Critical Reflection

Our podcast for Wissenschaft kurz erklärt was created using Notebook LM, an AI-driven tool that transforms text into audio content. Using AI for this podcast, rather than recording it ourselves, served a dual purpose: it allowed us to engage with AI not only as an object of inquiry (examining AI itself) but also simultaneously as a medium of inquiry (demonstrating its role in content creation). By combining these two layers, we were able to explore both the capabilities of AI and the ethical implications of using it to convey complex ideas—an approach that would not have been possible with a traditional, human-produced podcast. We specifically chose Notebook LM because its results are nearly indistinguishable from human-produced podcasts. This quality emphasises how closely AI now approximates human-like content creation, underscoring its potential as a pressing topic for ethical considerations in AI.

To produce the podcast, Maximilian Kiener wrote a 1,900-word draft focused on ethics by design and the key content the Institute wanted to communicate. We uploaded this draft into the AI multiple times, prompting it iteratively and selecting the best segments. Using audio editing software, we then rearranged and refined these segments to create a cohesive final product. The original source text is included in this space for reference.

Throughout production, we faced several challenges, as the AI occasionally introduced errors—such as conflating ethical terms (e.g., confusing transparency with explainability), substituting terms incorrectly, or presenting misleading examples. These issues required careful editing and extensive curation to maintain accuracy and integrity. But ultimately, our thorough oversight allowed us to stand by the final content, ensuring its accuracy and alignment with our intentions.

Our experience working with AI-generated podcasts has raised several questions, prompting us to reflect on the role of AI in content creation. Although all original content originated with us,  should we consider AI a co-creator, or merely a tool? The app introduced elements such as an enthusiastic tone, humour, and occasional emotional inflections, along with some examples that we had not initially included. How should we assess these additions? And what about key themes that the AI overlooked, such as ‘Technology for Humanity’, a central topic for TUHH, which were excluded in favour of others? Did the AI put a spin on our intended message?

From there, several broader reflections emerged as well. While AI can enhance accessibility, it requires careful oversight. Do the AI-generated enhancements make the content more engaging, or do they risk trivialising or distorting complex topics? Here, the conversational tone in AI-generated content merits consideration. Often, an American conversational style is embedded in the AI’s responses. How might this style affect our audience’s perception and interpretation? And what are the implications when podcasts are created by AI systems that lack (communicative) intentions and the ability to take responsibility for what the produce? Is this a different form of communication, or something else altogether?

We also noticed that AI tends to smooth out differences of opinion, making hosts consistently reinforce each other’s perspectives—sometimes in a superficial manner. So, does AI inherently reduce dialectic and diversity of opinion, which are central to meaningful human discourse and, arguably, vital for democratic engagement?

Finally, we reflected on possible gender dynamics within the AI-generated content. Does the AI assign equal credibility to male and female voices, or does it subtly position one as validating the other’s ideas? What is the risk of one voice being relegated to confirming another rather than introducing new perspectives on its own?

Ultimately, we hope our listeners and readers not only enjoyed the podcast and gained insights into our work on ethics by design but also found value in reflecting on AI’s role in content creation, additionally supported by the possibility of comparing the podcast with our own source text. We believe that this dual approach—AI as both the object and medium of our inquiry—will become increasingly prevalent, shaping future conversations about technology and its ethical dimensions.