AI Is Good at Design That Already Exists. Silber Says Breaking New Ground Is Still Human Work.
OpenAI's design chief says the human edge is knowing what people need before the data exists.
Lenny's Podcast: Product | Career | Growth · OpenAI’s Head of Design: This is the best time in history to be a designer | Ian SilberIan Silber, OpenAI's head of product design, was asked the awkward question for a design chief: how good will AI get at design itself. Silber said, "I think it already is an incredible product designer," and that it is now accessible to every single person as a tool for being more creative and building more.
He meant something specific. In his view, AI is already great at product design, though not necessarily the best at visual design, information hierarchy, typography, or interaction design. It will keep getting better, he said, after the rapid progress of recent years.
The question came up when Silber recalled asking the product manager for Codex why AI was not better at design. Silber said the Codex PM argued that a model trained on today's most interesting websites learns from sites that are no longer fresh, so its output is, in his words, "not interesting anymore." The other issue raised was that AI researchers are not innately great at design, and may not know how to teach a model that craft.
Silber pointed to products that broke precedent. The iPhone created a multi-touch interaction paradigm with no training data behind it. Instagram embraced the fact that everybody had a mobile phone with a camera. Snapchat introduced what first looked like a strange interaction design and then completely flipped how people wanted to communicate.
The same problem faces designers now, he argued, in figuring out how to talk to AI with voice or how to have an agent carry out tasks. There is not much precedent for that either. Silber said humans "are able to kind of do a really good job evaluating what good is." The human work he named was observing how people use products and understanding unmet needs.
Silber also tells designers to hone a point of view, asking, "what is my point of view? Who am I building this for and why?"
Questions people ask
Why wouldn't AI have training data for something like the iPhone?
It means AI learns from large collections of past work, so it is strong at patterns it has seen many times. When Apple introduced multi-touch, by definition there were no prior multi-touch apps to learn from. A human had to guess what would feel natural and test it with real people.
Does this mean human designers are doomed?
No. Silber says AI is already great and accessible to everyone, and expects it to keep improving. His argument is that designers who use it to prototype and explore, while supplying the new idea and the understanding of users, will do more and better work.