Música de IA: autoria, royalties e o futuro dos artistas

The episode opens with host Carlos Merigo introducing guest Felipe Vassão, a composer and music producer, to discuss AI-generated music and its impact on the music industry. The conversation begins by establishing context: streaming platforms like Spotify and Deezer receive tens of thousands of AI-generated tracks daily (Deezer reports nearly 90,000 daily), with AI uploads sometimes exceeding 50% of new uploads during peak periods. The hosts explore how AI entered music production historically, with Vassão explaining that before generative AI, there were already systems using machine learning for tasks like noise removal, tone balancing, and AI-assisted mastering tools like Landr (since 2012). These tools have been used for years without replacing human creativity. Vassão traces AI in music back to David Cope's 1970s-80s work with the Experiments in Musical Intelligence system, which could decompose and imitate classical composers' styles through data analysis. The conversation distinguishes between different types of AI: subtractive AI (that identifies and separates elements), analytical AI (that provides suggestions), and generative AI (that creates new content from scratch). The emergence of generative models around 2015 shifted the landscape, moving from MIDI-based systems to spectrogram-based models that can generate full audio. The hosts discuss the Suno controversy, where the platform was caught generating music that was nearly identical to copyrighted songs when given the song title and genre. In one example, "Low Bega Mambo No. 5" produced by the AI was virtually indistinguishable from the original. This led to legal pressure and Suno's agreement to train only on licensed material, resulting in a downgrade of their model's capabilities. Vassão emphasizes a critical distinction: he is "anti-preguiça" (anti-laziness), not anti-AI. He supports tools that expand creative possibilities or solve specific problems (like separating instrumental stems for learning), but opposes tools designed purely to automate and replace creative processes. The conversation explores why generative AI feels fundamentally different from previous technological tools. Unlike the sampler (which was repurposed from its original intent to create hip-hop), the electric guitar (which accidentally enabled rock and roll), or drum machines (which led to new genres), generative AI doesn't open new creative possibilities—it recombines existing patterns toward the average. Vassão argues that genuine creativity requires friction and struggle; the learning process through repeated failure (Ed Sheeran composed 240 songs to select 12 for an album) is essential to artistic development. The hosts discuss the Daft Punk narrative about process being central to music creation, noting that the actual experience of making music—studying, practicing, developing unique approaches—cannot be shortcuts. There's a deeper concern about cultural homogenization: when everyone uses the same AI trained on the same data, the output converges toward a bland middle, reducing diversity. The episode addresses the economic harm, particularly to entry-level musicians. Jingle composers and soundtrack producers for advertising—positions where musicians typically start—are being displaced by AI that can generate convincing content instantly. This eliminates the training ground where artists develop skills and build experience. The conversation highlights the distinction between using AI as an intermediary tool (like Moises' stem separation, which has 20,000 .edu email clients and 80 million active users) versus using it as a content generator. Tools like Moises actually enhance learning by allowing musicians to isolate and study individual instruments, similar to how slowing down recordings used to require equipment but now enables practice. However, the host notes that Moises itself could become a shortcut if users never progress to actually learning instruments. The hosts explore the paradox of remixes and covers: Street Fighter Latin Jazz Bolero versions exist and are genuinely appreciated as creative recontextualizations of existing music. This is different from Suno-generated content because human intention and artistic vision guided the recontextualization. The difference lies in whether there's a specific artistic perspective animating the work. A broader ethical issue emerges around training data: these AI systems were trained on billions of copyrighted songs without consent or compensation. While companies argue the mathematical vectors extracted during training contain no original data, Vassão counters that the training process itself was a violation—creators and rights holders paid for their material to be used. He notes that just because data gets transformed into vectors doesn't erase the appropriation. The conversation discusses regulatory responses: the European Union is pushing for transparency requirements, Deezer is implementing artist-centric payment (where your subscription money goes entirely to artists you listen to) and filtering out fully AI-generated content, while Spotify is introducing an AI Persona label to identify AI-created artists. However, these measures may not be sufficient. The hosts note that large play farms (bot networks that artificially inflate streams) have become an industry, with Brazil being one of the world's largest sources. This infrastructure allows people to generate thousands of songs through AI and use bots to create fake plays, generating revenue despite the music never being heard by real humans. The episode contemplates the future of musicianship: if creating music becomes as effortless as taking a phone photo, what does being a musician mean? The analogy extends that owning a professional camera doesn't make you a photographer—the eye and intention do. Similarly, generating a song on Suno doesn't make you a composer. The hosts discuss specific cases where AI music is emotionally engaging (like the Street Fighter Latin Jazz playlist) but acknowledge that once you learn it's AI-generated, the meaning changes for many people. This reflects a fundamental human need to connect with intentionality and human experience in art. Vassão emphasizes that art is about interpretation—the artist expresses their specific experience, and the audience connects with that uniqueness. An AI system trained on averages cannot capture this specificity; it can only recombine existing patterns. The conversation touches on entry barriers: while creating music is now more accessible than ever (Fruit Loops cracked software, Splice, affordable equipment), the AI trend suggests making it so easy that people never develop actual skill. The risk is a cultural implosion where nobody has the depth to create anything meaningful because everyone skipped the struggle that builds mastery. The hosts conclude by discussing what matters: not the technology itself, but the presence or absence of human intentionality, the specific perspective and vision that makes a work singular. They recommend several cultural works (film, theater, music) that exemplify intentional creativity, positioning human artistry as irreplaceable.