Group 3 Copy
0 Results
Schließen
Dec 21, 2018

AI in focus: The cards are being reshuffled

It's the talk of the day: artificial intelligence (AI). If you want to join in the conversations, you shouldn't miss our new series "AI in focus" on this groundbreaking topic.
AI is a large field. From a technical point of view, AI is associated with terms like "machine learning", "deep learning", "decision support", "chatbots" and "autonomous driving". However, creative, social and ethical aspects also play a role: How do we want to interact with AI? How does AI change our everyday lives? In the following months, Michael Burmester will introduce you to the challenges of human-AI interaction. In the first part he explains what AI actually is. Moreover, he illustrates the changes we have to face whenever we interact with an AI System.

Let's start from scratch: What is AI?

We often put AI on a level with humanoid intelligence. However, there is still no universal definition of this type of intelligence – so how can we possibly define AI?
There are certain core capabilities that constitute AI: perceiving, understanding, acting and learning*. Traditional computer systems always follow the same principle: input (perceiving) – processing (understanding) – output (acting). This system needs to be adapted and upgraded with the signature core ability "learning" in order to account for AI.

News Kikonkret Teil1 Bild1 En 935

Perceiving

People sense their surroundings. AI also uses various sensors to collect data: microphones to gather audio files such as spoken language, and cameras to record visual data of faces or objects. In addition, numerous other data can be accumulated through machines and products.

Understanding

AI processes, manages and understands data. This enables the system to construe meanings and connections. If, for example, the word "complaint" appears in an e-mail, the system forwards this mail to a company's customer service department. This comprehension skill is made possible by machine learning, a subfield of AI: A system is fed with data and trained to independently generate new knowledge from experience.

Acting

This process of understanding entails different actions on the side of AI systems. Robots move, drill, mill, weld, glue or take objects from one place to another, i.e. they support us in our work. And they increasingly appear in private contexts, for example as useful robot vacuum cleaners. Other AI systems make decisions and give informed advice using written or spoken texts. Moreover, AI is able to control entire production plants or smart homes.

Learning

AI systems use the learning capability in two ways: During the process of understanding, a system learns when it is trained on the basis of data. In this phase, a learning algorithm is created and then used by the AI system to extrapolate knowledge. In addition, AI systems learn from feedback loops within a process. Successful actions are reinforced, actions that have proven to yield less success are avoided or amended. When we buy a suggested product online, the system will judge this action as being positive. The consequence: We will receive further suggestions for products that are often purchased together with the original product.

Changes in human-AI interaction

The following changes in human-machine interaction can be observed when AI comes into play:

News Kikonkret Teil1 Bild2 935

Learning Systems

In contrast to traditional computer systems, AI systems analyze data, continuously enlarging their knowledge to adapt their actions. Users notice these changes since systems act differently than before. That way, a common principle in dialogue design is attenuated: conformity with user expectations. The more predictive a system's behavior, the better for usability. In fact, current studies on the usage of AI show that users perceive AI systems as showing little conformity with their expectations.

Initiative to interact

Up to now, humans took the initiative and computers reacted. With AI, the initiative to interact increasingly comes from the system. It makes suggestions to solve problems, points to correlations or manages background processes.

Human-machine Team

Humans and AI systems team up: More initiative from technology means that humans and AI systems share tasks in a way that requires less interaction and more cooperation from humans. Work processes are harmonized and alternated. This cooperation has benefits for humans and AI alike: Whenever semi-autonomous systems don't know how to continue, they can ask humans, allowing them to act even in unfamiliar contexts.

News Kikonkret Teil1 Bild3 En 935

Voice user interface

Humans and machines not only communicate using language. Chatbots have only improved to reach current quality levels with the advent of AI. In some cases, voice interaction has become so natural that it is hard to tell humans and machines apart. However, this kind of interaction is not suitable for every setting, for example open-plan offices.

Autonomous Systems

AI systems have reached a level of autonomy that hardly requires any human interaction. Nevertheless, humans still need to be able to interact and communicate with them – just think of autonomous cars. It is necessary for AI systems and other road users, such as pedestrians, to coordinate their actions in order to avoid accidents.

News Kikonkret Teil1 Bild4 935

Perception of AI

Studies have revealed that humans do perceive computers as social beings. But if we see human-machine interaction as social interaction, we have to define the role of the AI system in this relationship. Is it more like a butler? Or rather like a friend?

As we have seen, AI changes human-machine interaction. Humans and AI systems teaming up? This may be hard to imagine. In our second part of "AI in focus" we will show you what this might look like.

News Buzzword Burmester 935

The author

Michael Burmester, professor at the Stuttgart Media University HdM and UX expert, writes for UID.

* A Bitkom paper (in German) published in 2017 defines several capabilities that distinguish AI from traditional computer systems. The interaction of these characteristics is presented in a model.