Artificial intelligence (AI) and increasingly complex algorithms currently influence our livesand our civilization more than ever. AI creates huge opportunities for businesses globally across all sectors. However, the use of AI also brings the potential for significant legal, ethical and reputational exposure. This is exactly the subject addressed by Jürgen Döllner, Professor at the Hasso-Plattner Institute (HPI) in the Digital Engineering faculty at the University of Potsdam, at the RiskNET Summit 2019. Döllner spoke about "Big data, machine learning and artificial intelligence: Reality, opportunities and misunderstandings". His introduction was clear: "We are experiencing a paradigm shift." He believes that this will involve artificial intelligence, AI, fundamentally changing the industry. He concludes: "In the coming years, everything will change." By this he means the use of AI in business, but also in terms of research and social challenges.
But where are the differences in the field of AI? Döllner began by recalling the history of AI developments, a story that stretches right back to the 1950s. Machine learning was in use from the 1980s, and today we also have deep learning. These are three areas that Döllner believes need to be kept separate. Deep learning is a special form of machine learning. According to the academic, it can trace its origins back to image recognition.
Döllner is focused on the AI areas of machine learning and deep learning. This is linked to the central issue for AI research: How can we convert an input into an output in a way that is non-procedural (not by specifying instructions)? This is the main theme of machine learning and deep learning. Essentially it is about predicting the output without previously specifying the rules used to obtain it.
Using the example of recognising a dog in a photo, he described how we get from initially classified training data to feature vectors. The final step is to evaluate these vectors to obtain a probability for the category of dog. Throughout the entire process it is crucial that the training data is right. Döllner describes this as the crux of machine learning and artificial intelligence. Döllner said: "It's all about using the right training data." It is also important to use the correct amount of training data. Otherwise, a system can actually be over-trained with data.
Big data, machine learning and analytics: The dream team
At the same time, large data volumes have distinct advantages. Without big data, machine learning would not be possible. Döllner said: "Data is the new oil. A resource that has to be processed." He added: "Machine learning is the refinery." In other words, data is first refined using machine learning.
Furthermore, these large data volumes are the basic prerequisite for analytics, such as prescriptive analytics and associated location intelligence. "With big data and analytics, we can reorganise many of our business processes", Döllner said. He is talking here about the dream team of big data, machine learning and analytics. For him, this means for example, predicting risks more effectively or predicting scarcity and expectations as early as possible.
At the same time, proposed actions can be generated and patterns identified. These huge data volumes combined with analysis methods need a corresponding amount of computing capacity, i.e. hardware, to obtain valid data and thus meaningful conclusions. The expert believes that specialist hardware working quickly and in parallel is vital for handling the large data volumes. In this context, Döllner also touched on the energy consumption of large server farms. And in other areas too, as Döllner emphasised by discussing the consumption associated with every single Google search.
Software engineering and the key to success
Döllner views the field of "software engineering" as a major success factor. "Software is the key to success", he explains. He believes that every sector is dependent on software in one way or another. One of the risks associated with this is that of software defects. Döllner warns: "Software defects could bring down entire companies." He cites the example of a mobile phone provider. "If a mobile phone provider is unable to start up its servers for just one day, it will suffer huge damage to its reputation." That is to say nothing of the pure financial losses involved. Despite the key function of software, engineering is still a type of craftsmanship. Software is a classic example of working in silos in an organisation. There are a large number of pots of data. And AI can provide valuable services at this interface. AI can be used to bring together the data from the different pots", concludes Döllner.
The key is to model human activities in software engineering and to answer the central questions of: "Who does what? and: How many people are working on a project in a particular period of time? This involves looking at the central element, namely recognising patterns. Döllner thinks that machine learning can make monitoring easier and faster here. Furthermore, it can help organisations to make more effective use of development resources and to put together optimum teams. Ultimately, Döllner says this will lead to errors being identified earlier and to processes being improved. In this context, Döllner summarises: "I see machine learning as the key as it can do more than people."
RiskNET Summit 2019 at Schloss Hohenkammer. A place full of history. According to legend, in the 8th Century there was a grand stone house at this site on the River Glonn – it was known as chamara. The name was derived from the Latin camera, meaning a room or chamber, which was translated as kamer in Middle-High German. The house was considered "noble" or "high", resulting in its modern name, meaning "high chamber".
Frank Romeike built a bridge between the ancient and modern worlds and provided a number of ideas for the ongoing development of risk management. Odysseus teaches us about scenario thinking, enabling us to learn from the future.