The Human Signal #48 – The Truth About AI Detection Technology - AIWTF? 3/11

This In this episode of The Human Signal (part three of my eleven part series, AIWFT?) I’m looking at AI detection technology and the role it’s beginning to play in the music industry. I talk through the rapid rise of AI-generated music on streaming platforms, and how detection tools are being developed in response—but not without serious limitations.I reflect on the reliability of these systems, the risks of misidentification, and what happens when decisions about access, visibility, and income are being shaped by technology that isn’t fully accurate. I also connect this to wider examples of AI detection failures, from education to facial recognition, and what that might mean as these systems become more embedded in everyday life.Throughout the episode, I’m thinking about what this means for artists trying to release work in a way that is both ethical and sustainable, and how fragile that process can become when control is mediated by platforms and systems being driven more and more by AI.Full Episode Description This eleven part series is looking at AI in music, where the digital music ecosystem is at right now, and what some of the most pressing issues with this topic are right now. I’m digging into it from the perspective of somebody who has spent the past two or three years working on a collection of 6 albums which I am hesitant to release under the current streaming model.I need to make serious decisions about how I want to release them, sooner rather than later, or they will be doomed to sit abandoned on a dusty hard drive forever. As a result, a lot of this investigation is about helping myself make more informed decisions - ones that feel ethically aligned, while still maintaining visibility and access to listeners. To be honest, at this moment in time, I’m not entirely sure that option exists!In the past two episodes, part one and part two, I discussed personal experiences shared online by musicians who have had royalties cut off due to AI-related interventions. I also spoke about Deezer and how they’ve been collecting data on AI-generated uploads. Their most recent statistics now show that 44% of all music being uploaded to their platform is AI-generated. This suggests we’re approaching a point where the majority of music on streaming platforms could be AI very, very soon.This is happening incredibly fast. The ease of producing AI music and the difficulty in distinguishing it from human-made work are accelerating its growth. This has serious implications for human artists for a number of reasons, but one of the most practical issues is that there is a limited royalty pool. Every AI-generated track that gets streamed is taking from that same pool and limiting the income of a human artist who’s work was most likely scraped without permission to train the AI in the first place.Deezer’s detection technology has allowed them to identify and label AI-generated music. This has helped them gather data and maintain some level of visibility for human artists. They’ve now developed this into a commercial product, selling their detection software to partners.There are thousands of AI detection tools out there, many focused on niche use cases. Deezer’s system is strong because it has been trained on a massive dataset. But this raises questions about incentives—allowing AI to flood the market also creates demand for tools to regulate it.The key question is how reliable this technology actually is. In text detection, studies have shown that none of the major tools exceed 80% accuracy. All of them produce false positives and false negatives. That means human work can be flagged as AI, and AI work can pass as human.There are also real-world consequences. I came across a case where a grandmother in Tennessee was jailed for nearly six months due to a facial recognition error linking her to a fraud case. This is where these technologies start to become deeply concerning.If these systems are being used to make decisions that affect people’s livelihoods, or even their freedom, then the margin for error becomes critical. I don’t trust it. I don’t think it should be allowed to operate without serious safeguards.In music, we’ve already seen examples where albums are removed from platforms due to suspected AI involvement. If an artist invests significant time and money into a release and then has it taken down, the consequences can be devastating. It can make it impossible to even consider investing in the next album if no costs can be recouped from the lastThere are multiple layers of erosion happening. Access to royalties is being reduced due to AI saturation. Music can be removed entirely, cutting off income streams. In some cases, artists’ work is cloned and re-uploaded, with someone else collecting the royalties, leaving the original artist powerless and burdened with the responsibility of trying to fix it themselves through platform systems that refuse to engage, using tactics like automation shielding and service hollowing, which make it even harder for artists to meaningfully challenge these situations.All of this ties into ongoing legal battles. Major labels have filed lawsuits against AI companies for using copyrighted material without consent to train their models. Independent artists are also being represented in class action cases. The outcomes of these cases are still uncertain, with AI companies claiming fair use.While all of this remains unresolved, new systems continue to emerge. This brings us to the next part of the episode, where I introduce AIMPRO, the first organisation set up to collect publishing royatlies for AI music, and I begin to ask: Are we witnessing the birth of AI rights? 🫣If you enjoyed this podcast please share it with a friend or consider leaving a comment or review, it really helps me out. You can also support me and my work by becoming a paid member, it’s €5 per month on Substack and Patreon. And lastly, you can still find me on YouTube / Instagram as @the_persona_project__ & @laurasheeran_ieThat’s all for now. Thanks for being here, and remember: Put humans first. Don’t feed the machines. This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit laurasheeran.substack.com/subscribe