Deutsche Telekom has several hundred AI systems in use. This text presents a small selection. The activities are coordinated by the AI Competence Center (AICC), which is part of the Board of Management for Technology and Innovation. T-Systems' AI Factory focuses primarily on products for B2B customers and also is intended to support the Group's AI projects with the right skills and resources.
AI in Customer Service
Frag Magenta (Ask Magenta)
Frag Magenta - Deutsche Telekom's digital assistant - serves as a central gateway for our customers' service requests. It can be reached via www.telekom.de, the MeinMagenta app and, in some cases, via the service hotline. Frag Magenta helps 24/7 and without waiting.
In customer service, Frag Magenta has been used as a chatbot since autumn 2016, initially only for selected topics. Since 2018, it has been expanded to include artificial intelligence based on IBM Watson. He now learns automatically and can process customer inquiries even better and more competently. In the meantime, Frag Magenta uses Rasa's "Natural Language Understanding". This allows our customers to address their concerns to Frag Magenta in natural language in their own words. This means that he can help with over 380 different concerns – from relocation, new offers, questions about invoices or appliances and also in the event of malfunctions. Here, for example, it automatically measures the lines and, if necessary, restarts them.
Since 2020, Frag Magenta has also been connecting to the service hotline by voice. It takes over callers who previously reported a landline fault in the voice menu. In other words, the digital assistant understands people's natural language even on the phone. For example, in the case of account data, the "fifty-five" as well as "five, five" or "four times the zero", instead of "zero, zero, zero, zero" etc. With T-Duplex, an in-house development, it is possible to understand different ways of speaking and expressing customers.
Since autumn 2023, " Frag Magenta Voice" has been available with Rasa's voice skills. The previous language portal understands simple words and uses a predefined menu to select topics such as: "Then please state your request now so that we can help you in a targeted manner. If something doesn't work technically, please say 'malfunction' or give another reason for your call, for example 'router configuration' or 'move'." Now it is experiencing a quantum leap in terms of customer experience and efficiency: Customers are no longer asked to present their concerns in several stages in keywords, but can - as in chat - formulate them freely in natural language. For the time being, the intelligent new voice portal is available on the mobile phone hotline.
In 2022, Frag Magenta conducted over four million customer dialogues. This frees up our employees time to deal with more complex issues.
The chatbot solves more than a third of the concerns immediately. In the other cases, and also at the customer's request, human advisors take over and continue the dialogue almost seamlessly. Incidentally, ComputerBILD considers it to be the best service bot on the German market: "Deutsche Telekom's test winner Frag Magenta convinces with its enormous and readily available expertise."
Almost one million customer inquiries reach our service every year by mail. That's more than 4,000 letters that need to be opened on working days and processed as quickly as possible. The colleagues are supported in this by an artificial intelligence: Sherloq. This evaluates the documents electronically and forwards them to the right place. The letters are opened and scanned and then evaluated. The AI already reliably recognizes 80 percent of all customer requests and is constantly learning. The automated detection rate is increasing, and the time-consuming, manual post-processing is becoming less and less. Thanks to the intelligent mix of man and machine, customer requests can be processed more quickly. In the next step, Sherloq will also automatically forward e-mails from Deutsche Telekom mailboxes to the responsible contact persons and thus contribute to a quick response.
AI for our networks
T-Car: AI in fiber-optic expansion planning
Deutsche Telekom uses AI in the planning of fiber-optic routes. Deutsche Telekom's measuring vehicles with various sensors and laser scanning technology are driving along the paths where fiber-optic cables will soon be laid. They record the road situation in detail. For example, the recording of surfaces, houses and trees in the surrounding area plays an important role. The AI then uses the collected data to create potential routes for laying the fiber optics. A process that previously required numerous on-site appointments. Network planning can do a lot with these AI plans. This is because the majority of the costs of fibre-optic expansion are attributable to civil engineering. Therefore, detailed information is worth its weight in gold for cost-effective route planning. The planner can also use the information from the AI to check which alternatives to complex civil engineering can be implemented, e.g. shallow installation methods. With this method, not every road has to be torn up with an excavator, which can significantly speed up the expansion. And last but not least, it is also possible to estimate the costs incurred when laying a fiber optic cable.
AI for enterprise customers
Preventing Blood Poisoning with Artificial Intelligence
The AI-powered solution from Telekom and TCC Analytics reduces the risk of blood poisoning (sepsis) in intensive care units. Using a self-learning algorithm, it analyzes patients' vital signs in real time, predicts the individual sepsis risk up to ten hours in advance, and alerts medical staff via a user-friendly dashboard. This allows physicians to intervene early, improves the chances of survival, and relieves them of manual processes so they can focus on patient care.
Welding robots: AI solution from a single source for production and logistics
The research project with a leading car manufacturer demonstrated how a laser welding robot can improve its behavior with the help of AI. Traditionally, the robot receives a fixed set of parameters, but our solution allows it to adapt to different requirements on its own.
Our AI architecture detects and evaluates welds and then optimizes the robot's laser welding process. Not only does this innovation make the process more efficient, but it also allows robots to learn and adapt in changing environments.
This autonomous functionality not only improves the quality and cost-efficiency of the welding process, but also makes the factory more agile and productive. Our project revolutionizes the performance of industrial robots through AI and at the same time strengthens cooperation with our partners.
The "AI Engineer" - Programming with Artificial Intelligence
The AI solution "AI Engineer" from T-Systems programs and translates old programming languages into new ones. For example, 40-year-old FAME programs from the financial industry into current Python code. It also automatically creates documentation and tests parts of the program. The "AI Engineer" is based on three different large language models that have different strengths and weaknesses. It has been used in daily work at T-Systems since 2023.
AI in Cyber Defense
Cybercrime is becoming more and more professional and uses automated attack methods. At the same time, increasingly "smart" devices are creating more and more attack targets: Companies today are moving data and applications to one or more clouds. Since the pandemic, millions of people have been working from home in the corporate cloud every day. Examples include video conferencing or collaborating on documents. Sensors in the Internet of Things (IoT) or wearables are added as targets for hackers. In search of gaps in the corporate network, attackers are now using artificial intelligence on a large scale.
Deutsche Telekom also makes use of this principle and uses AI platforms to find vulnerabilities in itself or in customers. Honeypot sensors, on the other hand, provide data that allows conclusions to be drawn about attackers' strategies and tools. These are built into the Group's network. In addition, Deutsche Telekom can access data from more than 6,000 other sensors around the globe. In total, they register more than 100 million attacks a day at their peak. In addition, Deutsche Telekom's cyber defense team analyzes around one billion pieces of security-relevant data every day for signs of cyber attacks. Of course, a human being can no longer do this manually.
Computers are much better at recognizing patterns and very quickly identifying deviations that indicate attacks. This is a valuable support for analysts in cyber defense. Deutsche Telekom developed and trained the AI itself, in cooperation with Telekom Innovation Laboratories in Israel. For several years, Deutsche Telekom has been using AI to protect the security of its own infrastructure and thus also the security of its customers.
More than two dozen masterminds developed by Deutsche Telekom are working on the defense. Artificial intelligence keeps a constant eye on the network: it sounds an alarm if a line is faulty or breaks. AI helps analyze attack patterns in real time and advises on countermeasures. Later, she also writes reports about it.
AI and Ethics
With the nine self-binding guidelines, Deutsche Telekom was one of the first companies to define a framework for the use of AI. And that was already in 2018.
The guidelines describe how we at Deutsche Telekom want to deal with AI and how we develop our AI-based products and services.
The basic idea is that AI is only a tool and is neutral in itself. So it's up to us to use them in a positive way, without ignoring the risks and dealing with them responsibly.
These guidelines stipulate, for example, that it is clearly defined from the outset who is responsible for which AI system and which AI function. We are also committed to transparency. In this way, we disclose how we use customer data and make it clear to our customers when they communicate with an AI system. And we remain in control and are able to pause or shut down our AI systems at any time.
These are just a few examples. You can read all nine guidelines here.
In addition, a guideline helps to align AI-supported business models in compliance with data protection regulations. Employees can use this as a guide as early as the business or product idea phase and gain confidence in their actions. For example, the guide shows how data should be processed in principle, warns against automated case-by-case decisions and suggests appropriate protective measures.
AI and arts
Completion of Beethoven's 10th Symphony
Unfortunately, Ludwig van Beethoven was never able to complete his 10th Symphony. At the time of his death, only a few sketches were finished. On the occasion of the composer's 250th birthday in 2020, Deutsche Telekom dared to experiment. Together with international music and AI experts as well as scientists from the Beethoven-Haus in Bonn, Deutsche Telekom is developing an artificial intelligence that has continued the symphony in interaction with the experts. The world premiere took place on October 9, 2021.
In November 2022, the project continued: "Angels" by Robbie Williams was supplemented with compositions in the style of Beethoven with the help of AI.
For the benefit of society
AI-powered wildfire detection
Wildfires can have devastating consequences and threaten lives and property. And they are a growing problem, because they burn almost twice as much forest area today as they did 20 years ago. T-Mobile US, Pano AI, and Portland General Electric are deploying 5G-connected cameras for AI-powered active wildfire detection. This detects the first signs of forest fires. This allows emergency services to react before the fires get out of control.
Early detection is key to stopping a wildfire before it spreads. But with traditional methods, identifying and evaluating a fire can take hours and delay response. With T-Mobile US's 5G network, Pano AI can collect and analyze massive amounts of data in near real-time to detect new foci in the critical first minutes of a wildfire. And hopefully contain the fire before it endangers lives and property.
Generative AI for Deutsche Telekom employees
At the beginning of 2023, Deutsche Telekom and the Group Works Council adopted the AI Manifesto. It serves as a framework for the processing of employee data in AI systems. It ensures that employees' interests and their data are protected. It also regulates which standards apply to the introduction of AI systems. And it stipulates that all decisions will continue to be made by humans and not by machines.
AI systems should be useful for the interests of employees and the value creation of the company. The manifesto is intended to promote trust in the use of such systems and to reduce reservations about the use of AI by creating conditions for the AI systems at Deutsche Telekom that are jointly supported by the operating parties and by increasing transparency and comprehensibility.
AI systems must not be used to collect information about workers regarding their political opinions, ideological beliefs, trade union membership or sexual orientation. Nor may they be used to monitor behaviour, control performance or influence.
The AI Manifesto develops initial principles that apply to the introduction of AI systems as well as to systems that have already been introduced, according to which a standardised operational regulatory framework for such systems is to be developed. It is based on national and international (EU) legal regulations and (technical) standards.
The manifesto pursues the goal of making it easier to work with such systems and saving time where employees are concerned by providing a uniform regulatory framework for the large number of applications to be expected. In the future, this will enable the design work to focus on risk-relevant systems and to agree on an adequate participation process for systems with different risk relevance. The joint efforts are aimed at speeding up coordination processes, avoiding duplication and misinterpretation, mitigating risks and making work easier.
For the use of ChatGPT Back in February 2023, we gave our employees a framework for using the free trial versions of ChatGPT. This permits use under certain conditions. In particular, only texts that are freely available (data protection class 0 or information protection class "OPEN") may be entered into ChatGPT – i.e. neither internal nor personal data!
Most importantly, the use of ChatGPT does not release employees from the responsibility to check every answer, to judge for themselves and to act transparently. It requires conscious use and the knowledge that technology must serve people in a positive sense.