The first article enumerates the ways in which AI can be used in museums:
- Robots to record visitors’ emotions and preferences towards artworks
- Interactive robots to answer visitors’ questions and interact with visitors
- Data analytics to predict no-shows in order to release more tickets
- Image recognition matching pictures with artworks of the gallery
- AI for museum operations to measure and forecast visitor behaviors to save operating costs
- Sentiment analysis of visitor comments
Personally I am somewhat skeptical about 1 and 2, as having an object of human-form running around and interacting with / observing you can be an intrusive experience, and that might not work well with museum settings if some visitors prefer a quieter environment to enjoy the art. 3 reminds me airline oversales, but it seems to be a less defined problem in the museum setting. An aircraft has a fixed capacity limited by the number of seats, and if oversales result in exceeding the capacity, the airline would offer incentives for volunteers to give up their seats. However, the capacity of museum spaces is less well-defined, and oversales without adjusting for the capacity (it’s unlikely that museums would offer incentives or enforce a strict capacity) might result in a worse visitor experience. I am the most excited for 5 and 6, which use AI to improve both the visitor experience and the operational / curation practices, without being too visible and intrusive to the museum experience.
The second article describe a specific museum’s experiments with using machine learning. The blog was written 9 months before the release of chatGPT, before the widely available access of OpenAI APIs. I wonder how the results of the experiment will change with the better machine learning models. I find the following statement interesting, “Throughout the project, it became apparent that the commercial machine learning services are primarily geared toward the needs of commercial customers, for uses in marketing, customer management, call centers, etc… We hope that future experiments with using machine learning tools customized through training on our collection can address these stringent quality requirements.” Based on the numbers in the article, a total of $115000 was spent on this project, including compute and development costs. The budget is very small compared to typical commercial ML projects. Given the budget constraint of the museum (and most likely other museums), I wonder to what extent can models be fine-tuned for museums.