No juggling, no religion. It’s the Edinburgh Fringe.
- Published
- in Performance
The Edinburgh Fringe programme is out and the first task for visitors is to work out what to see. The Edinburgh Fringe boasts over 3,500 different shows, and expects to sell more than 2.5 million tickets over its three weeks of performances. That means there’s a decent chunk of data to play with. You can see (or scrape) all of the shows on the edfringe.com site, or you can pick up a printed programme. So how do you choose? It felt that this should be a useful example of seeing how data and AI are being used in the arts community.
It’s almost a year and a half since ChatGPT became public and the world started agonising over the possibility of AI taking over from humans. This year marks the first real opportunity to see how AI influences the Edinburgh Fringe – the world’s largest arts festival. Will it be used to write new plays, or to tell audiences what’s hot and what’s not?
To help you choose, the Fringe programme is divided into ten categories, ranging from Children’s Shows, through Comedy, Dance, and Music to Theatre. To help punters narrow down their choice, each show can select up to two additional “genres”, from a list of 100 options. That structure will be familiar to any who’s ever used a database. In the AI database world, the “genres” are key to what is more usually called tagging. Tagging is generally done by poorly paid employees in the third world (read Madhumita’s “Code Dependent” if you want to know more about the shadowy world behind tagging AI datasets), but in this case it’s left to the performers. What could possibly go wrong?
As shows may span more than one category, the genres repeat the main categories, letting a company inform their audience that it may be a children’s show that’s also theatre, (there are sixteen of them), or theatre piece which is also a children’s show (of which there are five). When a company chooses its genres, it’s clearly told not to repeat the main category in its choice of genre, which most adhere to, with the prominent exception of Cabaret and Variety, where 15% of the shows want to reinforce the fact that they are either cabaret or variety. They could be a sign of insecurity or just that cabaret and variety performers don’t read the instructions. In the latter case, there’s a show for them at Greenside – RTFM (Read the F****** Manual).
To be fair, their primary category is “Cabaret AND Variety”, so their choice may reflect which of those two they think they are. However, one production has chosen both. That combination of multiple possibilities in the primary category leads to another issue. Where it includes two nouns, as in Alternative Comedy, or Science and Technology, there are two different genres to choose from – one with an uppercased second noun and one with a lower case one.
As an example, if your interest is Science and Technology, you have a choice of nine shows. If you prefer Science and technology, you’re only presented with six. How companies choose between the pair seems fairly random. Alternative Comedy shows have a heavy preference for all uppercase (fifty versus twelve), possibly because it’s generally shouty; Musical Comedy and Character Comedy have a slight preference for uppercase, while Dark comedy tips into a lower case preference. Live music, Children’s shows and Food and drink all prefer the lower case option.
This is all a bit of fun until you think about what these tags are being used for. In the printed programme they are just pieces of additional information which may or may not be useful. On the web page, you can use them in the advanced search to help you find a show. Here, the uppercase/lowercase choice affects the results. There is no commonality between the results returned from the pair of seemingly identical sets. The two options lead to different recommendations, so the performer’s preference affects their search results. The shows you find when you search for Alternative Comedy are totally different from the ones you get when you choose Alternative comedy and vice versa.
In the new Fringe app, you can do the same sort of search, although what is called a “Category” on the web search is called a “Genre” on the app and the app doesn’t show any sub-genres. But it does offer me shows “based on my preferred genres”, even though I haven’t selected any.
If you’ve ever used a web commerce site, you’ll know what to expect. Here’s an example of Amazon’s recommendations:
And here’s the Edfringe app’s version:
There’s an obvious difference from a first glance. Amazon gives ratings, whereas Edfringe does not. As the Fringe Office has a policy of being totally impartial, I would have assumed it selects the options it gives me to be purely on what it knows about me. However, that doesn’t seem to be the case, as it keeps on recommending tickets I’d already bought. So, my guess is it’s just random. Although it looks like Amazon, it’s a dumb imitation. Amazon’s has a complex algorithm driving it. The Edfringe app doesn’t. I wonder how many users will appreciate that?
Looking through the sub-genres, I counted exactly 100 different tags. Their frequency of usage is illustrated in the work cloud at the start of this article. Whilst that nice, round number may look planned, I suspect it’s not, as ten of those were uppercase / lowercase duplicates. That leaves 90 options for a show to choose from. That seems an awful lot, particularly as almost 60% of them are chosen by fewer than 1% of the 3,500 shows. The point of tagging is to use the minimum number of relevant tags, not hand them out like candy. The implication here is that anything that a company suggested as a category has been added, regardless of data design. There are also some interesting omissions. There is no category for Juggling, but if you do an image search for the Edinburgh Festival, a large percentage of the images are of jugglers. They also feature strongly in the Fringe’s official trailer. There’s also no category for religion, although I doubt that will bother many festival goers.
The bulk of these sub-genres are about the type of show, not the content, which is also interesting. Over the years, Fringe companies have learnt that a good way to attract an audience is to offer nudity or to be offensive. So why are there no tags for Nudity or Offensive? Or even Controversial? There’s certainly no reticence about pushing those features in the listings, with eight shows promoting nudity and five promising to offend. I suspect there are several more that decided to be coy about their copy, but plug it in their PR.
The opposite of offensive is probably Family-friendly, which takes us away from form and into safe spaces. Intriguingly, only 18% of shows in the Childrens’ category use that tag, although having sat through some of them in the past, that may be honest. Our daughter preferred Titus.
If we look at the most popular sub-genres, which have over 100 uses, we have Stand-up, Comedy, Storytelling, New writing, Solo show, Drama, LGBTQ+, Musical theatre and Theatre. That prompts the question of what LGBTQ+ means as a performance category? LGBTQ+ performers have been responsible for many of the Fringe First winning shows over the history of the Fringe, but would have described their shows with one of those other genres. It’s been suggested that its use here is similar to “Family-friendly”, indicating a safe space for a community. With no guidance for performers or audience, it’s difficult to tell. That ambiguity takes us on to gaming the system.
TERF – a new play based on JK Rowling’s views on trans issues, promotes itself as “the most provocative to be staged at the Fringe for years”. Its online description has already been sanitised from what is in the printed programme, removing any mention of JK Rowling. I’m looking forward to seeing it, but doubt it will be a “safe space”, even though it carries the LGBTQ+ tag. This is where letting performers select tags falls over, as without consistency in their application, they have limited meaning. A company wanting to maximise ticket sales should probably look to see which tags last year’s sold-out shows used and copy those, even if they’re only marginally relevant. If you want to offend, choose the “safe-space” ones, as the resulting complaints should fuel your PR coverage.
With AI and apps so prevalent in our daily lives, it’s interesting that nobody has come up with a good way of highlighting the best shows to see. Back in 2009 we had Edtwinge.com – a webpage which scraped Twitter reviews to try and rank shows. I always thought it worked remarkably well, probably because not enough people were aware of it to game it. But by 2015 it had disappeared. It worked by looking at the thousands of opinions expressed every hour on Twitter to provide a crowd-sourced real-time Edinburgh Fringe review service. As well as monitoring the general Twitter “noise” for each act, Edtwinge also generated a “karma” rating to rank the performance, generating the following scoreboard.
They explained that their “Karma” rating was “the lower bound of the Wilson interval computed for the ratio of positive reviews to the total of combined positive and negative reviews as contained in tweets mentioning Fringe acts, calculated with a statistical power of 0.05 or, in other words, a 95% confidence level”. You can read more about this on their web-archived page. Today we’d call it AI rather than Karma. It was way ahead of its time and worked rather well. It’s a shame it only lasted a few years. It feels that there’s a good project here for some of the students in Edinburgh’s burgeoning education industry.
Despite the noise around AI, there’s not a lot of it in evidence at the Fringe. Of the 3,500 shows, only 11 mention AI, with two claiming to have used it to create the show. Only one mentions data – a talk about data privacy in AI. None mention ChatGPT or large language models.
There’s an old adage that if you gave a thousand monkeys typewriters, they would write the complete works of Shakespeare within a thousand years, but produce a fair number of Fringe shows within their first day. That joke and their jobs have been made redundant by the advent of ChatGPT, so you might have expected a bit more focus on AI and its ramifications. But maybe we’re already there and just don’t realise that’s what we’re seeing.
As always, the best way to find out what’s worth seeing is to put down your phone and talk to the people next to you in the queue. Word of mouth still rules. So, I’ll finish with what I’ll tell anyone standing next to me that they should go and see:
- Or What’s Left of Us by Sh!t Theatre
- A Letter to Lyndon B Johnson or God: Whoever Reads This First
- The Sex Lives of Puppets
- Johnny and the Baptists
- A Brief History of Difference
- TERF (but book early before it’s cancelled)
Enjoy your Fringe.