I'd Like to Thank the Algorithm that Made this Film Possible


Just under a year ago, I wrote about algorithms producing producing rap lyrics and said, “Perhaps the next project will be an algorithm that produces films.” In fact, that project has arrived, and its name is Benjamin (formerly known as Jetson).

Benjamin is the self-chosen name of “the world’s first automated screenwriter,” according to its own Facebook page. Benjamin’s site gives a slightly longer description of the screenwriter as “a self-improving LSTM RNN [Long short-term memory recurrent neural network] machine intelligence trained on human screenplays.”

Benjamin already has written a film entitled “Sunspring” with some prompting from Ross Goodwin, “creative technologist, artist, hacker, data scientist,” and the filmmaker Oscar Sharp. To qualify for the Sci-Fi London 48 Hour Film Challenge, the entire film was made from start to finish in just 48 hours with three human actors playing the roles.

It is available for your viewing pleasure on YouTube. As the writer Neil Gaiman tweeted, “Watch a short SF film gloriously fail the Turing Test.”

The lack of coherence between and within lines gives the film and script (posted here) is reminiscent of the dada movement. At one point, the main character (identified as H) declares: “It’s a damn thing scared to say. Nothing is going to be a thing but I was the one that got on this rock with a child and then I left the other two.” That is followed by the stage direction, “He is standing in the stars and sitting on the floor.”

Benjamin, however, is not deliberately channeling Hugo Ball or Tristan Tzara. He is merely composing what the algorithms dictate. As Goodwin explained in an emailed response to my question about the possibility of influencing the outcome of the algorithm:

    The type of algorithm I used, an LSTM recurrent neural network, can be influenced in certain ways -- for example, by selecting only science fiction materials for the corpus, the output will have a science fiction feel to it. However, the way it generates text is by referencing a statistical model with millions of parameters and predicting which letter comes next over and over again. That process doesn't lend itself to control over story structure or aspects like number of characters, etc.

Algorithms based on the model fed into Benjamin were also used for "Home on the Land," the song sung at the end of the film. Benjamin did not write the music, though. That was done by humans from Tiger and Man. The combination is not bad, though human creativity has to get at least some of the credit.

Whether or not one says the same for the film as a whole really depends on one’s expectations. Certainly, I wouldn’t have sat through all nine-plus minutes to watch the film if it were just a not-ready-for-primetime human production. But because it is an AI production, I was curious to see the results and had much more tolerance for its shortcomings. If it’s viewed as a work-in-progress, it could be interesting to watch for improvements down the road as algorithms are adjusted to produce something closer to what we expect from human writers.

Goodman does believe that AI will eventually be able to achieve the level of coherence one expects from human-produced films. However, he does not consider that to signal the end of human creativity. He contends that we should be thinking about “how we intend to use such technology well in advance of its arrival.” We should be considering in what way that technology can “augment our creativity” rather than rendering it irrelevant.

In light of that assessment, perhaps Benjamin’s song is what best encapsulates the possibility for bringing together human and machine intelligence to produce something that makes the best of both. What do you think?

Ariella Brown,

Ariella Brown is a social media consultant, editor, and freelance writer who frequently writes about the application of technology to business. She holds a PhD in English from the City University of New York. Her Twitter handle is @AriellaBrown.

AI, Machine Learning Power Transformation

AI and machine learning are transforming how companies work with big data. Here's how one organization is implementing the technology.

Car Makers Drive Hard Towards AI Advances

One of the most visible applications of AI and machine learning is in the area of connected cars. Here's how data and analytics are driving these advances.


Re: AI music trods forward
  • 2/6/2017 5:20:08 PM
NO RATINGS

..

Broadway writes


...I wonder too the risk that these momentary music makers will be used by marketers and more nefarious parties to manipulate our moods and coerce our behaviors. Certainly, this is already done in retail stores, for instance, but this leads to a whole level of targeting.


 

Maybe I'm naive, but I doubt that music can coerce behavior ... but I'm pretty sure it can influence behavior.

However, while robot music may be a source for cheaper music for applications such as TV and radio ads (very likely helping influence consumer behavior), my speculation is that the most serious adverse impact might come in the movie industry, if it makes major incursions in an industry looking for ways to cut costs. Quite a number of modern-era serious "classical" composers – e.g., Miklos Rozsa, Bernard Hermann, Franz Waxman, John Williams – have depended on writing film music for a major part of their livelihood. Replacement of human music by AI-generated scores could have a major impact on this aspect of culture.

On the other hand, who knows? Maybe future audiences in concert halls will enjoy listening to symphonies and concertos composed by really clever algorithms, performed to the rave reviews of (robot) critics ...

..

Re: AI music trods forward
  • 1/23/2017 11:07:03 PM
NO RATINGS

@Lyndon, I wonder too the risk that these momentary music makers will be used by marketers and more nefarious parties to manipulate our moods and coerce our behaviors. Certainly, this is already done in retail stores, for instance, but this leads to a whole level of targeting.

AI music trods forward
  • 1/23/2017 9:13:10 PM
NO RATINGS

..

Sane IT writes


The market for song writing bots is pretty small so I don't think competition is driving them to do better than a 4 note melody.  When someone teaches a bot the circle of fifths then we'll probably get more complex songs.  I think the musical side will be easier than lyrics though given the clunky movie dialog that spawned this discussion. 


There's new info to update this thread, provided in a New York Times article today 23 Jan. 2017) titled From Jingles to Pop Hits, A.I. Is Music to Some Ears.

Here's a sample quote:


Jukedeck is one of a growing number of companies using artificial intelligence to compose music. Their computers tap tools like artificial neural networks, modeled on the brain, that allow the machines to learn by doing, rather as a child does. So far, at least, these businesses do not seem to be causing much anxiety among musicians.


 

The discussion raises some intriguing possibilities. According to one of the AI music experimenters, Ed Newton-Rex, future systems could respond automatically to the listener's situation and concoct "music" accordingly:


Mr. Newton-Rex sees artificial intelligence changing the way we listen, especially if computers eventually "understand music enough to make it respond in real time to, let's say, a game, or you going for a run," he said. "Recorded music's brilliant, but it's static. If you're playing a game, Hans Zimmer isn't sitting with you composing. I think responsive systems like that will be a big part of the music of the future."


 

But I wonder about this. Would having those favorite tunes bouncing around in your head disappear? Would you hear something you like, but it goes poof after you complete your video game, workout, or whatever else you're doing? Maybe music is destined to become part of the new throwaway world?

 

 

Re: What machines think of humans
  • 8/27/2016 10:58:30 PM
NO RATINGS

..

SaneIT writes


The market for song writing bots is pretty small so I don't think competition is driving them to do better than a 4 note melody.  When someone teaches a bot the circle of fifths then we'll probably get more complex songs.


 

I agree. I think AI-driven machines will be quite capable of producing acceptable elevator music or even scores for B-rated films.

Maybe even songs for the latest rage in pop music performance.

..

Re: Music by algorithm
  • 8/25/2016 6:58:57 PM
NO RATINGS

@PC That's my take on the relative merits of the film and song, too.

Re: What machines think of humans
  • 8/24/2016 9:59:55 AM
NO RATINGS

The market for song writing bots is pretty small so I don't think competition is driving them to do better than a 4 note melody.  When someone teaches a bot the circle of fifths then we'll probably get more complex songs.  I think the musical side will be easier than lyrics though given the clunky movie dialog that spawned this discussion. 

 

Re: What machines think of humans
  • 8/23/2016 10:59:25 PM
NO RATINGS

..

SaneIT writes


Here is Google's first "song" produced by AI, https://cdn2.vox-cdn.com/uploads/chorus_asset/file/6577761/Google_-_Magenta_music_sample.0.mp3 , a piano melody with just four notes.  Eventually I'm sure these will be less clunky and more complex but the machines are learning and it's better than video game music from the 80s. 


 

Yes, it is definitely better. Not a high bar to surpass, however ...

 

Re: What machines think of humans
  • 8/9/2016 7:55:40 AM
NO RATINGS

The uncanny valley is much harder to overcome than most people who take on a hobby project like this realize.  We can synthesize voices and instruments, we can make synthetic skin, organs, etc. but when you start putting the pieces together the sum of the flaws in all of the engineered parts gives us that creepy feeling.  Here is Google's first "song" produced by AI, https://cdn2.vox-cdn.com/uploads/chorus_asset/file/6577761/Google_-_Magenta_music_sample.0.mp3 , a piano melody with just four notes.  Eventually I'm sure these will be less clunky and more complex but the machines are learning and it's better than video game music from the 80s.  

Re: What machines think of humans
  • 8/8/2016 10:41:20 AM
NO RATINGS

..

SaneIT writes


From what I understand there are computer generated pop stars in Japan, the music, the voices the persona, everything is artificial aside from the writing. 


 

 And then there's a robot that sings: 

Here's an article about it:

This Japanese Robot Is Proof That Mankind Is Doomed

Watching its pathetic twitching kind of makes you feel sorry for it, but to me the "song" is really, really creepy...

I'd say it needs more work before it's ready for Japan's Got Talent ...

..

Re: What machines think of humans
  • 8/1/2016 8:24:12 AM
NO RATINGS

I played a game with someone recently where we sent each other text messages just from the autocorrect/autosuggest keyboard.  It was possible to have a very surface level conversation and even reply with somewhat sensible texts.  It did get repetitive though and it might take 10 extra words to get the point across but it could be done.  This was just a keyboard and us picking from the 3-4 suggestions it gave without typing a single letter.  An AI that is trying to pass the Turing test should be able to write much better dialog. 

Page 1 / 7   >   >>
INFORMATION RESOURCES
ANALYTICS IN ACTION
CARTERTOONS
VIEW ALL +
QUICK POLL
VIEW ALL +