Cognitive Systems

Cognitive systems, which are already an indispensable element in many areas today, will have a major influence on growing numbers of industries and economic sectors in the future. According to estimates from International Data Corporation (IDC), $77.6 billion will be invested in cognitive systems by the year 2022, making them the basis for future technologies such as

• autonomous driving and other autonomous systems,
• Industry 4.0,
• the Internet of Things (IoT).

At the Fraunhofer Institute for Cognitive Systems, we view cognitive systems as technical systems capable of independently solving and developing strategies for human tasks. To accomplish this, these systems are equipped with cognitive capabilities for context comprehension, interaction, adaptation and learning. Cognitive systems can utilize artificial intelligence (AI) methods such as machine learning, neural networks and deep learning.

Cognitive systems have a wide range of applications, such as voice assistants or in the analysis of large quantities of data. Our focus however is on cognitive systems that also take over safety-critical functions. In these environments, cognitive systems offer the greatest value-add when conventional approaches and methods no longer intercede. In order for human-robot collaboration to be successful for instance, an industrial robot should move out of the way and ideally continue to operate if a human comes too close, instead of simply stopping in its tracks.

Three criteria must be fulfilled before cognitive systems can be used in safety-critical applications such as autonomous driving or Industry 4.0:

• They have to be safe,
• They have to be dependable and continuously available,
• And at the same time, they have to be cost-effective to develop and utilize.

The example of autonomous driving can be used to illustrate the three criteria:

A self-driving automobile has to operate safely in traffic without causing accidents. At the same time, it has to be dependable from an availability standpoint – meaning it has to remain usable. Although an autonomous vehicle that fails to operate in all uncertain situations, and instead sits on the side of the road, is safe, it’s not dependable.

This raises the additional question of the costs. Autonomous vehicles have to undergo millions of kilometers of testing while being monitored. This approach is especially popular in the US, where it is used to acquire as much as experience as possible with operation of the vehicle. The number of test kilometers alone makes these tests extremely complicated however and requires considerable time and money. For example, every possible scenario has to be tested, such as different incidences of light, weather conditions and situations in the vehicle’s surroundings. Furthermore, every single software change, such as an update or a new version, has to be verified again with extensive road tests. This method alone will not lead to the success of autonomous driving.  

With this in mind, AI-based methods and cognitive systems have to be validated with a corresponding system and software architecture, which forms a protective framework in which wrong decisions by the AI technology will not cause any damage.

In cognitive systems, the most important characteristic is the ability to reliably and optimally respond to unexpected events and unknown changes, particularly in safety-critical applications. For instance: when a self-driving automobile detects an object on the road that it is unable to identify, it has to react, brake if needed and then avoid the obstacle, all in a dependable manner. If an automobile is capable of reacting dependably, despite such unexpected and unknown challenges, the Fraunhofer Institute for Cognitive Systems IKS describes this as »resilience« in keeping with the work of computer scientist Jean-Claude Laprie.

The research activities of the Fraunhofer Institute for Cognitive Systems IKS focus on ensuring that cognitive systems can also be utilized to assume safety-critical functions, a capability that has often been lacking to date given that verifying the reliability of cognitive systems represents a major challenge. With this in mind, three aspects form the foundation of the institute’s work in making cognitive systems suitable for safety-critical applications:

• Cognitive connectivity
• Cognitive architecture
• Cognitive behavior

The real strength of cognitive systems lies in the fact that subsystems can form a large cognitive system, a so-called cyber-physical system of systems (CPSoS). Individual subsystems cooperate on a temporary or permanent basis in order to jointly carry out a task. For instance, cooperative transport systems can carry out a joint maneuver in order to move a heavy load. To do that, the autonomous transport systems share their own perception, forward information and coordinate with one another. Once the task has been completed, the collective disbands again. To ensure the resilience of such systems, the Fraunhofer Institute for Cognitive Systems IKS conducts research into areas such the integration of dependable fallback scenarios for the collective behavior. An individual system must be capable of taking over safety-critical functions on its own in case coordination or communication with the overall system is not functioning. The collective should furthermore optimally and efficiently complete its tasks without putting safety at risk. Cognitive systems with flexible and adaptive architectures are therefore a prerequisite.

Because cognitive systems often have to work together within an overall system, flexible and adaptive software architectures are another aspect of the Fraunhofer IKS research activities. In a large overall system, new services and technologies are constantly being added or eliminated, which cognitive systems have to be able to deal with. Functions and systems can suddenly be unavailable in autonomous vehicles as well, such as when a distance measurement sensor is concealed. In this case the software architecture must intervene and suggest alternative dependable behaviors until the distance sensor is once again functional. The goal of the Fraunhofer Institute for Cognitive Systems IKS is to develop service-oriented architectures that are not only flexible enough to adapt to the current requirements, but which are fail-operational – in other words, architectures that continue to function even during a malfunction.

These environments are more than about having a system architecture that functions dependably. The heart of the cognitive system – the artificial intelligence – must be safe. With this in mind, Fraunhofer IKS conducts research into ways to develop artificial intelligence such that it is absolutely safe and able to identify its own weak points and vulnerabilities. One of the goals is to develop sufficiently-robust artificial intelligence technology. In addition, it’s essential to identify uncertainties within the AI in each situation in order to be able to react adequately. The research also involves examining specific test and verification measures for validating the AI technology. This raises the question: when is my system safe enough? In order to answer this question, Fraunhofer IKS is developing a methodology for systemization of the vulnerabilities in a cognitive system. With this systematic, developers understand which situations have already been taken into consideration and sufficiently tested, and know what the artificial intelligence has yet to learn.