The AI that plans to shield us all from spoilers


When you fall behind on your favorite TV show, there’s nothing worse than having someone spoil it for you. Whether you accidentally see something online, or a coworker just gets a bit too chatty in the office, it can really ruin your day. Luckily, it seems that a team of researchers in San Diego are on the verge of removing spoilers for good. At least, the ones that are found online, anyway.

Photo: Unsplash.com

Stopping the spoilers

Aware of how annoying spoilers can be, this team from the University of California decided to find a way to stop them. They figured that the best way to do that was to design a form of artificial intelligence which would intercept the information and hide it from people. That way, no-one would have their favorite movies or TV series ruined by anyone on social media. Achieving this wasn’t going to be easy, but if they were successful, the payoff would change people’s lives for the better.

SpoilerNet

The outcome of their hard work was SpoilerNet, a neural network capable of detecting spoilers on its own. Early tests of the software weren’t without their problems, with the accuracy rate far from 100%. However, despite not working perfectly, the network still achieved far more success than failures. It managed to detect spoilers a majority of the time, putting potential users in a far better position than they were without it.

A library of spoilers

To try and teach their network how to detect spoilers, the San Diego team started by getting it familiar with literature. They compiled over 1.3 million different book reviews and put them into the system, giving SpoilerNet a good idea of what to look out for. The accuracy rate of this test was incredibly positive, with the researchers achieving successful results around 90% of the time. Unfortunately, the case wasn’t quite the same with TV shows.

Not as successful

When the team moved on to television, they filled the system with over 16,000 summaries of nearly 900 series. The hope was that SpoilerNet would produce similar results as it did with literature, but the accuracy rate was only 80% at its best. The problem – according to the researchers – was down to some of the words used in these summaries. Apparently, the system got confused by them, believing them to be more revelatory than they actually were.

Photo: Unsplash.com

Something to be proud of

However, while things might not have worked out so well with this test, the future’s still bright for SpoilerNet. An accuracy rate of 80% is quite promising, especially given it’s the first system of its kind. Mengting Wan, one of the students working on the project, said nothing like this has ever been done before “at this scale” or “fine-grained granularity.” It might still have issues to iron out, but it’s revolutionary nonetheless.

Aspirations for the future

The hope for SpoilerNet is that it will one day be able to prevent people from having things ruined on social media. However, that’s not all that the team has planned for this system. They’re determined to try and make it work on other platforms too, as well as use it to determine how people write spoilers. There’s a lot they think this system can achieve; it just needs more time.

It’s exciting to think that there’ll be a day in the not-too-distant future where spoilers will be a thing of the past. That day can’t come soon enough!