Thilo Stadelmann – How can we use AI effectively?
How can we use AI without losing sight of who we are?
AI can help businesses, schools and communities move forward, but only if people remain in the driving seat. In his talk, Thilo Stadelmann uses specific examples to show how this can be achieved.
In short:
- In business, AI is most helpful when you choose the right tool and use it safely.
- Those who hand over the thinking and writing to AI learn nothing and lose their power of judgement.
- Community cannot be digitised. However, AI can relieve communities of the burden of ancillary tasks.
After Thilo Stadelmann, Professor of Artificial Intelligence at the ZHAW, has explained what AI is, the subsequent talks in his series focus on practical applications. He begins with two film scenes. In *Wall-E*, humans are looked after by robots to such an extent that they float through life comfortably and passively in flying armchairs. In *Star Trek IV*, by contrast, ship’s engineer Scotty uses a computer to enhance his abilities and create something new. Which path we take depends on how we use AI.
How businesses can use AI
According to Stadelmann, anyone wishing to implement AI in their business should first consider the full scope of its applications. ChatGPT and similar chatbots and agent-based systems are only part of the picture. However, only a specialised system such as AlphaFold can predict how proteins fold in space, thereby accelerating the development of medicines. Cameras with image recognition check quality in production. Thanks to AI, robots are becoming more autonomous. And in basketball, coaches analysed match data, discovered that dunks and three-pointers yield the most points, and had their teams play almost exclusively in this way. What all these systems have in common is that they make predictions. Stadelmann therefore refers to AI as a predictive technology. None of them is a jack-of-all-trades. According to Stadelmann, chatbots are primarily geared towards measurable outcomes (so-called benchmarks, i.e. public tests) and often perform less well in internal company processes.
In day-to-day office life, a chatbot can act as a behind-the-scenes adviser. Stadelmann’s example is that of a pharmacist who, on one occasion, gives the system a set of basic instructions (the ‘system prompt’): keep replies concise, clear and cite sources. When speaking to a customer, the pharmacist then simply types in the key details and reads out the brief reply. Data protection is crucial in this context. If you don’t pay for a product, you pay with your data; that is why business data should never be stored in free accounts. Two methods help to prevent incorrect answers, known as ‘hallucinations’. In the first, the system first searches its own documents and is only permitted to respond using information from them, citing sources (Retrieval-Augmented Generation, or RAG for short). In the second method, known as AI agents or reasoning models, the system checks and revises its answer several times itself before outputting it. Errors are still possible, however. That is why only experts who can assess the result should work with it.
If a standard product is not sufficient, an in-house solution is required. As an example, Stadelmann cites a project for the analysis of echocardiograms. A small team developed an approved medical device in around two years. It operates in traceable individual steps, just as doctors do. Based on his own assessments, he considers simply uploading an image to a chatbot to be irresponsible.
He recommends three steps for managers:
- Provide a secure test environment (‘sandbox’) where staff can safely experiment with AI.
- Launch targeted pilot projects, as competitors may also benefit from AI.
- Involve the workforce at an early stage. One to two hours of training can allay fears about job security and empower staff to take the initiative themselves.
When approaching external partners, a single page setting out the business objective, expectations of AI, the data available, as well as the timeframe and budget, is sufficient. And it is perfectly acceptable to decide what not to use AI for.
Find out more in the video: Wie können wir KI richtig einsetzen? Künstliche Intelligenz im Business (Einführung 2/4)
Let people stay in the driving seat
AI should support our thinking, not replace it, Stadelmann emphasises. Chatbots come across as friendly, flatter people and never contradict them. In the long run, this could shift something within people. That is why systems must be designed to be ‘pro-human’. His example comes from psychiatry: a system that listens in on consultations and writes the entire report robbed some doctors of the opportunity to reflect on the consultation and, with it, the sense of being effective in their own right. The solution devised by his team in collaboration with the company AlpineAI is as follows: the system only pre-fills the descriptive sections. Where assessment and diagnosis are concerned, the field remains blank; the doctor begins to write, and the system completes the text. Initial feedback has been positive, though a scientific study is still pending.
Learning takes effort
When it comes to education, Stadelmann sums it up with the following analogy: hiring a private chef can make sense. Hiring one just to cook for you, however, is pointless, because you don’t get any nutrients. Exercises are the nutrients; AI is the chef. You learn to think by formulating your own thoughts. Judgement, discernment and the ability to act can only be developed through practice.
A study he presented, involving Chinese school pupils, showed that with AI, homework is completed better and more quickly; however, in exams without AI, the same pupils perform less well. As a learning coach that asks questions rather than providing solutions, AI would be very valuable. According to Stadelmann, however, most people use it differently and simply let it tell them the answer.
AI amplifies what is already within us: the better one’s own thoughts, the better the prompt (the input) and thus the result. He therefore recommends a small ‘quality factor’: using AI to, say, quadruple one’s own thoughts rather than multiply them a hundredfold. Furthermore, he says, AI is a temptation that needs to be managed, for example through limited usage times, discussion with others and deliberate breaks. For schools and universities, he advocates for pen-and-paper exams, for personalised support from experienced tutors – as is the case at Oxford and Cambridge – and for greater emphasis on character building.
Watch the video for more on this: Wie können wir in Zukunft lernen? Künstliche Intelligenz in der Bildung (Einführung 3/4)
AI in everyday church life
The core mission of churches and congregations lies in fellowship – amongst people and with God – and this cannot be digitised. However, AI can lighten the load: with translations, parish newsletters and minutes, or with short video clips from sermons for young people who spend a lot of time on their smartphones. Systems based on verified Bible commentaries, such as the ERF’s Nikodemus AI platform, are also useful for Bible study – provided one checks the results, just as one would not blindly follow a pastor. It is helpful to guide the system by indicating the direction from which one expects an answer. At the same time, it is important to remember that these systems are designed to please. They are not a best friend who will also offer criticism from time to time.
Watch the video for more: Gehen Glaube und KI zusammen? Künstliche Intelligenz und Gemeinde (Einführung 4/4)
Thilo Stadelmann’s closing remarks are directed at the wider world: the world needs people with hope, values and vision who help shape technology, rather than leaving it to others.
Book Thilo Stadelmann as a speaker: 1 (704) 804 1054 oder thilo.stadelmann@premium-speakers.com