Machines have long been more than just high-performance, but ultimately simple-minded robots. Machines are constantly learning, just like intelligent humans — and even better.
Even today, there is still no universal definition of intelligence, but experts are already putting all the computing power available to work on artificial intelligence, or AI for short. One practical example of this technology has been the often-mocked smart refrigerator, but AI has now evolved into a great deal more than just nice-to-have playthings. AI touches on existential issues, such as the future of one’s own job or entire sectors of the economy. With artificial intelligence, intelligent machines and the underlying “Deep Learning” technology either already affect everyone or will soon do so, whether consciously or unconsciously, voluntarily or not.
Are machines also just human?
Learning machines are similar to learning humans: Through diligent memorization, a student may be able to obtain a high school diploma, but that does not mean that he is intelligent. The same applies to a computer that — which although vastly superior to humans where speed and precision is concerned and able to work tirelessly, whether it is calculating an Excel table or the next moon landing — is simply performing the rules and commands previously programmed into it by humans. Until now, the following held true: Everything that can be formulated in the form of rules and calculated can be done better, faster, and more precisely by a machine. Intelligence, on the other hand, is reserved for humans, as only they can collect experience and learn from it, recognize problems, come up with solutions, make mistakes, and do things better.
However, machines have now learned how to learn, even in areas in which homo sapiens has thought itself irreplaceable: Understanding spoken language and communicating with it meaningfully, as well as recognizing emotions and reacting to them. Like humans, intelligent machines learn by observation, imitation, trial and error, and even find their own solution paths. This is made possible above all by powerful Deep Learning Systems (see explanation of terms).
What does this mean for the world of work?
The answer is: “Nobody can reliably predict how exactly individual professions will develop in the future”, says Ulrich Walwei, deputy director of the Institute for Employment Research (IAB). Despite this, there exist numerous studies on this topic, which occasionally result in sensational headlines: “The Job-Takers Are Coming” (Spiegel Online).
In the USA, 47 percent of jobs are seen as being in danger, including taxi, bus, and truck drivers — due to autonomous driving technology — as well as postal workers, which could increasingly be replaced with logistics drones. According to a study by ING DiBa, the situation in Germany is more dire, with 59 percent of jobs in danger of being replaced, above all administrative positions and jobs which mostly require low- to medium-level qualifications. On the other hand, management executives and academics in scientific and creative professions are unlikely to be replaced by robot colleagues — even if robot doctors are already appearing on the deep learning horizon which can make diagnoses and issue prescriptions. Generation Z, which is currently growing up in this era, might soon be able to look forward to online marriages or utilize divorce-to-go services, reducing the need for registrars, lawyers, and courts, or even making them superfluous altogether — all made possible by the blockchain (see Innovation Insights 1 / 2017).
Out with the programmers, in with the data scientists
According to the IAB, 70 percent of manufacturing jobs in particular could be automated now or in the near future, with similar figures for jobs in the finance and accounting sector. Even professions which until lately have been viewed as guaranteed job security are now becoming outdated faster than the latest iPhone model. In particular, jobs in IT and the natural sciences comprise a large number of routine tasks, giving them a high substitutability potential of more than 65 percent. On the other hand, those who stand to benefit the most are highly specialized IT and technical staff such as data scientists — for the foreseeable future, we will still require humans with natural intelligence in order to provide artificial intelligence with sufficient momentum until it is able to — someday — develop on its own. One other sector which stands to gain is the service industry, and numerous new tasks will arise which require both a great deal of digital know-how as well as competencies which (as yet) cannot be digitalized, such as creativity, communication skills, and the ability to work in a team. Currently in development are excavators which can excavate hazardous materials without human intervention, drones which can inspect vents from the air, and polymorphous robots which can independently service bridges, canal systems, and machines.
However, Ulrich Walwei dismisses highly dramatic future scenarios: “Digitalization does not mean mass unemployment”. In May 2016, he wrote the following on ZEIT Online: “From the current standpoint, a massive reduction in the demand for labor as a result of digitalization is rather unlikely. The loss of jobs … will probably be balanced out by new jobs in other areas.”
But what is clear is that AI can no longer be dismissed with a smile and a smug remark about the utility or absurdity of smart refrigerators. Startups emerge virtually daily, like new synapses in a learning brain, that translate the infinite possibilities of AI into concrete applications and develop new products, services, and business models.
AI Startups from Germany and Switzerland
With EyeEm, each of the photo app’s 20 million members (as of June 2017, according to the website) can sell their photos. In order to do so, appropriate keywords and a meaningful description are important. A neural network has learned to use millions of sample images to correctly identify objects in photos and to label the images correctly.
The company “Twenty Billion Neurons” develops deep learning solutions for German industrial companies, such as for the development of self-driving cars, highly efficient speech, image, and video analyses, for dialog systems and medical diagnostics; i.e. for all applications which require data to be analyzed in an intelligent manner.
This startup specializes in the analysis of aerial images and satellite photos. These analyses are then used e.g. by insurance companies or map providers who require precise information on the type of buildings.
The AI application of the startup analyzes e-mails and other messages from customers and provides service staff with suggested replies. The information is based on several million conversations which were evaluated using Deep Learning. Currently, the system understands more than 200 different reasons for establishing contact and is constantly learning.
This spin-off from ETH Zurich develops Smart City Technologies specifically for traffic congested (inner) cities. Smart Parking delivers occupancy information from parking lots in real time, which serves as the basis for optimal parking administration.
Starmind develops self-learning Deep Learning systems for companies. It is a type of central brain of all employees, which allows for collective knowledge to be utilized optimally for questions and solutions to problems. The development of Starmind was based on work done on virtual brain tissue and self-learning neural networks.
Brief explanation of terms
Artificial Intelligence (or AI for short) is a branch of computer science which aims to make it possible for machines to imitate intelligent human behavior (the fact that humans certainly do not always behave intelligently is another matter altogether).
Artificial neural networks: Layers of artificial neurons which are connected to each other in a manner similar to nerve cells. Artificial neural networks form the basis for machine learning processes based on how nerve cells are linked in the brain.
Deep learning refers to teaching artificial neural networks to identify patterns in information, to classify them, and to categorize them. Every piece of information and every decision will then be examined once again through Deep Learning. If an assumption is confirmed, a particular information link is assigned a higher level of importance; if it is revised, it receives a new, less important link. Over time, this results in the development of an increasingly intelligent system, such as what takes place in human brains — in ideal cases.