How Werder Bremen’s AI technology can also benefit SMEs

Jiri Pavlenka got his job thanks to a computer. The Werder Bremen goalkeeper was discovered using artificial intelligence (AI). Computers trawled through hundreds of thousands of reports and pieces of data, produced by the club’s scouts over the years, in search of a new goalkeeper. An impossible task for a human – the data set is simply too large and there is too much information.

Werder can thank Roland Becker and his team for this achievement. His company, JUST ADD AI (JAAI), developed the scouting software and successfully trialled it with Werder Bremen. Becker has been selected as a 2019 IBM Champion by the US tech giant for his expertise in artificial intelligence. He also develops neural networks, and his contributions to the Rasa open-source AI project have seen him become a ‘Rasa Hero’. His team is made up of talented young people who use their knowledge to introduce cutting-edge AI technologies in companies. Berlin-born Becker has been living in Bremen since 2012. “It is the perfect location for an AI company. Bremen probably has more AI experts per head of population than anywhere else in Germany,” he says approvingly.

He talked to us about his ambitions, the opportunities presented by artificial intelligence and the scepticism and faith that the technology engenders.

Roland, AI is currently a hot topic in the media. Is this all hype, or is the attention justified?

Roland Becker: Both. AI is a little overhyped in the public perception, whereas in the reality of business it is rather underhyped.
Various technological breakthroughs in recent years, including Google’s AlphaGo and Tesla’s self-driving cars, and events such as the appearance of IBM’s Watson on the TV game show Jeopardy have generated a lot of publicity. This can raise false expectations, but also false hopes or fears.

At the same time, AI is evolving so rapidly that the current technology will appear antiquated in just a few years’ time. The speed of progress in this field is generally hugely underestimated. And AI is also taking hold in all sectors of industry. Anyone who believes that AI is not relevant to their sector is either missing out on a huge opportunity or is already playing catch-up with the competition – at the national level, the former is more often the case, while internationally it is more likely the latter.

What can companies currently expect from AI?

Becker: There is a popular perception that AI currently has no productive applications, but that is entirely wrong. Extremely powerful technology already exists and is widely used. Amazon, Google, Facebook et al rely heavily on AI – so much so that Google has an ‘AI first’ strategy. Facebook runs every uploaded photo through four neural networks to identify objects and faces using AI.

Companies can use AI to analyse unstructured data. At the heart of AI lies the ability to learn from examples. Enter a million images and the corresponding object classes into a system and it will independently learn how to differentiate between them. And better than any human could. Or the system can learn to search text for specific statements, irrespective of how these are formulated. That is what we did at Werder Bremen, for example.

AI needs one thing above all: data. Take, for example, a mechanical engineer in an SME. What kind of data does he have, and how can AI help him to benefit from it?

Becker: He might have data that is recorded by the sensors in the machinery. This could be used for predictive maintenance, which involves using past experiences to make predictions about the maintenance intervals required in the future. In quality control, cameras and x-ray equipment are sometimes used to identify faults or cracks in parts, and AI is now able to do that much more accurately than a human can.

In customer services, responses to enquiries with recurring themes can be automated, or recorded conversations can automatically be analysed using AI. This helps to identify escalated conversations so that they can be discussed with the team and the customer can be contacted, or more generally to obtain better statistics on the quality of service provided.

Does every company have to collect data now?

Becker: Many companies already have valuable data sets, but so far they have been unable to use them. AI can make this type of unstructured data, such as text, images and videos, more accessible. Up to 80 per cent of all available data falls into this category, but until now it was virtually impossible to evaluate it and use it in analyses. But now we can.

AI helps to identify patterns that would be impossible, or at least very difficult, to spot manually. These huge volumes of data can be identified and processed automatically on a scale and at a speed that is simply beyond humans.

How quickly does AI produce usable results?

Becker: You can’t just plug AI in and expect everything to work. You have to invest a considerable amount of time and effort in the beginning before you can reap the rewards of such a project. A typical time frame for return on investment is between six and twelve months.

How and when does your company, JUST ADD AI, get involved?

Becker: We help companies to understand the latest AI research and use it effectively. This includes conventional machine learning systems that use rules-based approaches, but we focus especially on deep learning and similar innovative solutions in AI research and combine them with context variables and safety rules.
This is what sets us apart from major players such as Microsoft or Amazon. The sheer size of their installations means that they are far less flexible and quick to react when it comes to implementing the latest findings into production.

Data security is also a key aspect. The systems that we build do not pass any data through the cloud or send it overseas. The customers retain full control over their data and the models that have been trained. We have set up our own local, offline GPU servers that we use for training on projects that require a lot of computing power and involve sensitive data. We work for a number of health insurance providers, for example, and data protection is an essential part of our solution.

Can you give us a few examples of JAAI projects?

Becker: There is JAAI Scout, of course, the football product used by Werder Bremen and other clubs. Our JAAI Agent, an intelligent conversational AI solution based on open source tools, is also very successful. This is a system that can respond to enquiries through a dialogue format. The enquiries can come via chat interfaces, voice platforms like Alexa and even telephone. Our product analyses all the data that is already available and uses the existing context, making it significantly more intelligent than conventional systems. It can logically respond to entirely new conversations, and a particular feature is the ease with which it can be trained in new topics.

Our work also extends to the field of computer vision, that is to say video analysis, as used in surveillance cameras, for example. Is that just a tree moving in the wind or did a figure just jump behind a bush? If a human can tell the difference, then we can capture that using AI too, even if the images from the camera are out of focus or pixelated.

Artificial intelligence and robots featuring AI are frequently met with considerable scepticism – it emerged only recently that self-driving Google cars were attacked with rocks and knives in California. Others warn that automation could lead to job losses. Where do you stand on these issues?

Becker: We need a public debate on the ethics, but it needs to run alongside AI development, not block its path. The example of the self-driving car illustrates this quite well. There is no doubt that 95 per cent of accidents never even happen with self-driving cars because the causes no longer exist: speeding, tailgating, alcohol or drugs and distractions from mobile phones. Yet there are people who would like self-driving cars to be kept off the roads until we have clarified how the autonomous car should behave in the event of the remaining 5 per cent of accidents. They become embroiled in often unresolvable discussions about which one of two people should be saved if only one of them can survive a particular situation. I think it is irresponsible to slow down development with this type of discussion while allowing the avoidable 95 per cent of accidents to keep on happening.

I believe the most important thing is to give as many people as possible a basic understanding of what AI is and how it works. This is what we are trying to do with our information portal, http://jaai.de, for example. Only once people have this basic knowledge can an effective public debate of the various facets of AI take place.

We should not forget that AI systems are programmed by humans and learn from historical data. There is always the danger that humans are unable to set targets correctly or that the historical data contains discriminatory patterns, for example, that we are unaware of. These can cause AI to aim for the wrong targets. AI cannot be blamed for these wrong targets, but it is certainly good at achieving them. That is why it is important to discuss these matters with the people who are creating AI systems and to ensure that mechanisms are in place to prevent such effects or at least to bring them to light.

That is one of the reasons you established Bremen.AI, a network for the AI industry…

Becker: Bremen is home to many outstanding minds in AI research, for example at the German Research Center for Artificial Intelligence, the University of Bremen and the various institutes. Bremen is definitely at the forefront when it comes to scientific expertise. What could be better is the interaction between academia and business.

If you compare the process of creating an AI system to cooking, you could say that there are many people in AI who know how to make good induction hobs or even better microwave ovens. But there are very few people who are really good at cooking, or who could open up a restaurant. The actual application of AI in practice requires experience, and this has to be gained in the real world. If I have watched hundreds of cooking videos, but never actually cooked anything, then I cannot really cook, let alone be an award-winning chef. Our objective is to help more people to learn to ‘cook’ with AI.

Bremen provides us with the perfect opportunity to create the right environment. By establishing Bremen.AI, we want to make a contribution to this by organising events and workshops, facilitating the exchange of information and fostering ideas.

So you enjoy living in Bremen?

Becker: On a personal level, definitely. But the city is also an excellent place to start a business in the AI sector. While it is still difficult to find people with practical experience here, it is not impossible, which it is almost everywhere else. There is a good flow of talented young people from the university, the German Research Center for Artificial Intelligence and the other institutes. We are currently in the fortunate position of having people actively applying to work for us. Over the past three weeks, I have received applications from some outstanding candidates. We are aiming to employ 20 AI experts by the end of 2020, and we are also establishing an accelerator for AI start-ups in Bremen: https://WeserValley.com

One final question: What has been the best moment in your job so far?

Becker: Every day I am delighted to be working in a field that is so interesting and exciting. My work combines all my interests. I originally wanted to study philosophy, but I ended up doing a degree in politics and economics. At the same time, I got more involved in the world of technology and set up my first internet company with a friend in 1999. Later on I also started programming. I am now able to combine the three fields I am most interested in – technology, philosophy and politics – into AI. AI is technically complex, can present highly philosophical dilemmas and will lead to considerable socio-political changes.
I particularly enjoy it when an AI model has finished its training, is working well and we can present it to an amazed customer.

Roland, many thanks for talking to us.

About Roland Becker:

Born and raised in Berlin, the entrepreneur has many years’ experience in the digital world. He set up his first internet company in the late 1990s, at the height of the dot-com boom. He studied politics and economics in Kiel, Berlin, Hamburg and Cape Town, and then lived and worked in Hamburg, where he established a number of digital companies. He moved to Bremen for family reasons in 2012. He joined forces with Bremen-based We4IT GmbH in 2016 to establish the AI start-up JUST ADD AI. He was selected as an IBM Champion in 2018 and 2019 for his expertise in the field of AI, and his contributions to the Rasa open-source AI project (AI-based language processing) have seen him become a ‘Rasa Hero’. He regularly speaks at international conferences and runs AI-themed workshops.