Researchers at MIT’s Computer Science and Artificial Intelligence Laboratory have used a special algorithm to learn which characters in Game of Thrones are the easiest to remember according to visual cues alone.
Margaery Tyrell (Natalie Dormer) is the second-most memorable character on GoT, while Joffrey Baratheon (Jack Gleeson) is the 12th, so says MIT’s image memorability API.
Before we get into the mechanics behind the AI’s choices, here are the rankings:
1. Daenerys Targaryen: 9.42
2. Margaery Tyrell: 9.07
3. Tywinn Lannister: 8.81
4. Jaime Lannister, Cersei Lannister, and Bronn: 8.78 (three-way tie)
7. Sansa Stark: 8.7
8. Jon Snow and Ned Stark: 8.61 (two-way tie)
10. Gregor Clegane: 8.22
11. Tyrion Lannister: 8.15
12. Joffrey Baratheon: 7.25
To compile this list, CSAIL PhD student Aditya Khosla and graduate student Akhil Raju, along with MIT professors Antonio Torralba and Aude Oliva, used a technique called “deep learning” to get computers to find patterns on their own.
To teach their computer which images are the easiest to remember, the researchers fed the system with over 60,000 images representing the world’s largest memorability dataset (you can actually upload your own photos here to get your own memorability assessment). Each image was assigned a “memorability score” based on human participants’ abilities to recall them in a memory test. Algorithms meticulously analyzed the images, looking for discernible features that make them more or less memorable.
“Memorability is a concept that defines the extent to which anyone remembers an image,” Khosla told io9. “It is important because it applies to everyone in general, and we cannot consciously control it — we are all very likely to remember the highly memorable images, and forget the less memorable ones.”
Khosla says the most memorable images are the ones remembered by a very large percentage of the people they are shown to. In the case of Daenerys Targaryen, for example, that’s 94.2/100 people on average.
“While the applications in advertising are obvious, this research can be extended to apply to areas like education, where we can show students content in a more memorable way in order to make things easier for them to remember,” says Khosla.
[ MIT Computer Science and Artificial Intelligence Laboratory ]