How Explainable AI is revolutionizing trust in AI - Expert interview

This is exactly what Explainable AI (XAI) makes possible. Experts Julian Stoettinger and Johannes Mayer provide exciting insights into the potential and challenges of this technology.

The role of data science in automation

When asked how data science supports complex processes at 3E, Julian answers:

We try to automate and scale processes as much as possible. This starts with data procurement from various sources, analysis and classification, right through to integration into our customers' systems. With the help of AI, we can make decisions such as which dangerous goods stickers need to be placed on a truck or whether special regulations need to be observed. These decisions carry a high level of responsibility, and someone has to put their hand in the fire that they are correct.

Both speakers have worked intensively with Explainable AI - here are their insights and experiences.

The path to Explainable AI

How did you discover your passion for Explainable AI and machine learning?

Julian: I was always a visual and graphic type and originally wanted to go into computer graphics. But the environment didn't suit me. Then I came to the Institute of Automation Technology by chance, where oddball professors were doing graph theory. I liked that and found my passion for machine learning.

Explainable AI is often seen as the key to trust in AI systems. But how does it create this trust? Julian draws an interesting analogy:

"If the electrics in your house are causing problems, you call an electrician. You expect him to solve the problem without having to understand every technical detail yourself. At the same time, you want to be sure that everything is correct and safe - his certification and liability ensure that. It's the same with AI systems: We don't need to understand all the details, but we do need explanations and accountability."

Liability issues and Explainable AI

A central aspect of Explainable AI is the clarification of liability issues in AI decisions. Julian explains:

"In safety-critical areas such as the classification of dangerous goods, we have a lot of responsibility when it comes to our decisions. It's not just about whether a truck has the right sticker - the focus is on the safety of people and compliance with legal requirements. Explainable AI helps us to make such decisions transparent and comprehensible and to understand the cause in the event of errors."

He adds:

"If it comes to a question of liability, we must be able to explain exactly how the decision was made. It is not enough to simply say that the system has learned from hundreds of thousands of samples and that an error rate of one percent is acceptable. What we need are clear and comprehensible explanations.

Alpha-Go: Learning from AI

To illustrate the possibilities of Explainable AI, Julian refers to a well-known example:

"Do you remember AlphaGo? This highly complex system developed new strategies in the game of Go that mankind would not have discovered even in a thousand years. Go masters observed the AI's moves, were fascinated and drew completely new insights from them. This shows that AI not only provides solutions, but can also expand our understanding - provided we can understand how the AI makes its decisions.

Practical applications of Explainable AI

Explainable AI is not just a theoretical concept, but demonstrates its strengths in practice. Whether in industry, banking or the energy sector - the technology helps to make complex decisions comprehensible and efficient. The following examples illustrate how Explainable AI is used in practice:

Dangerous goods identification for 3E

How do you use Explainable AI in your company and what added value does it offer in the classification of dangerous goods?

Julian: We use deep learning to make preliminary decisions in the classification of dangerous goods. Our system is able to explain exactly which component of a product has led to a certain classification. This saves our experts valuable time, as they can concentrate on the crucial information. Explainability not only strengthens confidence in the system, but also enables us to take responsibility for the decisions made. 

Explainable AI in the banking business

What role could Explainable AI play in the banking sector to make processes more transparent and trustworthy?

Julian: "When I wanted to take out a loan, the bank advisor couldn't explain to me clearly how the interest rate was calculated - an experience that I found extremely frustrating. With Explainable AI, the bank could have transparently shown which factors influenced my conditions. This would not only have provided clarity, but also significantly increased customer confidence."

Intelligent fault detection & maintenance planning in wind energy

After discussing the use of Explainable AI in the banking sector, another example from industry shows how this technology can promote efficiency and trust - particularly in the predictive maintenance of wind turbines. Johannes explains:

"We use AI to detect anomalies in wind turbines. With Explainable AI, we can specify exactly which vibration data indicates potential problems. This helps technicians on site to take targeted measures and increases confidence in the technology."

The ability to identify specific factors that led to a decision not only improves the efficiency, but also the acceptance of AI systems in the industrial environment.

Explainability, ethics and society

Julian and Johannes discuss the ethical challenges posed by AI systems. Julian expresses doubts about the feasibility of EU-wide certification for AI models, but sees the need for gradual regulation. At the same time, he emphasizes the importance of a culture of error that enables the responsible use of AI.

His conclusion on the future: "There is no scenario in which we can completely avoid AI. The technology is becoming more and more powerful and more people will use AI in their everyday lives. EU-wide certification for AI models would be desirable, but its implementation seems extremely complex. It is more likely that regulations will develop gradually - similar to a neural network that is iteratively improved by its loss function. Problem cases could serve as the basis for necessary restrictions.

Julian also emphasizes the importance of a culture of error:

"The introduction of AI often also means the introduction of an error culture - and a clear process for dealing with errors. [...]"

He adds:

"The paradox of being human is that we are more willing to accept human error than machine error. Machine errors often leave us helpless because we don't trust technology in the same way we trust a human"

The future of Explainable AI

Julian looks to the future:
The use of AI tools will continue to increase in both the business and private spheres. This will make it inevitable to clearly regulate the issue of liability for decisions made by computer systems."

Developments in AI remain exciting - especially how legal frameworks and ethical standards will adapt. Explainable AI plays a central role in this: it strengthens trust in AI systems and creates the basis for their responsible use.

But what do you think? Will we see stricter ethical and regulatory standards in the future? Share your opinions and discuss the future of artificial intelligence with us!