GenAI is Foremost a Creative Tool
Concept Machines, Not Answer Machines
Starting April 2025, all full posts, including archived posts, will be available on my Substack, The Mediator.
Turn and face the strange
-David Bowie, Changes
For the average techno-curious Joe, making sense of GenAI is almost impossible. It is highly technical. The pace of innovation — new research, startups, use cases and products — is relentless. Using it doesn’t clear up much. Sometimes, it feels like magic, and others, it’s a waste of time.
Most confusing, even AI experts can’t agree on some of the most fundamental questions, like whether:
- AI valuations are in a “bubble;”
- the ongoing development of large language models (LLMs) puts us on a path to artificial general intelligence (AGI) or LLMs are just an “off ramp,” with fundamental constraints;
- the benefits of scale will continue indefinitely or we’ll get only “two more turns of the crank;”
- it will replace jobs or just tasks;
- consumers and enterprises are really using them or just trying them out;
- value will flow to the closed-source frontier models (such as those from Google, OpenAI and Anthropic) or open-source models will commoditize the foundational model layer; and
- it will or won’t kill us all.
For many professional creatives, it is more than just confusing. It is emotional and personal. Many have a viscerally-negative reaction to anything “AI.” They may consider their art as an extension of themselves and the very idea that a computer can “make art” as offensive; fear that GenAI will threaten creative jobs; and/or believe that training models on artists’ work without payment or attribution is theft.
GenAI raises real legal and ethical questions. But below I explain from a technological perspective why GenAI is foremost a creative tool.
Tl;dr:
- Fundamentally, GenAI models are impenetrable — because they are based on sub-symbolic systems that humans can’t easily understand or modify — and unpredictable — because their output is probabilistic. Their unpredictability is a feature, not a bug.
- The cutting edge of research is focused on ways to improve their reliability, such as through increased scale (of compute and training sets); agentic workflows that spread tasks among many models; and augmenting or conditioning them with known information. But today, they are primarily concept machines, not answer machines.
- As a result, they aren’t currently well suited to many use cases, especially high-stakes environments that require definitive, precise answers that are costly to verify.
- Instead, they are very well suited to the opposite: conceptual, low-stakes, iterative tasks where the quality of output is easily verifiable.
- In other words, GenAI tools are great creative assistants. They dramatically speed the creative process by providing faster feedback; they make it possible to try out a wider breadth of ideas, including riskier ones; they help give shape to partially-formed concepts; and they increase the “surface area of luck.”
- Creatives have a long history of rejecting new technologies as unnatural, threatening and unartistic that later become integral.
- It isn’t possible to stop technology, even if we wanted to. Legislating it, regulating it, shaming it or wishing it away probably won’t work. GenAI is just another tool. Progressive creatives would be wise to learn how it might help their process.
Click here to continue reading the full post on my Substack, The Mediator.