Artificial Intelligence – Digital assistants, robotic surgeons, individualised customer profiles – business and professional life is changing thanks to improved data analysis using artificial intelligence (AI). What’s possible today? What can we expect from AI in the future?

Machines have taken off. Really? In view of the recent success of Alpha Go, the artificial intelligence (AI) developed by the Alphabet subsidiary Deepmind, this statement does not seem to be so far-fetched. The third model, Alpha Go Zero, was fed and trained solely with the rules of the Asian board game Go, but not with human gaming strategies, when it played against itself. Within three days, it had reached the level of the predecessor model Alpha Go from 2016 and beat this human-trained system 100 to nil. The reigning Go World Champion describes Alpha Go as Go-God. „AI is a game-changer. For us researchers, it’s a bit like the Wild Wild West at the moment,“ says Damian Borth. He is the head of the Competence Center Deep Learning at the German Research Center for Artificial Intelligence (DFKI), which among other advises companies on AI systems. What was for a long time only possible in science fiction films now seems to be within our reach – AI has become the buzzword for the next technical revolution. Computers can suddenly detect cancer, drive our cars and know what kind of music we like better than we do. Silicon Valley’s major technology companies are turning out intelligent products and apps thick and fast.


According to a survey conducted on behalf of the US data analysis specialist Teradata, 80 percent of major international companies are already investing in artificial intelligence. The companies surveyed expect the biggest impact from AI in the ICT industry (59 percent), in B2B services and consultancy (43 percent) and in B2C and financial services. But for all the hype, there are still critics. The philosopher Nick Bostrom warned against smart machines in 2016: „We’re like children playing with a bomb.“ Back in 2015, Tesla founder Elon Musk and science legend Stephen Hawking were already demanding  that the new technologies should benefit society as a whole. Stefan Wess has been working with AI for more than 30 years and he remains calm. The CEO of software provider Empolis Information Management said: „I see it as more of an engineering issue“. AI concepts have been in use for a long time, for example in the programming language Java or in virus scanners.


Keep your feet on the ground

DFKI researcher Borth has no time for exaggerations either. Marketing likes using terms such as „superintelligence“ or „superhuman performance“.  But he goes to explain: „Although driverless cars recognise road signs better than we humans, they are not yet a kind of Hollywood superintelligence. We are still a long way from that”. The current hype is all about machine learning, especially deep learning, a small subfield of AI research. The goal is to simulate the abilities of the human brain through so-called artificial neural networks. These networks learn by watching and trying things out, just like a child. Originally they had three layers. In 2012, the AlexNet network, which has eight layers, won the visual recognition competition Image Net Challenge, a milestone in the development of deep learning. Today there are networks with hundreds or thousands of layers. The more layers a neural network has, the more data and computing power it needs for training and the better it gets, for example in visual and speech recognition. Science distinguishes between weak and strong AI. As deep learning expert Borth explains, weak or engineering AI can already be found in various applications today: a complex machine that looks intelligent but basically only processes data according to given rules. These machines are like idiot savants who convert spoken language into

text or direct a car into a parking space. „But this AI has no awareness“, says Stefan Wess.


Smart, smarter, AI

And yet this weak artificial intelligence is already changing our everyday lives as it is introducing automation into intellectual activities. „Smart software can search databases more efficiently, make a company’s knowledge more readily usable in the age of big data and consequently generate new business models and profits,“ explains Jörg Bollow, Executive Director Marketing DACH at Bisnode. In law firms, „Lawtech“ filters relevant data for legal cases from large files in just a few minutes. Intelligent language software helps personnel officers to select applicants. Insurance companies handle claims using robot-based process automation. In the retail sector, customers‘ requirements can be identified and served much more individually from unstructured data such as social media using AI. A robot consultant generates investment and savings tips from the customer data collected while the human bank advisor acts as a middleman between customer and machine.


Brave New AI World?

Machine learning is getting things moving in the economy. Google is pushing ahead and is about to disrupt its own business model since voice-controlled software eliminates the graphic interface as a platform for advertising revenue. Almost everything Google does involves AI, be it the Google Assistant, the picture service or the driverless car. Even if smart machines don’t shake up all companies quite so dramatically, these new technologies offer great opportunities. Start-ups can use the open source software from the major players to develop new AI-based applications. More and more young players are advising established corporate groups on machine learning. The learning machines personalise and improve customer service.

Do German companies have any chance at all against the top players from Silicon Valley? Borth thinks they do. „Of course, Google and Facebook have a direct line to the end customer and are therefore more visible. Here in Germany, however, we are very good in the less visible B2B sector.“ Nevertheless, a lot more should be invested in research and development in order to increase competitiveness. AI pioneer Wess also predicts that Germany will benefit more from the wave of digitalisation than the USA. On the threshold to intelligent products, German companies could play off their strengths in the production of goods. According to a survey by Crisp Research, 64 percent of companies in Germany were already investigating the possibilities of machine learning in 2016. However, according to the DIGITAL 2017 business monitoring report, only two percent of commercial companies use AI opportunities. Among ICT companies in Germany, the figure rises to 15 percent, but for the mechanical engineering companies it’s just three percent. 79 percent of all companies surveyed stated that AI is not yet an issue for them.


Modern working world

As with all other technical revolutions, machine learning is changing the working world. Occupational profiles are changing, some jobs will fall victim to automation and new job profiles are being created. The management consultancy firm Accenture expects three future AI job profiles: The trainer will make the AI systems smart, the explainer will evaluate and process the results of the smart software and finally, the sustainer will ensure that smart technologies do not cross ethical boundaries. In the current phase of upheaval, Wess sees a lack of appropriate political debate, for example on the subject of unconditional basic income. Facebook founder Mark Zuckerberg is now demanding such a basic income. Borth focuses on the younger generation: Everyone is talking about the three million US truck drivers who will lose their jobs because of driverless trucks. But nobody’s talking about the fact that these drivers will retire some day and nobody wants to do the job nowadays anyway.“ This is why it is so important to invest in the right education and training. Together with the German National Academy of Science and Engineering acatech, DFKI developed the first online course on machine learning in German, which attracted around 5,000 participants.


En route to AI certification

More and more people are also concerned about the ethical dimension of machine learning. Google, Facebook and Co. formed an AI partnership in 2016 where they are developing social and ethical best practice rules for AI. The scientist Borth states: „We have to discuss the goals and boundaries of AI technology. This must be decided by social consensus, not in some Silicon Valley corporation or by an isolated politician.“ Wess goes on to say: „The biggest danger is that AI systems may acquire our prejudices during the learning process.“ The DFKI is already collaborating with the Volkswagen Foundation on a classification system, a kind of AI certification. Companies would then have to abide by these compliance rules with their smart machines.


Networks are growing and growing

A lot of Germans are cautiously positive about this development. According to a recent survey by PwC, 77 percent of Germans expect that AI will help people to organise their everyday life better. However, 51 percent are also afraid of development. Is this fear justified? „Today’s neural networks simulate the brain of a bee. Thanks to better graphics cards and big data, they are growing and growing. In purely mathematical terms the performance of the human brain could be simulated by 2028,“ explains Jörg Bollow. „But these neuronal networks would still not have any real consciousness, so they wouldn’t be a strong AI,“ adds Wess. Borth agrees with him: „We don’t even really fully understand how the human brain and consciousness work.“ Is a strong AI a real possibility? „I change my mind about that every day,“ says Wess. But artificial consciousness aside, it is impossible to imagine everyday life without learning machines.



Artificial intelligence describes the capacity of an IT system to show human-like intelligent behaviour. This requires certain core skills in different proportions: perception, understanding, ability to act and learn. These skills are supplementary to the basic principle of all IT systems: Input – Process – Output. The really new feature is learning. Common to all today’s „real“ AI systems is that they are also trained in the processing component and can thus learn.