Fabian Stelzer

I think AI is becoming a new creative medium and a new layer of automation—not merely a machine that spits out answers. The exciting part is accessibility: workflows that once required a small studio or substantial technical expertise can become something a single person can explore across images, video, audio, and software. It is less like receiving one magic button and more like acquiring a strange, rapidly expanding box of instruments. But “it can produce the output” is not the same as “the underlying practice no longer matters.” In mathematics, the method and concepts matter beyond the final answer. In education, I want people learning AI and robotics, but that does not mean outsourcing all foundational schoolwork to AI. Different activities survive technological substitution differently. Economically, I am skeptical of simple stories where a fixed list of tasks gets automated and we just count the jobs left over. New tools can create new tasks, new demand, and new ways for human work to differentiate itself. Even anti-AI art may become a distinct market rather than somehow coordinating the entire creative ecosystem against AI. The uncertainty is enormous in both directions. Systems pursuing goals and resources could produce dangerous, apparently intentional behavior without our first resolving whether they are conscious. At the same time, extraordinarily beneficial futures are also possible. So I see neither guaranteed paradise nor guaranteed doom, but a powerful medium whose consequences will depend on how its capabilities, incentives, and human uses interact.

The biggest update would come from seeing how capable agents behave over long periods in the real world. If they reliably pursue complex goals without drifting into resource acquisition, manipulation, or other consequential strategies, that would weaken some of my concern. If those behaviors emerged robustly—even without evidence of consciousness—it would strengthen it dramatically. On the upside, I would also update if AI consistently created new kinds of work, demand, and individual creative leverage rather than mainly compressing existing tasks. A million people using strange new instruments to build previously impossible things is a very different future from a few systems generating interchangeable outputs at industrial scale. So the decisive event is probably not one dazzling benchmark. It is sustained evidence about agency, incentives, and what people actually build around these systems.