The participants in our expert discussion see a great need to inform people in particular about the opportunities and possibilities of Artificial Intelligence. Even if issues such as ethics and bias indeed pose a challenge – no one is worried about a super-intelligence that would replace human beings.
The image that people have of Artificial Intelligence is quite distorted. “On the one hand, the expectations placed on the capabilities of AI are huge; on the other hand, there is also the fear that super-smart Artificial Intelligence will take over and rule the world,” says Andrea Martin, Chief Technology Officer of IBM for the DACH region. Hollywood has especially shaped this image of a future ruled by machines, with blockbusters such as the “Terminator” series of films. However, statements by renowned scientists, such as Stephen Hawking, have also perturbed people: “When I hear that AI is going to kill us at some point, I just shake my head in dismay,” declares René Büst, Director of Technology Research at Arago. “Assertions such as that only serve to frighten people. And that makes it extremely difficult for us, as providers, to make clear just what AI is capable of and what not.” In this regard, the round-table participants also view the media as being obligated to not only report on horror scenarios in conjunction with Artificial Intelligence but also its opportunities and possibilities. “Unfortunately, though, my impression is that the press would rather focus on sensational doomsday theories because they simply attract more reader attention,” states Reinhard Karger, Spokesman for the German Research Centre for Artificial Intelligence (DFKI).
Yet, on the other hand, the providers of systems are also to blame for the distorted public image of AI: “With high marketing effort and expense, systems are being launched onto the market that are actually rather trivial,” says Thomas Staudinger, Vice President of Marketing at EBV. This creates expectations which many systems ultimately cannot fulfil.
“We should look at the opportunities and possibilities that Artificial Intelligence offers to us.”
Andrea Martin, Chief Technology Officer IBM DACH
More applications than assumed
A typical example is chatbots: many users expect one could conduct a conversation with the digital assistants as with a human. “We will probably not even be able to do this in the long term,” says René Büst. “We should instead try to start with smaller tasks, such as the automation of processes.” Andrea Martin also agrees: “Personalised interaction is only one aspect of Artificial Intelligence. In addition, we should see how AI can help us to gain new insights from enormous amounts of data and thereby assist us in making better decisions.”
The experts can list a whole range of applications where AI is already being used today, delivering real benefits to users: whether it’s predictive maintenance, the organisation of work, in medical research, or in many other areas where AI systems support people. In fact, many more AI solutions have already been deployed than most people realise. Oliver Gürtler, Senior Director Business Group Cloud & Enterprise, Microsoft Germany, makes this very clear: “As of today, 760,000 developers at our partners are already developing solutions that take advantage of AI. And this is only on our platform – there are, of course, other vendors.” This fits in with the figures which Andrea Martin cites: “In 2018, some 70 per cent of all developers will, in one way or another, be integrating Artificial Intelligence capabilities into their products.”
“The decisions made with AI must be transparent and reproducible.”
Oliver Gürtler, Senior Director Business Group Cloud + Enterprise, Microsoft Germany
Various technologies work together
This rapid progress is made possible through the interaction of various technologies, as René Büst explains: “With cloud computing, big data, analytics services and GPUs, the foundation was laid for today’s AI solutions over the past ten years.” Above all else, DFKI Spokesman Karger sees the possibility of massive parallel processing of graphics data, in order to compute neural networks, as having enabled a major breakthrough for AI: “Today, we finally have a supercomputer which we can, for example, integrate into a car in order to process sensor data in real time.” In addition to the possibility of providing computing power to an application via the processors, Thomas Staudinger sees a further building block: “Thanks to the developments in the field of semiconductor technology, the sensors have become so inexpensive that they can be integrated in applications on a wide scale.” Consequently, the data volumes required for AI solutions can be generated. “Data is the crude oil for Artificial Intelligence,” emphasises Oliver Gürtler. “In order to process it, for one thing, databases that can be accessed in milliseconds are required. For another, processing is done much more directly at the front end – on the devices. I also need connectivity to exchange data between devices and data centres.” The development will continue, as the round-table participants emphasised. Among other things, they cited mesh computing, in which end-user devices communicate directly with each other without an Internet connection, or quantum computers – with which initial testing has already been carried out.
“I see AI as a companion technology that helps us to more easily determine our lives.”
René Büst, Director of Technology Research, Arago
Secure AI systems
“Technology development has accelerated exponentially. A major challenge in this range of topics is transparency,” says Oliver Gürtler. If you don’t know why an AI application has made a certain decision, then the results could be easily manipulated without the user’s knowledge. This also means that the results must be reproducible. And, of course, that the data is protected to prevent tampering. “If there is no transparency, i.e. it is not apparent just how AI works at all, then the user has to accept the results without questioning.” The security of AI systems is based on many pillars, as Andrea Martin explains: “Security must be ensured in the hardware, in the software and in the connectivity.
And, then there is a further pillar – human beings. If we ignore people, then we have neglected the most important factor.” Reinhard Karger also views humans, or more specifically the lack of security awareness on the part of users, as being a major risk factor. “In all of the discussions regarding data protection and security, the focus is always on putting better locks on the doors – while the windows are wide open.”
“There is a plurality of assistant systems, but no singularity of AI.”
Reinhard Karger, Company Spokesman of the German Research Centre for Artificial Intelligence
The challenge of bias
Whereas data protection and data security are already familiar issues from many other areas, AI solutions pose an additional aspect: personal bias. “If facial recognition is developed by a team which, for example, consists of a group of white men, it could happen that the system returns incorrect results for people who do not have white skin colour,” explains Oliver Gürtler as an example. “This can be avoided if diversity is taken into account in the development teams. The industry is only finding out about this right now.” He stresses that guidelines are necessary for AI developers for this. “Microsoft, IBM, Google, Facebook, Amazon and a few other companies have founded the non-profit organisation ‘Partnership on AI’ to address this,” Andrea Martin mentions.
“There they jointly devise principles for best practices in the development of AI solutions. This is so that the things that we do can benefit society as well as business to the greatest extent possible, and not be in conflict with them.” Reinhard Karger points out just how difficult this can be: “Massive amounts of data are required to train neural networks. But how can we verify whether this data actually corresponds to the demands of today’s society? Should thousands of people be trained to individually check the data? What are the criteria? And, must this examination be repeated on a regular basis? It is very difficult to eliminate the bias of a system.”
“Especially with an aim towards productivity and efficiency gains, Artificial Intelligence can trigger very positive developments.”
Thomas Staudinger, Vice President Marketing, EBV
Rules for AI development
A similar challenge is the question of the ethics of decisions made by AI systems. What should an autonomously driven car do, for example, if it is in an emergency situation and must choose between the life of the driver, a child on the street, or an old man on the sidewalk? “In my view this is much more important than the discussion as to whether machines enslave us at some point,” says René Büst. However, Reinhard Karger is a bit annoyed by this issue: “The probability of such a scenario occurring is infinitesimally small. When humans get into such disastrous situations, they react reflexively and do not decide according to ethical principles, which do not even exist for such a difficult dilemma.” But here, he comes up against contradiction: “There is a lot of uncertainty in this context because there are so many unanswered questions,”
surmises Oliver Gürtler, for example. “These need to be answered before decisions made by an AI solution can be accepted – not only with regard to autonomous driving but also regarding applications such as those in the field of medicine.” Andrea Martin, too, considers the issue of ethics in conjunction with AI to be pertinent: “The question is, of course, raised by that segment of society which is not so intensely preoccupied with Artificial Intelligence. We still have a very long way to go, however; there are still many technical basics to be mastered before a car has to make such decisions. I therefore believe that we still have time to answer the question.” Nevertheless, just as with René Büst, she sees the need to form ethics commissions which are dedicated to resolving moral issues arising from AI use. “Not to frighten people, but instead to make AI socially acceptable,” Büst emphasizes.
Approaches to integration in companies
Consumers are one thing, says Thomas Staudinger: “But how can I, as a medium-sized company, integrate AI into my business processes? AI is not an end in itself.” Ultimately, it is a question of the strategy which acompany’s management must set forth, declares Oliver Gürtler, among others, and emphasises: “AI offers a great opportunity for an enterprise to differentiate themselves from the competition.” As a company, though, what can you really do with AI? How can you set up a new business model based on it? What processes can be improved with it? According to Staudinger, these are all questions which confront many decision-makers in companies today: “There is still great need for clarification. Best practice examples could help to make the benefits of AI solutions more understandable and the topic more tangible.”
Andrea Martin advises to switch the perspective: “The field of AI is so broad that it could crush companies. Instead of seeing it as a giant beast which needs to be slain, one should clearly understand that AI is a modular concept. You can pick out individual elements and create dedicated solutions.” René Büst recommends to first start by automating processes in a company’s IT area with the help of AI. “With the information you collect while doing this, you can then extend AI solutions to other business processes and, to the same extent, create an awareness in the company.” Thomas Staudinger, who has his focus on the end-customers of his customers, sees the responsibility at the decision-maker level: “It’s all about providing the end customer with better service based on AI. I can’t achieve this with IT; instead, this is a business decision.” To the contrary, Oliver Gürtler recommends just getting on with it: “In our experience, it is extremely productive if both young professionals and experienced staff creatively work together, then simply implement the solution. To begin with, these are all mini-projects which do not require five years of software development, but can instead be implemented as a very “lean” solution to supplement the existing IT. Once companies start to see the benefits of AI, then they will also learn to understand AI for themselves.” This would lead to the emergence of truly transformative projects, according to Gürtler. Reinhard Karger also thinks that much of the potential that lies in staff: “For small businesses in particular, it makes sense for them to ask their own employees – through the company-internal suggestion programme, for example. Because, above all else, imagination is needed to introduce AI aas opposed to technological expertise.”
Realising opportunities
All participants agree that one should look at the opportunities offered by Artificial Intelligence – in production, research, and logistics, but also in the back office with regard to taxes, insurance, and much, much more. “It’s a long way between the fictional super-intelligence of a Terminator and a simple assistant,” stresses Andrea Martin. The round-table participants view AI above all as a possibility to supplement human capabilities – not replace them. Consequently, Reinhard Karger also surmises optimistically: “Yes, Artificial Intelligence will change our lives – and that is terrific.”