🎥AI as Co-Creator – Fostering Reflective Use of AI and Building Foundational Skills in Music and Audio Analysis

Description

Prof. Dr. Meinard Müller, Chair of Semantic Audio Signal Processing at the International Audio Laboratories in Erlangen, integrates AI methods holistically into research and teaching. The focus is not merely on the application of AI tools, but on understanding the entire process chain: from musical applications and mathematical modeling to implementation, the training of neural networks, and the critical evaluation of their results.

Students learn not to use AI as a black box, but to understand it as a tool that opens up new possibilities in music and audio signal processing. At the same time, Prof. Müller raises awareness of the risks posed by flawed data, distorted annotations, and superficial use of AI.

Translated with DeepL.com (free version)

„We should neither glorify nor demonize AI. To me, AI is a tool—but a very powerful one!“ – Prof. Dr. Meinard Müller

Initial scenario

With the increasing prevalence of neural networks, music and audio signal processing has become more complex. Today’s students need skills across multiple disciplines: mathematics, signal processing, programming, data annotation, software and hardware infrastructure, as well as the ability to critically evaluate results.

The challenge: This broad spectrum of skills can easily be overwhelming. At the same time, the high performance of AI tools tempts users to solve problems too quickly using a “black box” approach, without fully understanding the task at hand, data quality, or model limitations.

Objectives

The central goals of Prof. Müller’s teaching concept are the development of in-depth competencies and the reflective application of AI methods:

• Students should understand the entire process chain: from application scenarios to mathematical modeling and implementation.
• AI is taught as a tool that has flaws, makes assumptions, and must be questioned.
• Data quality, annotations, and plausibility checks are established as core competencies.
• The teaching aims to empower students not only to use AI results but also to critically evaluate and explain them.

Concepts, Implementation, Methods

Prof. Müller’s teaching follows a clearly structured sequence of skills: First, a musical task is defined (e.g., beat tracking or tempo detection), then mathematically modeled, implemented, trained using annotated data, and experimentally evaluated. In the process, students learn that problems often stem not from implementation but from data errors, biased annotations, or probabilistic training effects. To ensure a deep understanding, Prof. Müller deliberately separates the acquisition of fundamentals (mathematical models, data comprehension, logical reasoning) from the practical use of modern AI tools. In some teaching situations, therefore, work is done without computers—in a circle of chairs, at the blackboard, through analog discussions—to clearly articulate core concepts. AI tools such as ChatGPT are not prohibited but are used context-sensitively: routine tasks may be performed with AI support, but Müller deliberately refrains from using them when building competencies (e.g., academic writing, algorithmic thinking).

„Many people use deep learning without fully understanding the problem at hand. That is exactly what I strive to address in my teaching.“ – Prof. Dr. Meinard Müller

Experiences

• Students value AI tools but often use them too early—before they fully understand the task at hand and the quality of the data.
• The most critical errors arise from insufficient or distorted data; AI reliably reproduces these distortions.
• Oral exams and in-person code reviews are particularly effective at revealing actual competencies.
• Consciously alternating between analog and digital teaching methods strengthens understanding of abstract concepts.
• AI-supported routine tasks reduce the workload but must not replace core learning processes.

„Within a few minutes of talking, I can tell what has actually been understood—and what was just churned out by an AI ghostwriter.“ – Prof. Dr. Meinard Müller

Criteria for success

• Students can explain AI methods rather than merely applying them—including mathematical foundations and data dependencies.
• AI results are always evaluated against one’s own expectations and through plausibility checks.
• Data annotation and preparation are recognized as core professional competencies.
• The connection between research and teaching is becoming closer: new developments are incorporated into teaching in a timely manner without neglecting the fundamentals.

“Only those who understand the problem statement and the data can properly evaluate AI results.“ – Prof. Dr. Meinard Müller

Back to overview