Solving AI's toughest problem with the Harry Potter series

Bloomberg reports that a growing number of AI experts are taking advantage of the series' enduring influence on pop culture, as well as its rich linguistic data and diverse wordplay, to experiment with advanced generative AI technologies.

For example, algorithms are being built to selectively “forget” information, a difficult task for large language models, which are trained on large amounts of online data including copyrighted and uncopyrighted material.

The Harry Potter series is used as a literary reference due to its popularity and the richness of natural language in the stories.

Researchers Mark Russinovich and Ronen Eldan from Microsoft say they have demonstrated that AI models can be altered or modified to remove any specific knowledge, such as the existence of the Harry Potter books, including their characters and plot, without affecting the system's decision-making or analysis capabilities. The pair say they chose Harry Potter because of its popularity.

“Even people who haven’t read the book can be aware of the plot elements and characters, so our technique can be tested to confirm whether the algorithm ‘knows’ the book,” said Russinovich, CTO of Microsoft Azure.

In another study, researchers from the University of Washington, the University of California, and the Allen Institute for AI developed a new language model called Silo that can eliminate data to reduce legal risk. However, the model's performance dropped significantly when trained only on low-risk text, such as out-of-copyright books or government documents.

With Harry Potter, “the richness of the scenes, the dialogue, the emotional moments make it very well suited to the specific field of natural language processing,” says Leila Wehbe, a Carnegie Mellon researcher who conducted a series of experiments in 2014 collecting brain MRI data from readers of the series to learn about the mechanics of language.

Even when not the focus of a study, Harry Potter remains a favorite literary reference for researchers. It is used, for example, to test the “intelligence” of algorithms. Terrence Sejnowski, head of the computational neurobiology lab at the Salk Institute for Biological Studies, argues that AI models reflect the intelligence and biases of their users, much like the “mirror of illusion” in Harry Potter—which always reflects the desires of the person looking at it.

(According to Bloomberg)

Microsoft to launch AI-powered Surface laptop line in 2024

Microsoft's upcoming Surface Pro 10 and Surface Laptop 6 will come with a next-generation neural processing unit (NPU), along with Intel or Arm chip options.