AI on Prescription
It takes over such tasks that bore or overwhelm humans: artificial intelligence (AI) has the potential to become the metronome of the SME sector if the UDE spin-off TamedAI has its way. By Birte Vierjahn
‘People in research are enthusiastic, committed and jump right in. In the tough business world, the first thing people ask is whether something is profitable. We inhabit the space in-between.’
These are the words of Ole Meyer, PhD student at the UDE Department of Software Engineering. For his PhD, he is focusing on the transfer of AI from the academic world into practice. Together with his fellow PhD student Nils Schwenzfeier, he founded the spin-off ‘TamedAI’ in 2019 — short for ‘Tamed Artificial Intelligence’. Meyer explains: ‘We want to bring radical research and SMEs together.’
In doing so, they are filling a gap in Germany: ‘We certainly have innovative companies in the country. But as far as the AI sector is concerned, the US has been in the driving seat in recent years, while we’ve been asleep. Germany has missed out on the whole cloud business, but there’s still much untapped potential in AI.’
The company offers four predetermined fields of activity: language technology, sports, banking/insurance, medicine – and a wild card, marked with a subtle rocket symbol: whatever the problem a client brings to the table, the two founders promise ‘highly specialised solutions that really make a difference’.
Algorithms with a feel for Language
Another focus of the young start-up is language technology: AI systems that read texts, understand their content and, based on this, create headlines or brief summaries – or that point out linguistic slip-ups to human editors. More information on: tamed.ai
Please add feet!
The requests are indeed very specific. For example, football clubs in both amateur and professional leagues want to optimise the movement of their teams: Is the wall optimally positioned for a free kick? Which foot was used to take the penalty kick; is there a better way? Human images have been created by computers for a long time, but there’s a catch: somewhere along the line, 18 rather random locations on the body were set. The lowest one stands for the ankle and there’s nothing below. ‘Ears aren’t of that much interest, but feet would be nice. So, we augmented, retrained and eventually created a football-specific skeleton that can answer the questions asked in the first place.’
AI can also support mental training for athletic challenges: in professional sport in particular, the mental aspect is crucial, be- cause athletes reach the limit of physical training at some point. Mental coaches are supposed to help, and one of their goals is to identify that one moment when an athlete starts losing. ‘Let’s stick to football. 1–0 isn’t a losing score in the 10th minute, but it is a losing score sometime before the 90th minute,’ Meyer elaborates. ‘Somewhere along the way, there’s the tipping point.’ It’s about emotion recognition, about minimal gestures – when the shoulders slump half a degree, for example. ‘These changes are too subtle for our eyes. By the time we see someone giving up, AI has been aware of it for a long time.’ Currently, they’re working with football and tennis pros who want to identify their personal tipping points and work on them.
Cough medicine for the third time? Not likely.
Insurance companies are also among the customers of the fledgling start-up. The main issue for them is fraud. There are drugs that have to be strictly controlled because they are addictive, as is the case with some cough suppressants. Or substances like steroids, which can be abused in bodybuilding – a prescription is quickly forged.
‘A major challenge is the infinite number of combinations of diseases that can exist,’ explains the computer scientist. A therapy should fit the medical record: a person who receives an HIV drug is likely to have the relevant diagnosis in their record. What’s more: men and women sometimes undergo different therapies, some diseases are more likely than others, medicines can have different names and still be based on the same active ingredient. The location of the pharmacy with respect to the place of residence also plays a role: someone who wants to obtain narcotics by fraud is more likely to take their fake prescription to the pharmacy three towns down the road than to the pharmacist around the corner.
‘This preliminary assessment by AI is called dark processing – gigantic statistical systems that build up huge probability distributions,’ says Meyer. ‘If something drops below a certain probability, we move into light processing, which means that the data has to be checked by a human.’
In spite of its many successes, the start-up doesn’t yet have its own premises, as the pandemic set other priorities. But the move to Essen is already in the pipeline. Still, even the company’s own AI can’t predict which project will be the first to come knocking.