Can a computer create a musical? The simple answer is ‘no’. Not yet, at least. In 2016, a musical Beyond the Fence opened in London. AI wrote most of the musical, but it wasn’t a success. The computer can create some elements. Perhaps better than a person can. But it falls short on others. At least still in 2020.

In this article, I want to consider AI and the possibility of it creating a musical on its own. Let’s discuss! 

Can a computer create or produce anything?

I can really dive into the material, but I assume none of us here are true IT nerds who understand the deep dark worlds of computers. So let’s try to keep it as simple as possible. The simple answer is: a computer needs data and an AI.  

Computers can certainly ‘invent’ things and have been doing so since the 90s. And they can do so by artificial intelligence (AI). Where in the 90s AI was still a futuristic word fitted for a sci-fi movie, we can’t avoid it anymore in 2020. The topic is hot. And evolving fast. More and more data is getting generated and we need AI to collect and sort it all. It’s the Big Data Era! If you want to learn about how big data influences television, read my other blog here. Examples of AI in the modern world are self-driving cars and Apple’s Siri. 

But what is AI exactly? It’s an algorithm with the goal to create a computer that can think like a human. We are trying to imitate the human thought process with a computer. Sounds scary? Maybe it is. Look at the deep fakes that are circulating the internet right now. The ambition is that you don’t need to teach the computer step by step what to do, but that the AI observes, classifies and learns from its own mistakes. You write an algorithm and the computer acts according to that algorithm. And as it learns, it changes the code to improve itself. Up to the point where programmers don’t know why computers make certain choices.  

AI is creative, but also really stupid in a sense. The problem AI is really struggling with now is cause and effect. AI can associate events with each other by learning it from other examples, but it doesn’t understand how one thing makes another thing happen. Besides that, a team of humans start with a code and the computer acts according to the code. So if it’s supposed to analyze cat pictures, it will analyze cat pictures. But he won’t know what a dog looks like. 

Can AI create art?

But AI can also produce or invent. In 2018, a portrait of Edmond Belamy sold for 432,000 dollars. Never heard of him? That’s probably right, as a computer-generated the portrait after analysing 15.000 portraits from the 14th to the 20th century. See the picture below: what do you think of it?

It started a real debate about the essence of art. Isn’t art when an artist creates something from within – an introspective – and that adds value? Can a computer do the same, even though it has no internal world? So far, most computers seem to analyse other artworks and then mimics what they’ve learned. And research shows that the AI’s without human input create boring results. It misses intentionality and sensibility that the human brain has. But maybe it’s just as Andy Warhol once said: ‘Art is what you can get away with’.

Most likely, the computer and human will go hand in hand to create art. This has been done for many years now and has had very successful results. AI can help artists by saving them time. Programs like Adobe Photoshop, for example, start using AI to help automate repetitive tasks. The artists have more time to focus on the more authentic parts of their art. Other artists use AI to create the artwork itself. Take a look at https://aiartists.org/ to see examples of AI-created art.

But AI is mostly great in creating art that’s systematic. It can write poetry, which is about rhythm and rhyme, it’s mathematical. But at present, it’s incapable to write prose. AI is machine learning, but it doesn’t yet have the ability to create a narrative structure. 

Beyond the Fence: the computer written musical from 2016

In 2016, an experiment was done with the musical Beyond the Fence as a result. Scientists wanted to find out if a computer can write a musical. Remember, this was four years ago. In AI development, that’s ages. But let’s take a look as the process was quite interesting. 

The experiment started at Cambridge University, where the computer analysed thousands of musicals to find out what makes a hit or a flop. The results to create a hit musical were:

• a love story
• set in the 80s
• in Europe
• a female lead
• a death
• overcome the loss
• a happy ending

In the music styles, you need to have multiple styles and also different pacing in the tunes. You better end each of the two acts with a powerful song. 

Then came the story writing. The university Goldsmiths stepped in with their ‘What-if Machine’. They tried to spark creativity by giving the computer topics and concepts to choose from. The computer sprouted many ideas and then the scientists discarded most of them as they didn’t fit with Cambridge’s conclusion of a hit. In the end, the storyline was about a single wounded soldier.

Next in line was the University of Madrid. They tasked their software with refining the plot. Once again, it was fed thousands of musicals and the software analyzed the lyrics and the emotional course of each musical scene. Now, it could tell which emotions should be in the musical on which moment, thus setting more structure for the plot and music. 

But this is where the computer ended. Because in 2016, a computer could not yet put the music, story and lyrics together. It had to be done by human hands. So Benjamin Till and Nathan Taylor took the computer’s output and created the musical with those elements. The computer had some input for the music and lyrics, but the creators mostly used the input as inspiration.

Beyond the Fence was not a success. Even though it had all the success factors the computer calculated. Robert Workman wrote a review for Engadget whose opinion I saw reflected back in other reviews:  

‘By coming up with the perfect formula for a musical, you invariably create something that’s formulaic. When watching, every song felt calculated, each plot point carefully measured. Neither act shocked: There’s a chance meeting here, a romantic entanglement there, a death and redemption thrown in for good measure. Nothing moved the needle. Nothing felt fresh.’

Robert Workman

So it’s quite interesting that a computer can give all the success factors but in the end, creativity was what missed. The computer works mathematical and in return, the musical was mathematical. And clearly, that’s not why we go to the theatre. 

The key elements of musical

Let’s first see what elements a musical exists of. As a computer focuses on the creating part in the pre-production phase, this eliminates acting or any crew action during the show. 

A musical can be described as a play that combines songs, spoken dialogue, and dance to tell a story where each element is equally important. The three key items of a musical are the book, the music and the lyrics. The music and the lyrics together are called a score. 

Once these three are complete, the creative team will broaden with among others the director, the set designer and costume designer.

Let’s stay with the creative core of the musical, so the book, music and lyrics. Can a computer produce these three? If you would like me to write about AI in decor, costumes etc, let me know in the comment section. 

The book written by AI

Can a computer write a story? Well, I suggested the answer already earlier in the article when I said the computer can’t write prose. 

AI can already create short stories. Some newspapers in the USA already use AI to write local news stories. Often, they are pretty formulaic: who, what, where and when are the input and the computer writes the rest. Sports reports are perfect for AI. It has also written multiple poems that people don’t recognize as AI-written.

Can a computer fill 50.000 words enough for a novel? Yes, absolutely. Will they be words we understand? Also, absolutely yes. Will it be a nice read with context? No. At this moment, AI cannot write a good story with a coherent plot. Yet. To write a wonderful story, AI has to understand story structure and concepts like plots. 

However, AI is already helping authors write their story. AI can check grammar (think about Grammarly!), generate names and give ideas on what to write about. Especially if you have a writer’s block, AI can help you get out of it. 

Marlowe, the AI for fiction writing, has been released the summer of 2020. The AI has been fed thousands of bestseller books and has analysed what the key elements of each genre are. The idea behind Marlowe is to help writers write their story by telling them what the readers expect for a certain genre so that the writers can write a successful story. It tells the writer the formula of a genre (think about a beginning hook, build-ups, etc.), but also how the pacing of the storyline should be and how well developed the characters should be. Writers can upload their manuscript, and Marlowe will analyze it and provide feedback. 

But also here, we see the same problem that we encountered in Beyond the Fence: the story writing becomes formulaic. Yes, the writer writes the story and creates its own setting and character. But by analyzing the success key elements and adding that to the story, it loses creativity. It will make every story exactly the same. 

So can a computer write the book for a musical? In big lines, probably yes. Will it be an excellent book? Probably not yet. 

Music composed by AI

Creating a musical piece with AI is not a problem. There are multiple systems, like Jukebox and Amper, that can create a song for you within minutes. You tell the AI what genre, length and beat you want and voila! You have your song. Everybody with an account can do it with no fee. And copyright-free. Pretty great, right? 

A good example of how well computers understand music is the concert The Eternal Golden Braid: Gödel Escher Bach in 2019. In this concert, half the music was written by Bach himself. The other half was written by AI who was fed Bach’s music sheet and wrote the music by a machine learning process. The audience of the concert was asked to vote which song was originally by Bach and which one by the computer. And the results were amazing, as people were completely fooled. They thought Bach’s pieces were done by AI, and vice versa. The music was too similar to tell.   

But with analysing the perfect musical music, it gets more difficult. Musical theatre music isn’t like Bach’s music: one style.  The musical canon goes from Show Boat in the 1920s up to any musicals in 2020. That’s a lot of different songs. You can’t compare the rock songs from Jesus Christ Superstar with the jazzy songs from Show Boat. There is not one ‘genre’ as the canon is too diverse.

This was a problem for the creators of Beyond the Fence. The computer couldn’t really give them one fitting genre or a couple of songs. The AI suggested just using different types of songs. So one song could be rock and another a jazz number. There was no coherence among the music album.

So on music, I would say: Yes, the computer can create the music for a musical. It can do it fast and easy. However, you as a human will have to define the framework. Otherwise, you will get a mix of everything. 

Lyrics created by AI

And then we have the lyrics. Something that AI still hasn’t developed well. 

For the Eurovision song contest 2020, the Netherlands (the host of that year) created a side contest called AI Song Contest. Thirteen countries send in their AI created song. Where the music sounded pretty good, the language was mostly made-up. The Dutch submission took its learning from the American Reddit forum. The result was this chorus:

It’s not complete gibberish and it fitted perfectly on the music as the computer could analyze its rhythm. But the lyrics don’t fit. Sentences are unfinished, and what is abbus abbus?

And we see this more often. There are multiple lyrics generators online and funny as they are, they often don’t make sensible sentences. The lyrics generator Keywords to lyrics for example learns from pop hits and you can select keywords to use in the song. The computer then analyses where the words are used in pop songs and generates a new song based on that. I tried to make a song with the generator with the keywords ‘it’s not you, it’s me’: I mean, how pop songy could I be? The result was sensible, but I honestly felt it was still below expectations. You could, however, use it:  

I know what you want, I know what you need
I know what you like, I know what you see
I know what you want, I know what you need
I know what you want, I know what you need
I know what you like
It’s not you, it’s me
If it’s the way that I’m feelin’ now
It’s not you, it’s me
And if you don’t call me now
I know you’re busy motherfucker

Tell me what you want, I know what you need
I know what you like, I know what you see
I know what you want, I know what you need
I know what you want, I know what you need
I know what you want, I know what you need
I know what you want, I know what you need
I know what you want, I know what you need
I know what you want, I know what you need
I know what you want, I know what you need
I know what you want

The problem is that computers don’t understand the meaning of words. They’re just pieces of code. And according to the code, after the word ‘love’ comes ‘you’, ‘him’ or ‘her’.  So it can create the sentence ‘You love him’. But that’s only because it analysed thousands of songs. It can’t deviate from it, because it doesn’t understand how to use love, as it’s meaningless for the computer.  

However, the generators get better and better. When Beyond the Fence was created in 2016, the lyrics were truly gibberish. And so, humans wrote most of the lyrics. The makers acknowledged that the computer-generated about 25% of the lyrics; the lyricist added the rest. 

And the improvement gets faster and faster. Jukebox is the first program to generate songs with both lyrics and vocals. Although far from perfect, the creators expect the program to create larger musical themes like choruses and repeated phrases in April 2021. Soon, it will all match perfectly. 

So can a computer produce the lyrics for a musical? Yes. If we allow the lyrics to either be a bit gibberish or very simple with many repeated lines. If we look for deeper content, we still have to write it ourselves. However, with the development going fast, it may not take long before a computer can completely produce the lyrics itself. 

To wrap it up

So if we look at the three key elements of musical (lyrics, music and book) we see that AI can already do quite a lot, but not enough. It falls short on lyrics and book and a human needs to make it comprehendible and finish it. 

But a musical is not those three elements separate. Musical theatre is those three combined. The book moves into song and back. And this is something a computer can’t do. It’s either good at music or story writing. This has to do with algorithms and how code is written. Remember how a computer can recognize cats, but not dogs because that wasn’t the task? AI can often do one thing, the task written in the code. Of course, you can add multiple algorithms, but it hasn’t been done yet for musical theatre. The fact that Jukebox can create music AND fitting lyrics is already an enormous step of the last year. So a computer can’t (yet) create a complete musical: it still relies on humans to finish it. 

And then we have the problem of formulaic creating. The problem of Beyond the Fence hasn’t changed in the last four years: computers still create based on mathematics. And as we saw in 2016: it’s boring. Even when you include the success factors, it still doesn’t make it a success. So what will happen when a computer creates an entire musical? Will it be interesting? Will it be mediocre as it takes the best of previous shows? Is success a possibility? 

So my opinion is that AI can certainly help create a musical. And how much influence it can have! Computers don’t ask for royalties, so it can save you as a producer so much money if you have music and lyrics created by a computer. Suddenly your budgets can become cheaper. You can take more risk, so we might get out of the circle of putting the same musicals on and on again. But I do think musical needs human input. You need to have some emotional input, understand the deeper meaning of things, some experience. And I think that’s not something a computer will have any time soon.