reliability, security and privacy.

Machine Learning aims to generate "knowledge" from "experience" by having learning algorithms develop a complex model automatically from examples. The model, and thus the acquired knowledge representation, can then be applied to new, potentially unknown data of the same type. Whenever processes are too complicated to describe analytically, but enough example data – such as sensor data, images, or text - is available, Machine Learning comes in handy. The learned models can be used to make predictions or generate recommendations and decisions often surpassing the human level of performance – without any predefined rules or calculation rules.

Machine Learning is one of the core competencies at Fraunhofer IAIS.

Informed Machine Learning describes the approach of making the calculations of an ML model transparent by using world knowledge or well-understood components so that the decisions made by the model can be better understood by humans. This comprehensibility is important, for example, when Artificial Intelligence systems are used in medicine.

Informed Machine Learning is intended to avoid a so-called black box (into which humans cannot figuratively see) and one of the core competencies at Fraunhofer IAIS.

Artificial neural networks are Machine Learning models that are inspired by the brain's natural neural networks. They consist of many layers of interconnected computational units implemented in software, called artificial neurons. Using examples, a learning algorithm modifies the interconnectedness between neurons until the neural network produces good results. The number of neurons, layers, and their interconnections have a significant effect on the model's ability to solve problems.

Deep Learning describes the implementation of a Machine Learning process in the form of an artificial neural network with many layers composed of a large number of artificial neurons. The generation of features relevant for learning is done autonomously. Deep Learning is responsible for the successes in speech, text, image, and video processing.

Crucially, Deep Learning works particularly well when very large amounts of data – Big Data – are available to train neural networks. Thanks to its many years of experience with neurocomputing and Big Data Analytics and Intelligence Solutions, Fraunhofer IAIS is one of the pioneers in Germany in the development of Deep Learning approaches for industry.

Quantum computers base their elementary computational steps on quantum mechanical states – called qubits – instead of the binary states (bits) in classical digital computers. Qubits are processed using quantum mechanical principles, which is expected to provide a tremendous speed advantage for some applications. The new computer architectures also offer potential for Machine Learning and thus Artificial Intelligence.

Learn more about Quantum Computing at Fraunhofer IAIS.