Could AI ease Game of Thrones-esque backlash in future?
The dust might never settle from the public uproar following the conclusion to global phenomenon Game of Thrones, so in the meantime we should ask the question whether technology could save writers’ blushes in the future? Would HBO trade the imaginations and creative flair of D. B. Weiss and David Benioff for an algorithm — all in the name of erasing the one blip in an otherwise perfect journey?
Research emerged this week claiming natural language processing tools enable more valid and accurate computations, while visualization and analysis of word frequencies across text provide insights into how narrative is structured. It goes even deeper, with network theory able to calculate a guide to which characters are most difficult to kill without disturbing too many interrelations — techniques with the potential to sculpt the ultimate plot.
A preliminary exploration into applying AI and machine learning techniques to something as subjective as fantasy novels has been carried out by Peter Vesterberg, a data science expert at Swedish machine learning firm Stellapolaris. In Part 1 published this week, he aims to unearth how and if a numerical processing of the books on which the series were based can reveal hidden patterns within A Song of Ice and Fire.
The divisive study walks a tightrope between erasing the element of human originality and fundamentally improving a show’s appeal and therefore ratings. Unsurprisingly, the study has received its fair share of reproach on social media with words to the effect of “the machines are taking over”.
Of the three aforementioned approaches, calculating the importance of characters using network theory jumps out as the technique which studios would see as most tempting. Network theory is a field of mathematics which has appeared in telco networks and search engines, relating to the study of graphs as connections between discrete objects. But for the sake of brevity, let’s brush over the math-heavy detail and get straight to the point. These network theory algorithms create a measure of a relationship, on the basis of how frequently and closely two names occur — creating a graph (more like a web) of relationships. The thicker the line, the stronger the relationship.
But this complex web of over 2,000 characters (nodes) needs simplifying further, which the study does through four key measures of centrality in network theory. These are: Degree centrality (proportion of nodes), closeness centrality (degrees of separation), betweenness centrality (number of times a node acts as a bridge along shortest path), and prestige centrality (the importance of nodes based on connections). The latter is the method commonly used in web search algorithms.
So, which is most important? Betweenness centrality is of particular interest to study author Vesterberg, given Game of Thrones’ tendency for killing off key characters at a steady rate. Meanwhile, betweenness centrality provides a measure of how difficult it is to replace a character without significant impact on the rest of the network’s connectivity. But surely in a writer’s world, reverberating the web as hard and as unpredictably as possible, is precisely the point?
Nevertheless, network theory establishes that Jon Snow is the most important character, as shown in the graph below, and therefore killing him off would put the writer in a difficult position with regard to maintaining a connected and coherent story.
Despite offering an absorbing introductory discussion around applying data science to storytelling, what the study doesn’t factor in are the potential use cases for techniques like network theory in the real world of content creation. From a studio perspective, the embarrassing 4.3 out of 10 stars rating (as of writing) attached to the final episode of season 8 and the resultant backlash could be interpreted as a lesson in how to destroy a cash cow in just a few easy steps. The ramifications for HBO could manifest as swarms of subscribers abandoning the streaming platform HBO Now to make a point of their disgust beyond outbursts on social media. Attempts at future hit series could be held in the same regard, suggesting the AT&T-owned company may take years to recover from what unfolded at the beginning of this week.
The reputation of multi-million dollar cinematic experiences could therefore be protected by using technology to better understand what viewers want. As creative voices have cried, this may indeed mean sacrificing the element of surprise while dealing a crushing blow for budding writers the world over — points which are irrelevant from an investment standpoint. Complementing the work of humans and ultimately making us more efficient is the intrinsic benefit of AI and machine learning, not that scaremongering headlines would let us believe it. Writers could eventually sculpt their work with a helping hand from data science, not necessarily employ an AI to do the job for them, and that is an exciting prospect.
A sequel to Part 1 of Decoding Game of Thrones By Way of Data Science should arrive in the coming weeks, exploring how machine learning can construct different non-deterministic representations of the books, which we hotly anticipate.
This is an extract from Faultline Online Reporter — providing a framework of technical understanding to help predict the future of digital entertainment and the global broadband economy.
