Conducting the Mechanical Choir
In which I say a bit about my Mechanical Choir project, which uses recordings by Mechanical Turk workers: how I put it together, why, and how I think it went.
I’m pleased to finally share a project I’ve been working on for a while. I haven’t been very active on the ‘artistic’ front since during my PhD, but recently I’ve been thinking quite a bit about ways to connect the themes that come up in my university work with what I do creatively.
This project involves Amazon’s Mechanical Turk service, where ‘requesters’ post small tasks online, which ‘workers’ complete in return for (usually small) payments. Often workers are contributing tiny parts to larger projects by doing things like things like categorising images to inform machine learning, or filling in surveys.
I find the idea of people ploughing away at these communal tasks, separately, in rooms all over the world, intriguing and depressing. Mturk is the epitome of the gig economy, which gives workers attractive flexibility but often involves low-paid, mundane and insecure work.
The thing I kept coming back to was the idea that the people who work on mTurk are alone/together — working on parts of a larger whole, but often having no insight into what that whole is, and with no idea who else is contributing to it. And, all over the world, they are doing it all the time.
One of the things which best enables and expresses human togetherness is music. Finding rhythm and tune together unites us in ways we feel to be fundamental. I thought about a choir, with members unaware of the song or who else was singing.
I wondered how a choir like this would sound — could you recreate the feeling of singing together, or would it be cold and disconnected?
I also wondered for a bit what the choir should sing, and thought Amazing Grace would be appropriate. It’s a hymn that embodies congregation and overcoming oppression. I don’t think mTurk workers are oppressed — but they are, at least, pressed in many ways.
The making of
I made a MIDI of the tune and split each part up into section of two or three notes, randomised the order and assigned phonetics to each section for the workers to sing.
I was a bit undecided on how much to pay. In some ways it would have been natural to make a point by paying a ridiculously low hourly fee. But ultimately I just felt a bit bad doing that and decided to pay about the living wage of £9 per hour (December 2018, outside of London — which I thought workers would be). So $1 for the 5 minutes I estimated the task would take.
I asked each worker to record three separate sections, which meant twelve workers between them could record a full melody and harmony. In total I commissioned this full set three separate times, yielding 3x3=9 combinations of melodies and harmonies. So 36 singers.
I mixed all these combinations together and applied pretty minimal processing, such as to remove some background noise and even out the volumes of the different clips.
I thought quite long about how to make the track available. Play-on-demand audio or video would both be possible, and I may still do one/both. But I was drawn by the idea of the choir singing all the time — a way to tune in to the gig economy. So I set up an internet radio station with just one song, and tuned www.mechanicalchoir.uk in to that. You can take a listen there now.
So how did it work out? I guess I will have an unusual perspective as I’ve had to listen through to the parts so many times, but there are a few things I’ve noticed.
While the overall feel is pretty discordant and perhaps a bit haunting, there are also surprising moments of harmony. I loved listening through to the clips as I originally downloaded them. Some people were clearly ‘phoning it in’, but others were incredibly committed.
I have no idea where the workers were based, but you can tell that there are a whole range of different accents represented. (I mostly collected the clips early in the morning UK-time, so night in the US, and afternoon/evening in Asia.) Some had intriguing background sounds, like exotic (to me) birds calling.
And for me the project has highlighted important ethical questions. I was clear in communicating who I was and that the work was for a crowdsourced music project, but people had only a limited idea of the words they were saying (phonetic instructions were things like “bur-now”), and no idea what the song was (I think). This was part of the point.
What does it mean to have such a limited view into the overall project we are contributing to? I hope people would feel contributing to Amazing Grace was pretty nice (although it is a religious song, so who knows?). But I could have been having people recite the phonetic units which make up Mein Kampf, or parrot other views they might find objectionable. It seems a reasonable worry that short-term attractive flexibility of gig economy-type work often makes us lose sight of the bigger picture, such as erosion of workers’ rights and fair pay.
Overall I feel like the project achieved what I set out to do. Anyone can go to the website and, in some way, tune in to the gig economy. If you do I hope it might prompt you (like it does me) to think a bit about the ones singing. I’m sure they represent a whole mix of people with a whole range of motivations for doing what they do, and I hope they are doing well.
I wanted to end by thanking all the mTurk workers who contributed their voices to the project. I’m not able to credit contributors as the mTurk T&Cs prohibit sharing worker IDs. However, if you are reading this and think you contributed to the project, I hope you like the result, and I’d be interested to hear your thoughts.
Thanks for reading, and I welcome any contact either via email@example.com or Twitter (@mikefsway). I’m also looking for opportunities to exhibit the work, so if you are aware of anything coming up which you think may fit with the themes, please drop me a line.
P.S. The image for the project — which appears at the top of this post — is also a composite a drawings done by mTurk workers, commissioned in a similar way to the choir.