An interview with Professor Dr.-Ing. Cristóbal Curio, Steinbeis Entrepreneur at the Steinbeis Transfer Center for Human-centered Artificial Intelligence
Artificial intelligence that focuses on people, despite thinking like a machine – is that utopian or an entirely possible scenario for the future development of AI? This was just one of the topics TRANSFER magazine discussed with Professor Dr.-Ing. Cristóbal Curio, an entrepreneur at the Steinbeis Transfer Center for Human-centered Artificial Intelligence and vice-dean of research at Reutlingen University.
Hello Professor Curio. You focus on human-centric artificial intelligence in your work. What exactly does that involve?
My work is driven by the concept of “learning from humans – learning for humans.” I think it’s precisely this interplay that gives rise to technical innovations for human beings, and what allows different sectors of industry to move forward. Ideally, this leads to insights into new methods, which can even be used to explore human cognition. It’s something that’s engrossed me since my time conducting fundamental research into human cognition at the Max Planck Institutes in Tübingen, Germany. The concept of human-centricity in artificial intelligence is nothing new, and can make an important contribution to Industry 4.0, or cyber-physical systems – or to be more precise: to any area in which human factors will continue to play an important role in the future. The challenge is to see human beings as a non-deterministic factor that behaves in very different ways.
Everyone is talking about autonomous systems at the moment – promising great things such as autonomous vehicles or systems based on a high degree of automation. By their very definition, autonomous systems don’t start out as human-centric. If you look at the regimes of rapid interactions, but also spaces in which people are found – such as urban living spaces – it makes you realize that these systems have plenty of potential to cause harm. So it’s all the more important for them to have a quasi-human understanding of people, so they don’t behave as selfish systems and assume they have right of way. Autonomous systems without a human-like understanding of human beings would be unacceptable when it comes to current transportation systems.
On the technical side, my focus lies in optimizing and harnessing machine learning algorithms to make sensor technologies smarter. To a large extent, humans provide me with a role model in this because of the way we optimize the interplay between our sensory perceptions during our entire “life cycles” – including our principles of learning. Even if we still haven’t yet understood all mechanisms of human perception, the way our senses interact does make it possible to work out some fascinating technical requirements for sensor systems in order to design them to think smartly.
That should also include social intelligence. And it results in some amazing new features. For example, this can be used to enhance sensory perceptions from a human perspective, and that improves system safety. Giving consideration to human characteristics is often an important factor for users when interacting with applications – for example, a system might provide visual support or a system can guide your attention in a car. “Human-centered” means considering everything, as part of the overall development system. Where it gets interesting is when you have to consider users beyond the norm. This requires AI-based technology that offers accessibility and is useful for everyone in an aging society.
Realizing AI-based systems is just as important to me as understanding their consequences. This is where being able to interact with new VR technologies provides us with a realistic experience of technology. At the same time, possessing the technology to simulate interactions lays a foundation for further innovation. It provides a basis for simulating critical situations – so-called corner cases – so you can safeguard smart systems. In lots of areas, it’s not possible to predict how much information will be required for a system. One way to get around this is to use complex data simulations that are as faithful as possible. The solutions in this area are getting more and more powerful all the time. A particular challenge is simulating sensors according to the depictions we have of people; there’s still no way to synthesize all the different facets of their behavior. But here, too, there are some exciting things coming out of research at the moment, and they’ll be suitable for transferring to a whole variety of applications.
In lots of the areas I’ve just mentioned, the research has all the right fundamentals in place, and that’s already led to a number of processes being used in a whole variety of sectors of industry.
You just mentioned machine learning, but also AI. In what ways do these two concepts differ?
Machine learning is seen as a subfield of artificial intelligence, and it’s currently moving forward extremely quickly. The language used at the moment to describe artificial intelligence often now includes a definition of machine learning. It involves learning techniques, especially when there’s only access to limited data, which are also referred to as observations. Machine learning also often extends previous intelligent search strategies, making it possible to solve complex optimization tasks.
What challenges do you believe the current AI developments pose for companies, particularly SMEs?
Many of the innovations in AI have an impact on methodologies. The methods are often developed by AI scientists conducting research by looking at specific benchmarks. Often the first step is to work out what level of abstraction you need in order to understand the relevance of a particular innovation for an application. Identifying the categories problems have to be solved under is usually a quick process for companies. In some cases, the solutions prove to be quite complex for companies at first; in others they don’t have access to the required expert knowledge. Also, empirical AI requires that you always have access to relevant data. The value of data often entails an understanding of specific events. But unfortunately, important events only occur infrequently, so having access to a large volume of data turns out to be much less valuable than was initially assumed.
Many industries are undergoing a transformation toward the use of AI. The automotive industry has still yet to prove that it’s in a position to introduce widespread autonomous driving, and it can’t be denied that AI plays a crucial role here; but there are some big markets yet to be uncovered, such as other kinds of assistance systems – like microelectronics used to automatically design analog circuitry. The picture that’s emerging is that combining the right kinds of simulations, expert knowledge, and data-driven AI methods such as reinforcement learning will play a major role in the future.
Once an initial solution to a problem has been found and is introduced, you have to ensure it keeps working and it’s maintained. There are still many things that need to be understood, especially with larger data models. Future service providers will be in a position to solve many of the technological challenges, but there’s still a lot that needs doing when it comes to new software and AI hardware. Whatever happens, there will be liability issues, especially in the healthcare industry, and they’re a major challenge when it comes to regulatory issues.
You also work on AI ecosystems. What potential does this area offer, especially for companies involved in manufacturing?
Complex innovations, especially at classic manufacturing companies, require increasingly complex alliances to allow contributions to be dovetailed. Particularly smaller manufacturing companies should establish a meticulous understanding of AI and identify the benefits it offers in a whole variety of areas. These range from automation to quality assurance aspects. We expect there to be demand for recurring solutions, and this will require the dovetailing of different providers. There’ll be demand for new AI ecosystems – AI provider networks, so to speak – that will optimally complement one another. Particularly for manufacturing companies, the hope is that being part of such AI ecosystems could result in new ideas and inspiration, also with regard to fundamental research into AI, and that new AI research benchmarks will emerge – which will result in breakthroughs for entire sectors of industry in the long term. One challenge will be agile competitors – who are skillful at using AI, integrate it early into their products, and thus produce faster and more efficiently, or even develop completely new business models.
Prof. Dr.-Ing. Cristóbal Curio (interviewee)
Steinbeis Transfer Center: Human-centered Artificial Intelligence (Tübingen)