Foto: Bax Lindhardt

“The world is changing under our feet”

Wednesday 09 Dec 20

Contact

Per B. Brockhoff
Professor, Head of department
DTU Compute
+45 45 25 33 65

Contact

Lars Kai Hansen
Professor, head of section
DTU Compute
+45 45 25 38 89

Contact

Thomas Bolander
Professor
DTU Compute
+45 45 25 37 15

Contact

Henrik Madsen
Professor, Head of section
DTU Compute
+45 45 25 34 08

Contact

Ole Ravn
Professor, Head of Group, Study Line Coordinator Automation and Robot Technology
DTU Electrical Engineering
+45 45 25 35 60

4 key ‘Safe AI’ principles:

• AI must be safe. It is technically possible to make AI safe through testing, validation, and verification.

• AI must be able to keep a secret. It is technically possible to incorporate privacy protection, so-called ‘privacy by design’.

• AI must have well-defined values. It is technically possible to test AI for values and to get AI to perceive, respond to, and communicate with emotions.

• AI must be transparent and communicative.

 

Read more on https://www.compute.dtu.dk/Kunstig-Intelligens-DTU.

Artificial intelligence (AI) has the potential radically to change the way we live our lives. But the growing use of AI is also a cause for concern.

Close your eyes. Imagine that you are coming home from work and your household robot has already cleaned and vacuumed the house, and cooked dinner. The robot has become a natural part of the family, and it understands what you expect from it.

The household robot is still a dream scenario, but it could become one of the next big future breakthroughs. Technologies that think like humans are already entering our everyday lives and are going to occupy a more active role than previously. In fact, artificial intelligence (AI) will give us the opportunity to solve some of our society’s small and major challenges, ranging from reducing CO2 emissions to fighting cancer, developing new pharmaceuticals, revolutionizing transport, and creating new teaching forms.

“In many ways, we’re seeing an acceleration of a development we’ve been going through for decades. Right now, the world is really changing under our feet—especially in data-based and machine learning-based artificial intelligence,” says Per B. Brockhoff, Professor and Head of Department at DTU Compute. 

“The volume and quantity of data generated—combined with computing power, communication speeds, and learning opportunities for data, have evolved so rapidly that we can suddenly do things that we could have done in theory 30 years ago. Only then, they didn’t work in practice.”

Among the best in Denmark

The development is clearly felt at DTU. One of the places is DTU Electrical Engineering, where the researchers work to integrate artificial intelligence into robots. Here, the experience is that robots are becoming increasingly autonomous. According to Ole Ravn, Professor at the Department of Electrical Engineering and head of DTU’s automation and robot technology research, we are facinga breakthrough with collaborating robots—theso-called cobots. The technology is now so developed that robots are equipped with situational awareness, which, for example, enables them to read their surroundings. This will mean that, in future, robots will become more usable and will be able to work more closely with humans.

In another place at DTU, you will find some of Denmark’s foremost research environments in artificial intelligence. Here the research is gathered in two large groups. In one section—Algorithms, Logic and Graphs—research is conducted into artificial intelligence at the intersection between logics, mathematics, and computer science. Here we find Professor Thomas Bolander from DTU Compute, whose work includes teaching DTU’s robot R2DTU to function socially and flexibly with other robots and humans.

In the Section for Cognitive Systems, Professor Lars Kai Hansen from DTU Compute and his team work with machine learning, cognition, and social behaviour. Here the intense basic research conducted for many years has given the researchers a better understanding of sound and images with the help of data. This has had great importance for the development of methods in, for example, medical imaging, where DTU was the first to use machine learning to interpret brain scans.

Today, the work with machine learning and artificial intelligence has created a fertile ground for many start-ups. One of them is Corti, which—using voice recognition—assists healthcare professionals when someone calls the emergency number 112. This is done using advanced machine learning technology that can analyse the way a person breathes or formulates himself/herself. This makes it possible to reduce the number of unidentified heart attacks by more than 50 per cent.

Another example of machine learning is seen at Amazon, which uses the technology to provide users with recommendations to buy products based on their preferences. In addition, Amazon makes use of nearly 200,000 warehouse robots, which are based on a combination of robotic technology and AI. Amazon also uses AI for stock management purposes by being able to predict what people intend to buy.

From black to white

However, machine learning can also take a more worrying turn. Recently, Lars Kai Hansen was himself surprised to see a post on Twitter in which an AI tool had reconstructed a pixelated image of former US President Barack Obama and turned him into a white man. The case is an example of an algorithmic bias that can occur when algorithms are trained using databases that are not representative, for example by having a large overweight of white male faces. This creates ethical dilemmas.

“We see that AI creates greater focus on fundamental values and provides new opportunities for discovering inequalities and injustice. That’s why we give great priority to educating our students in ethics, and we’ve also formulated a number of ‘Safe AI’ principles that constitute a vision for responsible artificial intelligence, based on tangible and realistic AI technology. We’ve done this to ensure that the technology meets democratic values and offers the possibility of control,” says Lars Kai Hansen.

Evil robots

He points out that—in addition to the ethical issues—there is also great focus on sustainability in connection with AI systems. On the one hand, new methods must be found to make AI less energy intensive, and, on the other hand, AI is the key to an effective green transition and a sustainable world in a wider sense. According to Henrik Madsen, Professor at DTU Compute, this is very much about making the next generation of data centres much greener. Therefore, DTU has spearheaded a couple of new major projects focusing on demonstrating that AI is the key to activating the flexibility in all parts of society necessary to implement a sustainable, fossil-free energy system.

A completely different concern often mentioned in the media is whether artificial intelligence and robots will take over world domination. But Thomas Bolander is less pessimistic:

“In principle, it’s relevant to be concerned. But it must be on the right scale. A horror scenario with evil robots is, in any case, very unlikely.”

“Such robots will not begin creating their own need to do something they aren’t programmed for. But that doesn’t mean that robots can’t do evil things. For example, if you develop technology that can be used for a driverless car, you have all the ingredients to create an autonomous ‘killing machine’. So we obviously need to pay close attention to this.”

What is artificial intelligence?

Artificial intelligence is primarily about getting computers and robots to do things that humans have been able to do until now, for example playing chess, driving, diagnosing patients, or having a conversation.

Machine learning is the backbone of data-driven artificial intelligence. Machine learning is the statistical way of creating artificial intelligence and it functions by training the computer to recognize data patterns. Based on data, Google learns, for example, what you want your mailbox to look like, and Twitter learns which tweets you are to see, just as Facebook decides which ads you are shown. Deep learning goes one step further and recognizes patterns using deep neural networks that can learn complicated connections based on very large data volumes.

In the symbolic branch of artificial intelligence, researchers are directly trying to create simplified models of some of the highest levels of human thinking: our linguistic, conscious, and logical thinking.

Work with artificial intelligence is being done at DTU Compute and DTU Electrical Engineering, among other DTU departments.

Source: DTU, Danish Technological Institute