It seems very logical to set AI limits for coming up with solutions — this will help make them more useful. And this is true when we set neural networks tasks "for performance", but if more creative solutions are required from AI, it is better to experiment with removing, imposing and changing restrictions in order to get the most effective result.
For example, brand managers are already using special AI tools like Lately or Jasper to create useful content on social networks. In order for the resulting result to correspond to various formats and brand image, they often set clear limits on the length, format, tone or target audience of the content.
However, striving for true originality, brand managers can reject formatting restrictions or, on the contrary, limit the output to an unconventional format. A great example is the Help Changes Everything campaign from GoFundMe
: the company sought to create a creative annual review that would not only express gratitude to donors and evoke emotions, but would also stand out from typical content at the end of the year. To achieve this, they set unorthodox restrictions: the AI was required to create visual effects that would rely solely on street art, and depict all the fundraising and donor campaigns conducted over the year. DALL-E and Stable Diffusion created separate images, which were then transformed into an emotionally charged video.