Critiquing an AI-Generated Architecture Critic

After requesting an AI-generated architectural review of the Perot Museum, Mark Lamster, writing in the Dallas Morning News, evaluates the result. Lamster annotates the review with comments, then explains the flaws found in the automatically generated critique.

“To my critical mind, the Perot’s most significant flaw is that it turns its back on the city around it, offering a cold concrete shoulder to the adjacent Arts District. Our computer-generated review suggests just the opposite. Not good.” The review remains glowing, including factual inaccuracies and misunderstandings of the museum’s role in the surrounding neighborhood. As Lamster points out, understanding a building or place requires being there. “But visiting a building in person isn’t possible if the critic isn’t a person at all.”

Because an AI program can’t ‘be there,’ it ends up, like a lazy college freshman, culling what material it can find floating around the internet and regurgitating it in a generic format.”

This may work for public relations, but not for criticism. “Indeed, one of the critic’s essential tasks is to cut through spin, to provide readers with unvarnished, informed opinion, and to do so with a bit of panache.” In Lamster’s opinion, “Of course we all want to be objective (however loaded a term that is, and especially when judging abstract arts), but who wants criticism without personal opinion?”

Ironically, when asked about the limitations of AI-produced critique, ChatGPT nails it, acknowledging its inability to replace, at least for now, human experience and knowledge. The self-awareness of that answer in itself, writes Lamster, “has me worried.”