Why AI is the Drum Machine
An instrument, not a tool
This past week during Cannes Lions, I had the honor of moderating a panel as part of the Salon Culture Conversations titled “Lights, Camera, Algorithm: How AI is Writing Hollywood's Next Golden Age”. On the panel was Doug Eck, Google DeepMind’s Senior Research Director who co-leads all generative media at the company (image, video, 3D, audio, and music generation) and Nik Kleverov, Foreign Native’s Chief Creative Officer and one of the leading AI filmmakers in the space.
We talked about how filmmaking is entering a golden age of creative possibilities, with new production workflows and new monetization opportunities. And watched some fun and engaging audio/visual AI samples. This conversation shed light on several fascinating themes but this one particular insight has been echoing in my head ever since:
“Technologies that get absorbed into the economy live. The ones that don’t? They fork off and die.” Doug Eck
The question for AI is not if it will be absorbed. It seems quite certain that the multi-trillion dollar investments have settled that. The question is how will it become a truly revolutionary platform that reshapes our economy from the bottom up?
The key to a true revolution hinges on one single design principle: programmability
Let’s wind back the clock so we can unpack this.
The Drum Machine
Eck framed the evolution of AI around this powerful metaphor: The Roland TR-808 Drum Machine.
In the late ’70s and early ’80s, the drum machine emerged, most iconically, the Roland TR-808. At the time, it wasn't a cultural force. It was a utility. In fact, it was literally a way to replace a drummer. A tool specifically for musicians creating demos and for use in studios to keep costs down.
Speed and cost efficiency, sound familiar?
But what made it powerful was its programmable step-sequencer. Artists could meticulously place each sound, the kick, the snare, the hi-hat—in relation to every other sound. They could tweak it, bend it, and amend it. They weren't just playing a rhythm; they were building one from its atomic parts.
The result? The birth of hip-hop, remix culture, and electronic music. The drum machine wasn’t just a tool, it became an instrument, it gave rise to new genres, subcultures, and commerce. And from there, it became an economy.
The Preset Rhythm Box
Now contrast the 808 with the preset rhythm section on a 1970s home organ. It was a panel of buttons: “Bossa Nova,” “Waltz,” “Samba.” Press one, and it played a canned, unchangeable beat.
It was a novelty. But you couldn’t bend it or deconstruct it. It was a closed loop. No one started a cultural movement with a Hammond organ’s "Foxtrot" button because there was nothing to build with. It faded into irrelevance, a kitschy memory rather than a living part of our cultural DNA.
Tool vs Instrument
I’ve been thinking a lot about the definition of a tool vs an instrument in the context of AI. Many of us in the space refer to AI as a “tool.” In fact, we often emphasize this point as a means to highlight the importance of human creativity: “AI is just a tool.” But let’s consider what a tool and an instrument actually are for: a tool is for a task and an instrument is for exploration.
A tool solves a known problem: a hammer is for a nail, a spellchecker is for an error. You use a tool to achieve a predictable outcome efficiently. An instrument, however, is a medium for expression. Its purpose isn't just to be correct, but to explore what’s possible.
A piano is a pre-set arrangement of keys but can be played with infinite variation. Its potential is defined not by its maker, but by the virtuosity and voice of the artist. This is why the best creative technologies, from the electric guitar to the film camera, ultimately succeeded by transcending their function as mere tools to become instruments—conduits for partnership, discovery, and the creation of things the world has never seen before.
The drum machine is an instrument, the preset rhythm box was a tool. As I’ve thought more deeply about these definitions, I believe AI in the context of creativity is more akin to an instrument vs a tool and there’s a technical rationale to prove this point.
The Transformer
The reason AI became an instrument and not just a tool is a revolutionary architecture called the transformer. (Yes, that’s the “T” in ChatGPT: Generative Pre-trained Transformer).
Eck explained that over the last 30 years, it was the transformer that was actually the tipping point for AI in 2017 because it fundamentally changed how machines understand and generate language, images, and sequences. Here’s a breakdown of why it mattered so much, in simple terms:
Before Transformers: AI Was Limited by Sequential Thinking
Prior to Transformers, most state-of-the-art models used something called RNNs (Recurrent Neural Networks) or LSTMs (Long Short-Term Memory). These models processed data one word at a time, in order, kind of like reading a sentence one word at a time without being able to look back or ahead easily.
As a result, they struggled with complexity, performance, and scalability.
But that all changed in 2017 with the research paper “Attention is All You Need” which introduced a revolutionary idea:
What if, instead of reading one word at a time, the model could look at all the words at once, and decide what to pay attention to?
Now, instead of “reading” a sentence from left to right, the model creates a relational map of how every word connects to every other word—regardless of position.
Why This Was a Tipping Point
Massive Performance Gains: Transformers outperformed RNNs and LSTMs on nearly every NLP task: translation, summarization, question answering, etc. They enabled models to handle much longer sequences (e.g. documents, code, even video).
Scalability: Because the model processes all tokens at once, training could be massively parallelized on GPUs. This allowed researchers to scale models by orders of magnitude — which led directly to large language models (LLMs) like GPT, BERT, and eventually ChatGPT.
Multimodal Expansion: The same transformer architecture could be adapted to other domains: images (Vision Transformers), audio, video, 3D, protein folding, etc. It enabled cross-domain models like text-to-image (DALL·E), text-to-video (Sora), and multimodal models (GPT-4, Gemini).
Programmability Through Natural Language: The transformer’s contextual attention mechanism made it programmable through natural language. Instead of needing hardcoded rules, you could prompt the model making it usable by non-technical creatives, writers, designers.
If AI is the new drum machine, the transformer is its programmable engine. But what does "programmable" even mean here?
How is AI "Programmable" Like a Drum Machine?
First, think about programming a drum machine. You have basic sounds (kick, snare) and a grid of steps representing time. "Programming" is the act of arranging those sounds on the grid. The magic comes from the relationship between the sounds. A kick on beat 1 feels different from a kick on every beat. You program the context.
The transformer architecture mirrors this exact process.
An AI breaks down all information including words, pixels, notes, into basic building blocks called tokens. This is its set of “sounds”. The heart of the transformer serves as a grid of relationships. When you give it a prompt, the attention mechanism looks at every single token and creates a complex map of how it relates to every other token.' And then, when you write a prompt, you are arranging tokens to create context. The AI is the instrument that intelligently builds upon the context you programmed.
The AI platforms that will win the race for dominance will maximize allowing users to fine-tune and prompt them in unexpected directions, enabling outputs from one model to become inputs for another, and will be easy enough for creators to adopt and for audiences to understand.
Most importantly, these platforms need to be economically generative. Not just tools to replicate old work, but platforms to build entirely new markets. They need to result in greater opportunities like marketplaces for fine-tuned models, new jobs like ‘AI personality director’ as this NYT piece references, and new services integrating tools into legacy systems.
Those Who Create Technology Lose Control of It. And That’s a Good Thing.
Just as Roland couldn’t predict that the TR-808's sequencer would fuel genres and global cultures in ways they never imagined, the companies and researchers building today’s AI tools won’t be the ones who dictate how the world will absorb them. When innovative technologies are programmable, the power always shifts to the creators who master the new instrument.
Ironically, Roland did not benefit much financially from the initial sales of the TR-808. It was a commercial flop that was discontinued after only about 12,000 units were sold. The companies that benefited most were the independent record labels like Def Jam that had the vision to embrace a new sound, and fashion and lifestyle brands that benefited from hits like Run-DMC's "My Adidas," a track built on an 808 beat, which led to an unprecedented endorsement deal with the sneaker company, cementing the link between music and fashion and giving birth to a new form of branded content.
This Future Is Already Starting to Take Shape.
AI is on a similar track and you can already see the signs:
A new wave of "spec ads" are achieving a level of quality once impossible without a major production studio. Solo creators are now producing cinematic quality commercials for major brands without permission—and defining their aesthetics in the process. A great example of this is Kelly Boesch’s Apple spec ads where she plays with the logo with fluid transitions having various Apple products transitioning into and out of the design in a calming, eye-catching, and mesmerizing ways. You could imagine this work on a big LED screen in Apple stores nationwide or perhaps as branded content in the Uber app. This isn't just a creator making an ad; it's a creator showing a trillion dollar company what the future of participatory marketing could look like.
Individual creators are becoming "full stack" studios. AI filmmakers like PJ Accetturo produced an extremely unhinged Super Bowl-esq ad with Google Veo 3 for a financial services company which aired as a broadcast spot during the NBA finals this month. Not only did he create the ad as a “full stack solo creator studio” in 2 days and for a very small fraction of a normal production budget, but he also built a character universe with supplemental (fake) AI BTS videos to extend the conversation across social. He promoted the main spot on his LinkedIn and other channels with captions like “I can’t believe Disney allowed us to run this AI ad during the NBA Finals 😂” and “This is GTA VI meets cocaine crazed Florida man. Here’s how I made it:” The ad became a story and a controversy, driving more conversation and engagement for the brand. He used AI as an instrument to bring social to broadcast and broadcast back to social as a new transmedium for brand storytelling.
Creators have developed extensive hybrid workflows and are even democratizing their own unique approach. AI/XR/VFX artist Paige Piskin moves fluidly between traditional tools like Procrate, Blender, and Adobe Photoshop and AI tools like Sora, Higgsfield, and even custom LoRas she’s fine tuned on Stable Diffusion. She has mastered a diverse technical skillset ranging from AI filmmaking, custom AR effects. She’s created hundreds of some of the most popular AR filters with over 300B impressions and over 1B shares and has now adopted AI as a medium to extend her capabilities as an artist. She brings to life animation, hyper realistic fantasy worlds, and a unique blend of immersive multi-media that feels like a new visual language of this era. Paige is not only an artist delivering these high concept outputs, but a sort of professor who democratizes her process, transparently breaking down her workflows for her community so they can try to create their own versions too. She doesn’t gatekeep her talent, she recognizes her own artistic brand as a platform for others to see themselves in her work and IP.
Brands who lean into this next-gen creator landscape empowered by AI— the spec ads, the full-stack creators, the ‘Brand as platform’ approaches will realize the massive commercial opportunities they will bring.
The Imagination Economy
The internet gave us the digital economy. AI is giving us the imagination economy. In this new world, the winners won't just be those who build the smartest AI, but those who build on top of it with the most audacity.
Just like with the 808, the real value isn't in having created the machine; it's in the culture, the genres, and the industries born from its performance. In fact, it’s not the output that’s the art, but the input that matters most. The intention, the iteration, the crafting and molding. The work you decide to align your name with. The artistry is defined by what we program, human will and virtuoso.
AI is an instrument just like the Drum Machine. And as of today, we can only imagine what’s to come.