The final season of Game of Thrones is upon us, and that means that heads will roll. Figuratively as well as literally, since impaling, immolation, bludgeoning, crushing and so on are all options on the table. The fantasy series is famous for its bloodthirstiness and willingness to off beloved characters just when you think that they're on the up, so who's next for the chop? Brace yourself: spoilers for previous Game of Thrones seasons are coming.
A team at the Technical University of Munich (TUM) has an ongoing project to answer this question using the magic of machine learning to crunch the available data and predict the deaths of GoT characters with alarming accuracy so far. The best odds of survival belong to none other than the mother of dragons herself, Daenerys Targaryen, with a slight 0.9% chance of death. At the other extreme, the omens are grim for cynical sellword Sir Bronn of the Blackwater: 93.5% chance of death.
The method behind these calculations is similar to that used for predictions for medicine and life insurance. But rather than look at things like diet, hereditary diseases or exercise habits, TUMs data scientists scoured fan-made Wikis (many of which are obsessively detailed) to gather every available scrap of information that seemed to affect a GoT character's lifespan in Westeros. This being a narratively determined world, of course, relevant data points include which house a character belongs to, their allies and enemies, marital status, gender and location, whether they are a major or minor character and how often they are cited in the fan Wikis.
After analyzing this data and extracting the salient characteristics, the data scientists fed it into a deep neural network, which spat out the odds of death. Tyrion Lannister, for example, has a three per cent chance of death. Being a Lannister and a major character boosts his longevity. Yet being a male counts against him: 22 per cent of male characters have bitten the dust already, compared to 11 per cent of women. Being a major character improves survival rates by 16 per cent, which is less than you might expect, but part of the appeal of Game of Thrones is that major characters aren't safe.
TUM has a decent track record of accuracy, since in 2016, the algorithm managed to predict the deaths of (spoiler alert) Tommen Baratheon, Peter Baelish and Stannis Baratheon. It even predicted Jon Snow's resurrection. However, it was wrong about Daenerys, who was given a 95% chance of death three years ago. This was all using data from the Song of Ice and Fire novels. Speaking to Wired, TUM researcher Christian Dallago mentioned that the show moving away from George R.R. Martin's books and being shaped by other writers has really messed with the data.
Upon hearing this news, our editorial office immediately sounded (ok, the office Slack channel) with cries of allegiance to one character or house to triumph. Pick your favorite and see the odds of death below:
To access all the predictions and see a breakdown of the data for each character, you can visit the dedicated site, A Song of Ice and Data. And don't get too cocky, either. It's the season finale and the writers will have some dramatic reveals up their sleeve. Valar Morghulis.