When Game of Thrones returns to HBO for its sixth season this Sunday (April 24), characters will die. We don’t know who, or how, or whether they will be resurrected at a later date, but it’s a near certainty that some of the hit fantasy show’s major players won’t survive the season.
There has been plenty of internet speculation, but if us feeble humans can’t accurately predict who will die, maybe a computer can. To that end, a computer science class at the Technical University of Munich developed a machine learning algorithm that forecasts the likelihood that each Game of Thrones character will live or die.
The project, called A Song of Ice and Data (a play on A Song of Ice and Fire, the name of George R.R. Martin’s fantasy series, on which the HBO show is based), uses a variety of statistical data scraped from the show’s expansive wiki site. The algorithm was then fed a number of features about characters, including their social statuses and connections to other characters (both dead and alive), to determine which ones were most likely to bite the dust.
The most likely to die is poor Tommen Baratheon (pictured above left), the younger brother of deceased former ruler and all-around terrible human, Joffrey Baratheon. At 97%, the computer is rather certain that young Tommen is doomed. With all due respect to Tommen, who’s a much cooler dude than his older brother was, it makes a lot of sense given how many very powerful and violent people are after his place on the Iron Throne.
But some of the other characters the algorithm says will die might surprise you. The four characters with the next highest death likelihoods are Stannis Baratheon (96%), Daenerys Targaryen (95%), Davos Seaworth (91%), and Petyr “Littlefinger” Baelish (91%).
Stannis’s fate shouldn’t be too shocking—the last time we saw him on-screen, he was about to be executed. (The scene cut away before he was killed, but it seems pretty clear that he’s a goner.) More surprising is that Daenerys Targaryen, the dragon-riding ingenue with a legitimate claim to the throne, scores so highly on the death meter.
“The algorithm mostly takes family history into account, and since her family is nearly extinct, her calculated likelihood of death is pretty high,” the project’s website says.
This may be where machine learning fails. Human intuition says the opposite: that the fact Daenerys’s entire family is dead—as well as her importance to upcoming storylines—indicates she’s more likely to live. Then again, Game of Thrones has never been afraid to kill off its most vital characters.
Still, when retroactively applied to past seasons and books, the algorithm was right 74% of the time, suggesting that Daenerys’ fate is not necessarily set in stone. Here are the likelihoods of death for some other fan favorites:
- Tyrion Lannister: 70%
- Arya Stark: 68%
- Jamie Lannister: 64%
- Cersei Lannister: 16%
- Sansa Stark: 3%
Sansa’s strong survival odds are pretty astonishing, considering she always appears to be teetering on the edge of being murdered (plus, half her family has already been offed). Cersei, another woman who has faced quite a bit of jeopardy, also has good odds—for now, her political wit is keeping her from a trip to the Seven Hells.
As for the man of the hour, Jon Snow, who’s totally dead, according to HBO—the algorithm puts his likelihood of death at only 11%. So while he may in fact be dead right now, there appears to be a high probability that he figures out a way to un-die.
This is the second time in the last month that math has been used to unravel the complex world of Game of Thrones. In March, researchers employed network science to figure out who the true protagonist of the show is.