One of the big problems we noticed when walking around was the difficulty of accurately explaining technical concepts in the museum format. One technological exhibit in particular demonstrated neural network inpainting with (what we presumed to be) epoch-by-epoch visualizations, but all of this would have been so unclear to someone without a background in machine learning. There was a diagram of the corresponding neural network architecture, but I have no idea how a casual observer would have any idea what it meant beyond being some aesthetically pleasing black and white boxes placed on top of each other.

I personally felt that the AI exhibit devolved into showing people “impressive results” from the history of computing, but with little context for how they worked - style over substance.

Ultimately, the piece in the exhibit that we were the most impressed by was a huge mechanical box created by Claude Shannon in the 1900s - essentially a hand-built computer. In contrast to the more digital/“modern” exhibits, we felt like this box better captured an interesting slice of the history of artificial intelligence, and it was so much easier to immediately understand what it represented.

This isn’t to say that technological exhibits can never work. They’re probably perfect for explaining how modern machine learning works - they just require more context. Museum curators are experienced with presenting physical objects well, but there is a slightly different style that integrated technology needs to use before it truly complements the exhibit itself.