Artificial intelligence

Interview – Michael Baeriswyl

“Man and machine have to work together”


Michael Baeriswyl is responsible for the development of artificial intelligence solutions at Swisscom and he regards it as presenting major opportunities for the economy but also as posing a few risks.


Text: Hansjörg Honegger,




You develop artificial intelligence solutions for Swisscom. Let's try to explain this term. What is AI?


It's a very broad term. Let's put it this way: Thanks to AI, computers are now able to perform tasks that they couldn't manage before as they weren't sufficiently intelligent to do so.


So machines are now doing what humans used to do?


It's not quite as clear cut as that. A lot more data is created these days. That's why we need intelligent systems that can sift through the enormous quantities of data and tell humans what is important and what is not.


Could you give us a practical example?


Swisscom call centre employees have been supported by artificial intelligence for several months now. AI understands the customer's problem description. It analyses what Swisscom services the customer uses and what similar problems other customers have had. The program uses this analysis and can suggest a solution to the agent very quickly.


Does AI recognise the content of the spoken word or does the agent have to enter the problem?


Switzerland benefits from very good data protection laws and for this reason the calls are not recorded. The agent uses keywords to note the problem. In theory, voice recognition would not be a problem. However, Swisscom has made a conscious decision not to use it.




Where is Switzerland compared with other countries with regard to the research and implementation of AI solutions?


A comparison is difficult. But we have a huge advantage over the giant US concerns: We have real corporate customers who we can ask about what they need.


What are the main requirements of the companies?


The companies first have to understand that data previously regarded as rubbish can be used today to improve business.


Which branches are the most dynamic?


There is a great pressure on prices in the Swiss machine building industry. Every efficiency improvement measure is worth a great deal. The entire business model of the financial branch is currently changing and there is a great dynamism here too.


Can a branch like the machine building industry afford the pending investments at all?


This is a major challenge. We are at the start of the development phase, experts are thin on the ground – with a price tag to match – and investment costs are massive. We will have to wait and see.




How are things looking with machines capable of taking decisions that also have moral implications?


That is a very important question that us specialists are confronted with every day. The fact is: if you apply specialist software to a different task, it fails. It can drive a car, but doesn't know how to fry an egg.


Adaptation is not yet possible?


No, software learns and trains for a particular use case. It can't do anything more. Intelligence is still a very relative matter when it comes to machines.


But this is the direction in which development is going.


Yes, of course. We need to be talking about this today to ensure that negative developments are kept in check.

Dr. Michael Baeriswyl

Michael Baeriswyl has been Head of the Artificial Intelligence & Machine Learning Group at Swisscom since mid-2016 and, together with his team, is responsible for developing specific AI solutions. He completed his doctorate at the ETH Zurich and at MIT in Boston.


His profile can be found here


«For us very rigid data protection is central.»



What is the answer?


For us – our team as well as Swisscom as a company – very rigid data protection is central. This is the only way to avoid misuse.


Another negative impact is that up to 30% of office workers could lose their jobs due to artificial intelligence. You can discuss this all you want – companies will make use of these efficiency improvement measures due to the pressure of competition.


This view of things is too simple. You should not underestimate the role of humans. Handling customers takes a lot of human empathy and understanding. The machine has to find out what is really important in this enormous mountain of data we produce. Customer contact cannot be left to machines.


Your daughter was born last year. Aren't you not afraid of how the future will shape the world she is growing up in?


No. I think that the world will change. Certain things will be much simpler than others, but there will be new cultural problems. But that was always the case. My great grandfather was a wainwright. He built wagon wheels. There's no need for this today. So that's why I have no fears.


OK, but your grandfather was worried about the future.


Yes, of course. Our task will be to position the new technology with employees. People who are afraid do not use new technology and so are left behind by developments. They should view machines as a sensible and welcome enhancement for their work. Machines will only be really useful if we work together.




That paints a very rosy picture. The fact is that the use of AI seems rather sinister to normal users. You don't really know what's actually going on.


True. Many companies have no interest in transparency, mostly in the interests of their profits. I don't think that this is the right way. The most important principle is that the decision of the machine must always be transparent. The user needs to understand them.


Swisscom is one of the leading companies in Switzerland when it comes to the use of AI. How transparent is the company in this regard?


Very transparent. Swisscom is partially state-owned and is subject to very strict data protection legislation. We have an ethics committee that examines these issues intensively and that comes to other conclusions as major US companies, for example.






What is AI?


Artificial intelligence is intended to enable computer systems to behave in an intelligent manner. Computer or algorithms should be able to solve problems independently and to learn from this and expand their skills. The methods of AI include knowledge-based systems, pattern recognition and neuronal networks.




More on the topic