The psychological issue with generating text, code and images using AI is that people become overly enthusiastic and fail to assess the quality properly. Engineers at Figma AI explicitly warn about this: "as AI makes it easier to generate designs, the risk is shipping something average". I see this as the main challenge and my role lies in quality control and predicting the effectiveness and appropriateness of using AI at any given stage of the work.
Let’s look at an example. The first image was generated by designer using AI and approved by the marketing team. The second image is the result of my art direction: an improved idea and the image.
One day, the marketing team sent us a list of 15 ideas for visuals for a marketing campaign. The ideas had been generated by AI, they were either trivial or irrelevant. In the design department we rejected them all. The problem is that LLM-based AI, by its very design, tends towards triviality, but many people don’t realise this and believe that if you generate a lot of AI-slop, an original idea will eventually emerge from it. In reality, it doesn’t work like that. Without an initial concept and human intellectual input, nothing good will come of it, and a lot of unnecessary time and money will be wasted on generation.
