Computers and robots are becoming increasingly powerful thanks to artificial intelligence. Computing Science expert Oliver Kramer is convinced that humans will benefit. However, social discourse about the risks is important. A contribution to the Science Year 2018 "Working Worlds of the Future".
QUESTION: Mr Kramer, artificial intelligence - AI - is a buzzword that has been heard frequently in recent years in connection with autonomous cars, search engines and robots. What exactly does this mean?
KRAMER: Basically, it's like any other machine. Technology takes on individual tasks to support humans. A car, for example, is a machine that enables people to get around faster. AI means that a computer or robot takes over things that it can do better than humans. Today, we often have to deal with extremely large amounts of data, such as climate data, in which computers, unlike humans, can recognise certain correlations. Cameras can analyse images faster than humans. This can be important for the quality control of components on the assembly line. Thanks to AI, machines are currently so advanced that they are taking on more and more cognitive tasks. An ability that was long reserved for humans. A team of researchers recently programmed software so that it can recognise architectural styles in images of buildings. In a comparative test, it performed better than the experts.
QUESTION: Once again, it sounds as if humans have to fear that computers will make them superfluous...
KRAMER: ...an old and frequently voiced fear. Of course, we have to keep an eye on the disadvantages and engage in a broad public discourse about the risks. In Germany, the first chairs focussing on the ethical aspects of AI are currently being established. But I also see the advantages, especially with regard to the world of work. Big data is currently a big topic. The internet collects vast amounts of information about all of us. This data will be utilised more and more in the future. In a positive sense, this will result in products that we don't even know about today. Car movement data, for example, could be an important source of data in the future. These new digital products will be highly profitable - and a lot of people will be needed to programme all of this.
QUESTION: Some visions of the future go so far as to suggest that the real and virtual worlds will merge in our work in the future; is this a realistic idea?
KRAMER: Absolutely. Augmented reality, AR, is on the rise. Data glasses are being used into which information is superimposed. There are already the first applications in which instructions for the assembly of components are fed into AR glasses. Thanks to AR, even unskilled people will be able to carry out high-quality work in the future. This offers new opportunities for the labour market. Of course, the better robots and machines become, the more tasks they will take on. But we also know that the further development of AI will continue to create new jobs.
QUESTION: Do you have a measure of how intelligent AI already is today?
KRAMER: Cognitive abilities are a good illustration of this. Currently, machines can recognise the content of images as well as an average trained person if they look at them for a second or so. On the other hand, voice assistants such as Siri from Apple or Alexa from Amazon clearly show the limits. They dutifully answer questions such as "What will the weather be like tomorrow? A human would answer in context and ask "You're flying to San Francisco tomorrow. Do you mean the weather here or there?" I think that in about ten years, such assistants could be as advanced as humans and answer meaningfully in context.
QUESTION: You are primarily concerned with deep learning processes, i.e. the ability of machines to learn in a similar way to the deep, branched network of nerve cells in a brain. What is your focus here?
KRAMER: We are working on so-called neural networks that mimic the information processing in nerve cells. A neural network like this is trained for a specific task - such as recognising images. It is fed images of dogs, for example, so that it learns what a dog looks like, regardless of whether the dog is seen in profile, from the front, lying down or standing up. We specialise in genetic algorithms. These are computational rules that change the neural network through a kind of mutation. An evolution takes place in the computer, in which the neural network becomes better and better through various mutations. For example, we change the connections between the individual neurons in the network and then look at how this affects performance.
QUESTION: And these findings are then channelled into new products?
KRAMER: Not quite. Our aim is to develop new algorithms, new methods with which neural networks can be optimised. These networks can then be used universally. My team uses our methods for a wide range of applications. For example, we took part in a competition in which the neural network had to be trained to recognise surgical instruments used in eye operations in camera images from an operating theatre. These pieces of cutlery are all very delicate and very similar. Hospitals need this information to document the operations and for training purposes. However, we can also use our methods for weather forecasting. For example, we have used the neural networks to forecast wind speeds. Normally, you need complex meteorological calculation models for this. If we want to make a forecast for a specific area, we simply need to input wind measurements from other locations in the region around this target area.
QUESTION: Where do you think we will be in ten years' time?
KRAMER: Let's take the robots. Their performance is likely to increase significantly in the coming years with the advances in AI - especially because their cognitive abilities are improving enormously. Robots will be able to react faster, recognise situations better and orientate themselves better. Communication with humans will also become easier - for example, because machines will be able to recognise gestures better and better. I don't think that humans will become superfluous. Humans and machines will interact in a kind of symbiosis. Technology should first and foremost help people. And it will be able to do this better and better in future.
