I don't yet see AI replacing humans

Mathematician Matthias Rottmann on the future of humanoid robots

AI, or artificial intelligence, is on everyone's lips, delighting some and worrying others. Development in all areas is gathering pace, but many people are no longer able to keep up. In Cornwall, for example, they are currently building the most human robot. The device is called Ameca, and it looks frighteningly human because it decides independently what it responds to. At the University of Wuppertal, Matthias Rottmann, Professor of Practical Computer Science and Digitisation in the Applied and Computational Mathematics working group, who was recently appointed to Osnabrück, is still working on ongoing projects until 2028 and says: "I can understand that it scares you at first, because you immediately ask yourself whether these decisions are safe or not. However, we are already using it in many cases through Chat GPT. I don't see AI replacing humans yet, but it can support us well, just as it supports me in my everyday life. I always have those aha moments with Chat GPT where I think, let's see if Chat GPT can do this or that for me because I don't feel like it. For example, I wanted to quickly create a graphic from a table with figures, build a plot that I could also modify if necessary. To do this, I simply pasted the screenshot to Chat GPT and said: "Write me a Python code (a Python code is a series of instructions written in the Python programming language to create software or automate a task, editor's note) so that I get a bar chart for this section of the table", and bang, it was done. I had saved at least a quarter of an hour." Rottmann works with artificial intelligence with regard to vision. "I use so-called deep neural networks, i.e. AI models, to analyse images. This is used to model the visual cortex (the visual cortex, or visual cortex, is the part of the cerebral cortex in the occipital lobe of the brain that is responsible for perceiving and processing visual information, editor's note). We want to analyse images, and if you look at what image analysis can do, there is already a lot possible."

A deer in the city centre

Rottmann's research is dedicated to recognising the environment in street scenes. Based on AI, this already works with high accuracy in street scenes with good weather, unless the AI is confronted with an object that it cannot categorise. Rottmann explains: "If an object appears that we have never seen before, or if it appears in a context that we are not familiar with, such as a deer in the city centre, this can lead to a misclassification." There are still a lot of diseases that humans would recognise immediately, such as the deer in the city centre. In contrast to humans, AI has to be fed with billions of data points. However, once this data has been entered, AI can work like an expert. This can also be seen in the medical field when it comes to processing medical data. Doctors are also subject to a high degree of uncertainty. "AI is getting better and better, it supports us a lot and can make a lot of work easier or give us the opportunity to concentrate on the difficult cases while the simple cases are treated with AI."

Perfect humanoid robots cause unease

A world in which humanoid robots make our lives easier is not considered desirable by everyone. The more perfect the artificial friend, the more discomfort it causes. But how far can we go in development? "That's a research question," laughs the mathematician, "we should find out step by step in studies. In the interests of people, you should start with something that you can clearly identify as a robot and then take a small step forward from time to time. I believe that the people who develop this have a strong responsibility." There need to be studies with test subjects and then the robots can be made better and more realistic over time. But Rottmann also asks: "Ultimately, the question is: is it necessary? Is it desirable for a robot to become completely human? You can still give it a lot of abilities, but it doesn't have to appear human. Development is going so fast and most people don't realise what's behind it. We need to do a lot of work on this, because science is constantly advancing and the gap is widening."
The Scottish model Ameca mentioned at the beginning even advertises the programming of emotions. But these feelings are also merely imitations. "We humans have real emotions," Rottmann clarifies, "but a robot imitates. An AI can imitate and this is called imitation learning. It then imitates the symptoms of the emotions."

Object recognition with digital cameras

Rottmann's research uses ordinary images from digital cameras, which show the robot's surroundings, for example, and then feed the software with information. "A digital camera initially summarises this as a huge box of zeros and ones, or numbers in general. Each pixel contains a number or even three numbers for red, green and blue and the AI then searches for patterns and recognises structures," explains the expert. This works in the same way that recognising a dog or a cat, for example, is data-driven. "You enter lots of examples of dogs and cats and the neural network learns to recognise these patterns using a kind of optimisation process. AI recognises that patterns belong together. So if I show a cat from ten thousand positions, I have a huge pool of data and then it starts to generalise."

Do robots learn from each other?

In America, there is the example of Iris, a Chinese robot from the company Unitree, which is already in use and is considered to be the most advanced of its kind. It can pick things up independently and is used in customer service. At the moment, it is still controlled by humans. However, the robots are supposed to learn from each other. "At the moment, it's not the case that a robot learns and can apply this directly," says Rottmann, "but what we typically do is work in a learning phase, which takes place in a supervised form. There are always researchers involved who monitor what has been learnt. I experienced this in a robotics lab, where doctoral students tried to teach a robot to pour tea into a cup. This was then learnt through imitation learning and the scientists sit at it for a very long time. In the end, the robot can reproduce it." However, the function is no longer changed, so it does not relearn. "But while I'm doing that, I have sensors and cameras and can record new data," says the scientist. In turn, all other researchers could do the same, so that this individually generated information could later be exchanged and used with a large number of robots. "Then the result has to be tested again, because otherwise it could happen that you have learnt in the wrong direction, the robot loses its abilities and then spills the tea off the edge of the table."

Robots are becoming part of our society

The American company "Open mind" is trying to create robots that are as independent as possible. The company believes that this will give people highly competent helpers. "You still have to implement a lot of safety measures for direct collaboration with humans," Rottmann qualifies, saying: "For example, it must not exert too much force when working with a human and there must be emergency switches. You have to regulate a lot of things and then we might end up working with robots. Robots could certainly become part of society at some point, there are great hopes for this. The first thing I can see is them replacing us in factories, where it is also very strenuous and dangerous for us humans. Then people can supervise better and perform tasks that are difficult to automate with more dedication. Robots can make our work easier. These are the great potentials. Step by step, robots will come into many different areas. And I trust that companies will do this responsibly." Rottmann sees a decline in expertise in many sectors, with a large proportion of society set to retire in the near future and many SMEs lamenting the loss of expertise. We are almost forced to support them with AI solutions. This is more about software than pure robots, which can be used to train new people again. The researcher can also imagine artificial helpers in the private sector, some of which already exist. "I can well imagine having a robot at home at some point that cleans, cooks and does everything I don't feel like doing. We already have robotic hoovers and robotic lawn mowers. And if there was a humanoid robot that could also do the things that we do manually, then that would be a pleasant solution."

Humanoid robots - a blessing and a curse

So what could the future look like with humanoid robots?
"Giving AI full power and autonomy without emergency stop buttons or anything like that, or allowing it to override the will of the person it is interacting with, that would be totally difficult. There has to be a limit. I don't know how "intelligent" we'll be able to make this AI at some point, but we have to be careful, even if we're still a long way from that." The work on this is becoming more and more detailed. Information must be prepared in a hacker-proof way, otherwise it will be dangerous, Rottmann states. "It's both a blessing and a curse, and it still means a lot of work. The technology is there, it's getting better and better and I don't think you can pull back. The interests and opportunities are too great."

Uwe Blass

Prof Dr Matthias Rottmann was a junior research group leader for reliable and efficient artificial intelligence in the Applied and Computational Mathematics working group in Wuppertal. In August 2025, he was appointed Professor of Practical Computer Science and Digitisation at Osnabrück University. His project work at the University of Wuppertal runs until 2028.