AI in Design and Society

In everyday work, AI is a new tool to be explored. In research it opens up an unfamiliar field quickly and helps order sources. In programming it takes over routines and makes unfamiliar techniques accessible. In conceptual phases it serves ideation and visual exploration by offering variants against which one’s own judgment can sharpen. Its real leverage lies in making an average expertise broadly available and so widening the individual’s range of competence. This widening remains a loose thesis whose consequences cannot yet be assessed. What is certain is only that the results call for a critical eye, precisely where one’s own expertise is lacking.

Research on creativity shows a flip side of this tool. Holzner, Maier and Feuerriegel pool 28 studies with 8,214 participants in a meta-analysis and find that people working with generative AI perform somewhat more creatively on average than without it, while the resulting sets of ideas lose markedly in diversity. Wadinambiarachchi and colleagues observe, at the individual level, how the mere sight of AI-generated examples fixes designers on a first solution and lowers the number, variety and originality of their sketches. Both findings point to a narrowing of the solution space. For design practice this raises the concrete question of how to resist the obvious first solution and work more iteratively. Design as problem-solving thinking does not change at its core, since it has always been about judgment, selection and framing. Its practices, however, shift toward a co-creative way of working in which proposing and discarding interlock between human and machine.

Beyond the tool, AI touches how we communicate and perceive. Hito Steyerl calls the generated pictures mean images, statistical averages that put probability in the place of facticity, and Lev Manovich describes how AI co-shapes the aesthetic decisions of an entire culture. When synthetic material is processed recursively, a homogenization threatens that levels the diversity of the source data. As access to information runs more and more through generative, language-based interfaces, the design of those interfaces helps decide which knowledge remains visible at all.

Skepticism is provoked by the pace. AI develops exceptionally fast and opens great potential for the individual, which at the same stroke empowers malicious actors to an unforeseen degree. Regulation does not keep step with this speed, and so corporations arise that fully exhaust ethical and moral limits. Kate Crawford describes AI in Atlas of AI as a technology of extraction that concentrates power in the hands of a few, while Emily Bender and Timnit Gebru show how models amplify the biases of their training data. The same loss of variety runs through these findings, from the solution space of design work through the homogenisation of image culture to the concentration of power among a few providers. Productive use and critical distance toward this technology therefore do not exclude one another, they depend on one another. Anyone who uses these tools and shapes how they are deployed can work against that loss where it becomes visible. What that requires shifts with every generation of models and in every field of application.